def assignor(url ,specific = None, talk=True, options=None):
if userLock(): return
if duplicateLock(): return
if not componentInfo().check(): return
UC = unifiedConfiguration()
CI = campaignInfo()
SI = global_SI
#LI = lockInfo()
NLI = newLockInfo()
n_assigned = 0
n_stalled = 0
wfos=[]
if specific or options.early:
wfos.extend( session.query(Workflow).filter(Workflow.status=='considered').all())
wfos.extend( session.query(Workflow).filter(Workflow.status=='staging').all())
if specific:
wfos.extend( session.query(Workflow).filter(Workflow.status=='considered-tried').all())
wfos.extend(session.query(Workflow).filter(Workflow.status=='staged').all())
dataset_endpoints = json.loads(open('%s/dataset_endpoints.json'%monitor_dir).read())
max_per_round = UC.get('max_per_round').get('assignor',None)
max_cpuh_block = UC.get('max_cpuh_block')
random.shuffle( wfos )
for wfo in wfos:
if options.limit and (n_stalled+n_assigned)>options.limit:
break
if max_per_round and (n_stalled+n_assigned)>max_per_round:
break
if specific:
if not any(map(lambda sp: sp in wfo.name,specific.split(','))): continue
#if not specific in wfo.name: continue
print "\n\n"
wfh = workflowInfo( url, wfo.name)
wfh.sendLog('assignor',"%s to be assigned"%wfo.name)
## the site whitelist takes into account siteInfo, campaignInfo, memory and cores
(lheinput,primary,parent,secondary, sites_allowed) = wfh.getSiteWhiteList()
## check if by configuration we gave it a GO
no_go = False
allowed_secondary = set()
for campaign in wfh.getCampaigns():
if not CI.go( campaign ):
wfh.sendLog('assignor',"No go for %s"%campaign)
if not options.go:
n_stalled+=1
no_go = True
break
if campaign in CI.campaigns and 'secondaries' in CI.campaigns[campaign]:
allowed_secondary.update( CI.campaigns[campaign]['secondaries'] )
if (secondary and allowed_secondary) and (set(secondary)&allowed_secondary!=set(secondary)):
wfh.sendLog('assignor','%s is not an allowed secondary'%(', '.join(set(secondary)-allowed_secondary)))
#sendEmail('secondary not allowed','%s is not an allowed secondary'%( ', '.join(set(secondary)-allowed_secondary)))
sendLog('assignor','%s is not an allowed secondary'%(', '.join(set(secondary)-allowed_secondary)), level='critical')
if not options.go:
n_stalled+=1
no_go = True
if no_go:
continue
## check on current status for by-passed assignment
if wfh.request['RequestStatus'] !='assignment-approved':
if not options.test:
wfh.sendLog('assignor',"setting %s away and skipping"%wfo.name)
## the module picking up from away will do what is necessary of it
wfo.wm_status = wfh.request['RequestStatus']
wfo.status = 'away'
session.commit()
continue
else:
print wfo.name,wfh.request['RequestStatus']
## retrieve from the schema, dbs and reqMgr what should be the next version
version=wfh.getNextVersion()
if not version:
if options and options.ProcessingVersion:
version = options.ProcessingVersion
else:
wfh.sendLog('assignor',"cannot decide on version number")
n_stalled+=1
wfo.status = 'trouble'
session.commit()
continue
original_sites_allowed = copy.deepcopy( sites_allowed )
wfh.sendLog('assignor',"Site white list %s"%sorted(sites_allowed))
override_sec_location = CI.get(wfh.request['Campaign'], 'SecondaryLocation', [])
blocks = []
if 'BlockWhitelist' in wfh.request:
blocks = wfh.request['BlockWhitelist']
if 'RunWhitelist' in wfh.request and wfh.request['RunWhitelist']:
## augment with run white list
for dataset in primary:
blocks = list(set( blocks + getDatasetBlocks( dataset, runs=wfh.request['RunWhitelist'] ) ))
wfh.sendLog('assignor',"Allowed %s"%sorted(sites_allowed))
secondary_locations=None
for sec in list(secondary):
if override_sec_location:
print "We don't care where the secondary is"
print "Cannot pass for now"
sendEmail("tempting to pass sec location check","but we cannot yet IMO")
#pass
presence = getDatasetPresence( url, sec )
print sec
print json.dumps(presence, indent=2)
one_secondary_locations = [site for (site,(there,frac)) in presence.items() if frac>98.]
#one_secondary_locations = [site for (site,(there,frac)) in presence.items() if there]
if secondary_locations==None:
secondary_locations = one_secondary_locations
else:
secondary_locations = list(set(secondary_locations) & set(one_secondary_locations))
## reduce the site white list to site with secondary only
#sites_allowed = [site for site in sites_allowed if any([osite.startswith(site) for osite in one_secondary_locations])]
sites_allowed = [site for site in sites_allowed if SI.CE_to_SE(site) in one_secondary_locations]
wfh.sendLog('assignor',"From secondary requirement, now Allowed%s"%sorted(sites_allowed))
initial_sites_allowed = copy.deepcopy( sites_allowed ) ## keep track of this, after secondary input location restriction : that's how you want to operate it
sites_all_data = copy.deepcopy( sites_allowed )
sites_with_data = copy.deepcopy( sites_allowed )
sites_with_any_data = copy.deepcopy( sites_allowed )
primary_locations = None
available_fractions = {}
set_lfn = '/store/mc' ## by default
endpoints = set()
for prim in list(primary):
if prim in dataset_endpoints:
print "endpoints from stagor",dataset_endpoints[prim]
endpoints.update( dataset_endpoints[prim] )
set_lfn = getLFNbase( prim )
presence = getDatasetPresence( url, prim , only_blocks=blocks)
if talk:
print prim
print json.dumps(presence, indent=2)
available_fractions[prim] = getDatasetBlocksFraction(url, prim, sites = [SI.CE_to_SE(site) for site in sites_allowed] , only_blocks = blocks)
#sites_all_data = [site for site in sites_with_data if any([osite.startswith(site) for osite in [psite for (psite,(there,frac)) in presence.items() if there]])]
#sites_with_data = [site for site in sites_with_data if any([osite.startswith(site) for osite in [psite for (psite,frac) in presence.items() if frac[1]>90.]])]
sites_all_data = [site for site in sites_with_data if SI.CE_to_SE(site) in [psite for (psite,(there,frac)) in presence.items() if there]]
sites_with_data = [site for site in sites_with_data if SI.CE_to_SE(site) in [psite for (psite,frac) in presence.items() if frac[1]>90.]]
sites_with_any_data = [site for site in sites_with_any_data if SI.CE_to_SE(site) in presence.keys()]
wfh.sendLog('assignor',"Holding the data but not allowed %s"%sorted(list(set([se_site for se_site in presence.keys() if not SI.SE_to_CE(se_site) in sites_allowed]))))
if primary_locations==None:
primary_locations = presence.keys()
else:
primary_locations = list(set(primary_locations) & set(presence.keys() ))
sites_with_data = list(set(sites_with_data))
sites_with_any_data = list(set(sites_with_any_data))
opportunistic_sites=[]
down_time = False
## opportunistic running where any piece of data is available
if secondary_locations or primary_locations:
## intersection of both any pieces of the primary and good IO
#opportunistic_sites = [SI.SE_to_CE(site) for site in list((set(secondary_locations) & set(primary_locations) & set(SI.sites_with_goodIO)) - set(sites_allowed))]
if secondary_locations and primary_locations:
opportunistic_sites = [SI.SE_to_CE(site) for site in list((set(secondary_locations) & set(primary_locations)) - set([SI.CE_to_SE(site) for site in sites_allowed]))]
elif primary_locations:
opportunistic_sites = [SI.SE_to_CE(site) for site in list(set(primary_locations) - set([SI.CE_to_SE(site) for site in sites_allowed]))]
else:
opportunistic_sites = []
wfh.sendLog('assignor',"We could be running in addition at %s"% sorted(opportunistic_sites))
if any([osite in SI.sites_not_ready for osite in opportunistic_sites]):
wfh.sendLog('assignor',"One of the usable site is in downtime %s"%([osite in SI.sites_not_ready for osite in opportunistic_sites]))
down_time = True
## should this be send back to considered ?
## should be 2 but for the time-being let's lower it to get things going
copies_wanted,cpuh = wfh.getNCopies()
wfh.sendLog('assignor',"we need %s CPUh"%cpuh)
if cpuh>max_cpuh_block and not options.go:
#sendEmail('large workflow','that wf %s has a large number of CPUh %s, not assigning, please check the logs'%(wfo.name, cpuh))#,destination=['*****@*****.**'])
sendLog('assignor','%s requires a large numbr of CPUh %s , not assigning, please check with requester'%( wfo.name, cpuh), level='critical')
wfh.sendLog('assignor',"Requiring a large number of CPUh %s, not assigning"%cpuh)
continue
if 'Campaign' in wfh.request and wfh.request['Campaign'] in CI.campaigns and 'maxcopies' in CI.campaigns[wfh.request['Campaign']]:
copies_needed_from_campaign = CI.campaigns[wfh.request['Campaign']]['maxcopies']
copies_wanted = min(copies_needed_from_campaign, copies_wanted)
if not options.early:
less_copies_than_requested = UC.get("less_copies_than_requested")
copies_wanted = max(1,copies_wanted-less_copies_than_requested) # take one out for the efficiency
else:
## find out whether there is a site in the whitelist, that is lacking jobs and reduce to 1 copy needed to get things going
pass
wfh.sendLog('assignor',"needed availability fraction %s"% copies_wanted)
## should also check on number of sources, if large enough, we should be able to overflow most, efficiently
if available_fractions and not all([available>=copies_wanted for available in available_fractions.values()]):
not_even_once = not all([available>=1. for available in available_fractions.values()])
wfh.sendLog('assignor',"The input dataset is not available %s times, only %s"%( copies_wanted, available_fractions.values()))
if down_time and not options.go and not options.early:
wfo.status = 'considered'
session.commit()
wfh.sendLog('assignor',"sending back to considered because of site downtime, instead of waiting")
#sendEmail( "cannot be assigned due to downtime","%s is not sufficiently available, due to down time of a site in the whitelist. check the assignor logs. sending back to considered."% wfo.name)
sendLog('assignor','%s is not sufficiently available, due to down time of a site in the whitelist. sending back to considered.'%( wfo.name ), level='delay')
continue
#pass
print json.dumps(available_fractions)
if (options.go and not_even_once) or not options.go:
known = []
try:
known = json.loads(open('cannot_assign.json').read())
except:
pass
if not wfo.name in known and not options.limit and not options.go and not options.early and not options.partial:
wfh.sendLog('assignor',"cannot be assigned, %s is not sufficiently available.\n %s"%(wfo.name,json.dumps(available_fractions)))
sendEmail( "cannot be assigned","%s is not sufficiently available.\n %s"%(wfo.name,json.dumps(available_fractions)))
known.append( wfo.name )
open('cannot_assign.json','w').write(json.dumps( known, indent=2))
n_stalled+=1
if options.early:
if wfo.status == 'considered':
wfh.sendLog('assignor',"setting considered-tried")
wfo.status = 'considered-tried'
session.commit()
else:
print "tried but status is",wfo.status
if options.partial:
print "Will move on with partial locations"
else:
continue
## default back to white list to original white list with any data
print "Allowed",sorted(sites_allowed)
if options.primary_aaa:
sites_allowed = initial_sites_allowed
#options.useSiteListAsLocation = True
options.TrustSitelists = True
else:
sites_allowed = sites_with_any_data
wfh.sendLog('assignor',"Selected for any data %s"%sorted(sites_allowed))
if options.restrict:
print "Allowed",sites_allowed
sites_allowed = sites_with_any_data
print "Selected",sites_allowed
else:
if set(sites_with_data) != set(sites_allowed):
## the data is not everywhere we wanted to run at : enable aaa
print "Sites with 90% data not matching site white list (block choping!)"
print "Resorting to AAA reading for",list(set(sites_allowed) - set(sites_with_data)),"?"
print "Whitelist site with any data",list(set(sites_allowed) - set(sites_with_any_data))
#options.useSiteListAsLocation = True
#print "Not commissioned yet"
#continue
#print "We could be running at",opportunistic_sites,"in addition"
##sites_allowed = list(set(sites_allowed+ opportunistic_sites))
### check on endpoints for on-going transfers
if endpoints and options.partial:
sites_allowed = list(set(sites_allowed + [SI.SE_to_CE(s) for s in endpoints]))
print "with added endpoints",sorted(sites_allowed)
#if options.partial:
# continue
if not len(sites_allowed):
wfh.sendLog('assignor',"cannot be assign with no matched sites")
#sendEmail( "cannot be assigned","%s has no whitelist"%(wfo.name))
sendLog('assignor','%s has no whitelist'% wfo.name, level='critical')
n_stalled+=1
continue
t1_only = [ce for ce in sites_allowed if ce.startswith('T1')]
if t1_only:
# try to pick from T1 only first
sites_out = [SI.pick_dSE([SI.CE_to_SE(ce) for ce in t1_only])]
else:
# then pick any otherwise
sites_out = [SI.pick_dSE([SI.CE_to_SE(ce) for ce in sites_allowed])]
## one last modification now that we know we can assign, and to make sure all ressource can be used by the request : set all ON sites to whitelist
###sites_allowed = original_sites_allowed ## not needed, afterall as secondary jobs go their own ways
wfh.sendLog('assignor',"Placing the output on %s"%sites_out)
parameters={
'SiteWhitelist' : sites_allowed,
#'CustodialSites' : sites_custodial,
'NonCustodialSites' : sites_out,
'AutoApproveSubscriptionSites' : list(set(sites_out)),
'AcquisitionEra' : wfh.acquisitionEra(),
'ProcessingString' : wfh.processingString(),
'MergedLFNBase' : set_lfn,
'ProcessingVersion' : version,
}
## plain assignment here
team='production'
if os.getenv('UNIFIED_TEAM'): team = os.getenv('UNIFIED_TEAM')
if options and options.team:
team = options.team
## high priority team agent
#if wfh.request['RequestPriority'] >= 100000 and (wfh.request['TimePerEvent']*int(wfh.getRequestNumEvents()))/(8*3600.) < 10000:
# team = 'highprio'
# sendEmail("sending work with highprio team","%s"% wfo.name, destination=['*****@*****.**'])
## SDSC redirection
#if "T2_US_UCSD" in sites_with_data and random.random() < -0.5 and wfh.request['Campaign']=='RunIISpring15DR74' and int(wfh.getRequestNumEvents()) < 600000 and not any([out.endswith('RAW') for out in wfh.request['OutputDatasets']]):
# ## consider SDSC
# parameters['SiteWhitelist'] = ['T2_US_UCSD','T3_US_SDSC']
# parameters['useSiteListAsLocation'] = True
# team = 'allocation-based'
# sendEmail("sending work to SDSC","%s was assigned to SDSC/UCSD"% wfo.name, destination=['*****@*****.**'])
## SDSC redirection
#if wfh.request['Campaign']==R'unIIWinter15GS' and random.random() < -1.0:
# parameters['SiteWhitelist'] = ['T3_US_SDSC']
# team = 'allocation-based'
# sendEmail("sending work to SDSC","%s was assigned to SDSC"% wfo.name, destination=['*****@*****.**'])
if False and 'T2_CH_CERN' in parameters['SiteWhitelist']:
## add some check on
### the amount pending to HLT
### the size of the request
### the priority of the request (maybe not if we decide to overflow during runs)
parameters['SiteWhitelist'] = ['T2_CH_CERN_HLT']
team = 'hlt'
## reduce the splitting by factor of 4, regardless of type of splitting
sendEmail("sending work to HLT","%s was assigned to HLT"%wfo.name)
##parse options entered in command line if any
if options:
for key in reqMgrClient.assignWorkflow.keys:
v=getattr(options,key)
if v!=None:
if type(v)==str and ',' in v:
parameters[key] = filter(None,v.split(','))
else:
parameters[key] = v
## pick up campaign specific assignment parameters
parameters.update( CI.parameters(wfh.request['Campaign']) )
if not options.test:
parameters['execute'] = True
split_check = wfh.checkWorkflowSplitting()
if split_check!=True:
parameters.update( split_check )
if 'EventBased' in split_check.values():
wfh.sendLog('assignor', "Falling back to event splitting.")
#sendEmail("Fallback to EventBased","the workflow %s is too heavy to be processed as it is. Fallback to EventBased splitting"%wfo.name)
sendLog('assignor','the workflow %s is too heavy to be processed as it is. Fallback to EventBased splitting'%wfo.name, level='critical')
elif 'EventsPerJob' in split_check.values():
wfh.sendLog('assignor', "Modifying the number of job per event")
#sendEmail("Modifying the job per events","the workflow %s is too heavy in number of jobs explosion"%wfo.name)
sendLog('assignor',"the workflow %s is too heavy in number of jobs explosion"%wfo.name, level='critical')
# Handle run-dependent MC
pstring = wfh.processingString()
if 'PU_RD' in pstring:
numEvents = wfh.getRequestNumEvents()
eventsPerLumi = [getDatasetEventsPerLumi(prim) for prim in primary]
eventsPerLumi = sum(eventsPerLumi)/float(len(eventsPerLumi))
reqJobs = 500
if 'PU_RD2' in pstring:
reqJobs = 2000
eventsPerJob = int(numEvents/(reqJobs*1.4))
lumisPerJob = int(eventsPerJob/eventsPerLumi)
if lumisPerJob==0:
#sendEmail("issue with event splitting for run-dependent MC","%s needs to be split by event with %s per job"%(wfo.name, eventsPerJob))
sendLog('assignor', "%s needs to be split by event with %s per job"%(wfo.name, eventsPerJob), level='critical')
wfh.sendLog('assignor', "%s needs to be split by event with %s per job"%(wfo.name, eventsPerJob))
parameters['EventsPerJob'] = eventsPerJob
else:
spl = wfh.getSplittings()[0]
eventsPerJobEstimated = spl['events_per_job'] if 'events_per_job' in spl else None
eventsPerJobEstimated = spl['avg_events_per_job'] if 'avg_events_per_job' in spl else None
if eventsPerJobEstimated and eventsPerJobEstimated > eventsPerJob:
#sendEmail("setting lumi splitting for run-dependent MC","%s was assigned with %s lumis/job"%( wfo.name, lumisPerJob))
sendLog('assignor',"%s was assigned with %s lumis/job"%( wfo.name, lumisPerJob), level='critical')
wfh.sendLog('assignor',"%s was assigned with %s lumis/job"%( wfo.name, lumisPerJob))
parameters['LumisPerJob'] = lumisPerJob
else:
#sendEmail("leaving splitting untouched for PU_RD*","please check on "+wfo.name)
sendLog('assignor',"leaving splitting untouched for %s, please check on %s"%( pstring, wfo.name), level='critical')
wfh.sendLog('assignor',"leaving splitting untouched for PU_RD*, please check.")
result = reqMgrClient.assignWorkflow(url, wfo.name, team, parameters)
# set status
if not options.test:
if result:
wfo.status = 'away'
session.commit()
n_assigned+=1
try:
## refetch information and lock output
new_wfi = workflowInfo( url, wfo.name)
(_,prim,_,sec) = new_wfi.getIO()
for secure in list(prim)+list(sec)+new_wfi.request['OutputDatasets']:
## lock all outputs flat
NLI.lock( secure )
#for site in [SI.CE_to_SE(site) for site in sites_allowed]:
# for output in new_wfi.request['OutputDatasets']:
# LI.lock( output, site, 'dataset in production')
# for primary in prim:
# LI.lock( primary, site, 'dataset used in input')
# for secondary in sec:
# LI.lock( secondary, site, 'required for mixing' )
except Exception as e:
print "fail in locking output"
print str(e)
sendEmail("failed locking of output",str(e))
else:
print "ERROR could not assign",wfo.name
else:
pass
print "Assignment summary:"
sendLog('assignor',"Assigned %d Stalled %s"%(n_assigned, n_stalled))
def closor(url, specific=None):
if not componentInfo().check(): return
UC = unifiedConfiguration()
CI = campaignInfo()
#LI = lockInfo()
all_late_files = []
check_fullcopy_to_announce = UC.get('check_fullcopy_to_announce')
## manually closed-out workflows should get to close with checkor
if specific:
wfs = session.query(Workflow).filter(Workflow.name.contains(specific)).all()
else:
wfs = session.query(Workflow).filter(Workflow.status=='close').all()
held = set()
for wfo in wfs:
if specific and not specific in wfo.name: continue
## what is the expected #lumis
wfi = workflowInfo(url, wfo.name )
wfo.wm_status = wfi.request['RequestStatus']
if wfi.request['RequestStatus'] in ['announced','normal-archived']:
## manually announced ??
wfo.status = 'done'
wfo.wm_status = wfi.request['RequestStatus']
wfi.sendLog('closor','%s is announced already : %s'%( wfo.name,wfo.wm_status))
session.commit()
expected_lumis = 1
if not 'TotalInputLumis' in wfi.request:
print wfo.name,"has not been assigned yet, or the database is corrupted"
else:
expected_lumis = wfi.request['TotalInputLumis']
## what are the outputs
outputs = wfi.request['OutputDatasets']
## check whether the number of lumis is as expected for each
all_OK = defaultdict(lambda : False)
#print outputs
if len(outputs):
print wfo.name,wfi.request['RequestStatus']
for out in outputs:
event_count,lumi_count = getDatasetEventsAndLumis(dataset=out)
odb = session.query(Output).filter(Output.datasetname==out).first()
if not odb:
print "adding an output object",out
odb = Output( datasetname = out )
odb.workflow = wfo
session.add( odb )
odb.nlumis = lumi_count
odb.nevents = event_count
odb.workfow_id = wfo.id
if odb.expectedlumis < expected_lumis:
odb.expectedlumis = expected_lumis
else:
expected_lumis = odb.expectedlumis
odb.date = time.mktime(time.gmtime())
session.commit()
wfi.sendLog('closor',"\t%60s %d/%d = %3.2f%%"%(out,lumi_count,expected_lumis,lumi_count/float(expected_lumis)*100.))
#print wfo.fraction_for_closing, lumi_count, expected_lumis
#fraction = wfo.fraction_for_closing
#fraction = 0.0
#all_OK.append((float(lumi_count) > float(expected_lumis*fraction)))
all_OK[out] = True
## check for at least one full copy prior to moving on
in_full = {}
for out in outputs:
in_full[out] = []
presence = getDatasetPresence( url, out )
where = [site for site,info in presence.items() if info[0]]
if where:
all_OK[out] = True
print out,"is in full at",",".join(where)
in_full[out] = copy.deepcopy(where)
else:
going_to = wfi.request['NonCustodialSites']+wfi.request['CustodialSites']
wfi.sendLog('closor',"%s is not in full anywhere. send to %s"%(out, ",".join(sorted(going_to))))
at_destination = dict([(k,v) for (k,v) in presence.items() if k in going_to])
else_where = dict([(k,v) for (k,v) in presence.items() if not k in going_to])
print json.dumps( at_destination )
print json.dumps( else_where, indent=2 )
## do the full stuck transfer study, missing files and shit !
for there in going_to:
late_info = findLateFiles(url, out, going_to = there )
for l in late_info:
l.update({"workflow":wfo.name,"dataset":out})
all_late_files.extend( late_info )
if check_fullcopy_to_announce:
## only set this false if the check is relevant
all_OK[out] = False
## verify if we have to do harvesting
(OK, requests) = spawn_harvesting(url, wfi, in_full)
all_OK.update( OK )
## only that status can let me go into announced
if all(all_OK.values()) and wfi.request['RequestStatus'] in ['closed-out']:
print wfo.name,"to be announced"
results=[]#'dummy']
if not results:
for out in outputs:
if all_OK[out]:
results.append(setDatasetStatus(out, 'VALID'))
tier = out.split('/')[-1]
campaign = None
try:
campaign = out.split('/')[2].split('-')[0]
except:
if 'Campaign' in wfi.request and wfi.request['Campaign']:
campaign = wfi.request['Campaign']
to_DDM = False
## campaign override
if campaign and campaign in CI.campaigns and 'toDDM' in CI.campaigns[campaign] and tier in CI.campaigns[campaign]['toDDM']:
to_DDM = True
## by typical enabling
if tier in UC.get("tiers_to_DDM"):
to_DDM = True
## check for unitarity
if not tier in UC.get("tiers_no_DDM")+UC.get("tiers_to_DDM"):
print "tier",tier,"neither TO or NO DDM for",out
results.append('Not recognitized tier %s'%tier)
sendEmail("failed DDM injection","could not recognize %s for injecting in DDM"% out)
continue
n_copies = 2
destinations=[]
if to_DDM and campaign and campaign in CI.campaigns and 'DDMcopies' in CI.campaigns[campaign]:
ddm_instructions = CI.campaigns[campaign]['DDMcopies']
if type(ddm_instructions) == int:
n_copies = CI.campaigns[campaign]['DDMcopies']
elif type(ddm_instructions) == dict:
## a more fancy configuration
for ddmtier,indication in ddm_instructions.items():
if ddmtier==tier or ddmtier in ['*','all']:
## this is for us
if 'N' in indication:
n_copies = indication['N']
if 'host' in indication:
destinations = indication['host']
destination_spec = ""
if destinations:
destination_spec = "--destination="+",".join( destinations )
## inject to DDM when necessary
if to_DDM:
#print "Sending",out," to DDM"
p = os.popen('python assignDatasetToSite.py --nCopies=%d --dataset=%s %s --exec'%(n_copies, out,destination_spec))
print p.read()
status = p.close()
if status!=None:
print "Failed DDM, retrying a second time"
p = os.popen('python assignDatasetToSite.py --nCopies=%d --dataset=%s %s --exec'%(n_copies, out,destination_spec))
print p.read()
status = p.close()
if status!=None:
sendEmail("failed DDM injection","could not add "+out+" to DDM pool. check closor logs.")
results.append( status )
if status == None:
wfi.sendLog('closor','%s is send to AnalysisOps DDM pool in %s copies %s'%( n_copies, out,destination_spec))
else:
print wfo.name,"no stats for announcing",out
results.append('No Stats')
if all(map(lambda result : result in ['None',None,True],results)):
## only announce if all previous are fine
res = reqMgrClient.announceWorkflowCascade(url, wfo.name)
if not res in ['None',None]:
## check the status again, it might well have toggled
wl_bis = workflowInfo(url, wfo.name)
wfo.wm_status = wl_bis.request['RequestStatus']
session.commit()
if wl_bis.request['RequestStatus'] in ['announced','normal-archived']:
res = None
else:
## retry ?
res = reqMgrClient.announceWorkflowCascade(url, wfo.name)
results.append( res )
#print results
if all(map(lambda result : result in ['None',None,True],results)):
wfo.status = 'done'
session.commit()
wfi.sendLog('closor',"workflow is announced")
else:
print "ERROR with ",wfo.name,"to be announced",json.dumps( results )
else:
print wfo.name,"not good for announcing:",wfi.request['RequestStatus']
wfi.sendLog('closor',"cannot be announced")
held.add( wfo.name )
days_late = 0.
retries_late = 10
really_late_files = [info for info in all_late_files if info['retries']>=retries_late]
really_late_files = [info for info in really_late_files if info['delay']/(60*60*24.)>=days_late]
if really_late_files:
subject = 'These %d files are lagging for %d days and %d retries announcing dataset \n%s'%(len(really_late_files), days_late, retries_late, json.dumps( really_late_files , indent=2) )
sendEmail('waiting for files to announce', subject)
sendLog('closor',subject)
print subject
open('%s/stuck_files.json'%monitor_dir,'w').write( json.dumps( really_late_files , indent=2))
if held:
sendEmail("held from announcing","the workflows below are held up, please check the logs https://cmst2.web.cern.ch/cmst2/unified/logs/closor/last.log \n%s"%("\n".join( held )))
def equalizor(url , specific = None, options=None):
up = componentInfo(mcm=False, soft=['mcm'])
if not up.check(): return
if not specific:
workflows = getWorkflows(url, status='running-closed', details=True)
workflows.extend(getWorkflows(url, status='running-open', details=True))
## start from scratch
modifications = defaultdict(dict)
## define regionality site => fallback allowed. feed on an ssb metric ??
mapping = defaultdict(list)
reversed_mapping = defaultdict(list)
regions = defaultdict(list)
SI = siteInfo()
for site in SI.sites_ready:
region = site.split('_')[1]
if not region in ['US','DE','IT']: continue
regions[region] = [region]
def site_in_depletion(s):
return True
if s in SI.sites_pressure:
(m, r, pressure) = SI.sites_pressure[s]
if float(m) < float(r):
print s,m,r,"lacking pressure"
return True
else:
print s,m,r,"pressure"
pass
return False
for site in SI.sites_ready:
region = site.split('_')[1]
## fallback to the region, to site with on-going low pressure
mapping[site] = [fb for fb in SI.sites_ready if any([('_%s_'%(reg) in fb and fb!=site and site_in_depletion(fb))for reg in regions[region]]) ]
use_T0 = False
if options.augment : use_T0 = True
use_HLT = False
if options.augment : use_HLT=True
if use_HLT:
mapping['T2_CH_CERN'].append('T2_CH_CERN_HLT')
#mapping['T2_IT_Legnaro'].append('T1_IT_CNAF')
for reg in ['IT','DE','UK']:
mapping['T2_CH_CERN'].extend([fb for fb in SI.sites_ready if '_%s_'%reg in fb])
for site,fallbacks in mapping.items():
for fb in fallbacks:
reversed_mapping[fb].append(site)
## this is the fallback mapping
print json.dumps( mapping, indent=2)
#print json.dumps( reversed_mapping, indent=2)
altered_tasks = set()
def running_idle( wfi , task_name):
gmon = wfi.getGlideMon()
#print gmon
if not gmon: return (0,0)
if not task_name in gmon: return (0,0)
return (gmon[task_name]['Running'], gmon[task_name]['Idle'])
def needs_action( wfi, task, min_idled = 100, pressure = 0.2):
task_name = task.pathName.split('/')[-1]
running, idled = running_idle( wfi, task_name)
go = True
if not idled and not running :
go = False
if idled < 100:
go = False
if (not running and idled) or (running and (idled / float(running) > pressure)):
go = True
else:
go = False
return go, task_name, running, idled
def getcampaign( task ):
taskname = task.pathName.split('/')[-1]
if hasattr( task, 'prepID'):
return task.prepID.split('-')[1]
elif taskname.count('-')>=1:
return taskname.split('-')[1]
else:
return None
def close( interface ):
open('%s/equalizor.json.new'%monitor_dir,'w').write( json.dumps( interface, indent=2))
os.system('mv %s/equalizor.json.new %s/equalizor.json'%(monitor_dir,monitor_dir))
os.system('cp %s/equalizor.json %s/logs/equalizor/equalizor.%s.json'%(monitor_dir,monitor_dir,time.mktime(time.gmtime())))
interface = {
'reversed_mapping' : reversed_mapping,
'modifications' : {}
}
if options.augment or options.remove:
interface['modifications'] = json.loads( open('%s/equalizor.json'%monitor_dir).read())['modifications']
if options.remove:
if specific in interface['modifications']:
print "poping",specific
interface['modifications'].pop(specific)
close( interface )
return
PU_locations = {}
PU_overflow = {
#'RunIISpring15PrePremix' : {
# 'sites' : ["T1_US_FNAL", "T1_DE_KIT" , "T1_IT_CNAF", "T1_RU_JINR" ,"T2_CH_CERN"],
# 'max' : 20000,
# 'pending' : 0
# },
'RunIIFall15DR76' : {
'sites':['T1_ES_PIC','T2_US_Purdue','T2_UK_SGrid_RALPP','T2_BE_IIHE','T2_DE_DESY','T2_IT_Legnaro','T2_US_Caltech','T1_DE_KIT',
'T2_UK_London_Brunel','T2_IT_Pisa',
'T1_US_FNAL',
'T2_IT_Rome','T2_US_Florida','T1_IT_CNAF','T1_RU_JINR','T2_UK_London_IC','T2_US_Nebraska','T2_FR_CCIN2P3','T2_US_UCSD','T2_ES_CIEMAT',
'T1_FR_CCIN2P3','T2_US_Wisconsin','T2_US_MIT','T2_DE_RWTH',
'T1_UK_RAL','T2_US_Vanderbilt','T2_CH_CERN'],
'max': 20000,
'pending' : 0},
'RunIISpring16DR80' : {
'sites':['T1_ES_PIC','T2_US_Purdue','T2_UK_SGrid_RALPP','T2_BE_IIHE','T2_DE_DESY','T2_IT_Legnaro','T2_US_Caltech','T1_DE_KIT',
'T2_UK_London_Brunel','T2_IT_Pisa',
'T1_US_FNAL',
'T2_IT_Rome','T2_US_Florida','T1_IT_CNAF','T1_RU_JINR','T2_UK_London_IC','T2_US_Nebraska','T2_FR_CCIN2P3','T2_US_UCSD','T2_ES_CIEMAT',
'T1_FR_CCIN2P3','T2_US_Wisconsin','T2_US_MIT','T2_DE_RWTH',
'T1_UK_RAL','T2_US_Vanderbilt','T2_CH_CERN'],
'max': 20000,
'pending' : 0,
'force' : True},
'RunIISpring15DR74' : {
'sites' : ['T1_ES_PIC','T1_DE_KIT','T1_US_FNAL','T1_IT_CNAF','T1_RU_JINR','T1_FR_CCIN2P3','T1_UK_RAL','T2_CH_CERN'],
'max' : 20000,
'pending' : 0}
}
set_to = SI.sites_AAA
LHE_overflow = {
'RunIIWinter15GS' : set_to,
'RunIISummer15GS' : set_to,
'Summer12' : set_to,
'Summer11Leg' : set_to
#'RunIIFall15MiniAODv2' : set_to,
}
pending_HLT = 0
max_HLT = 60000
pending_T0 = 0
max_T0 = 60000
try:
gmon = json.loads(os.popen('curl -s http://cms-gwmsmon.cern.ch/prodview/json/T2_CH_CERN_HLT').read())
pending_HLT += gmon["Running"]
pending_HLT += gmon["MatchingIdle"]
except:
pass
t0_special = [
'vlimant_BPH-RunIISummer15GS-00030_00212_v0__160129_135314_9755',
'pdmvserv_TSG-RunIISummer15GS-00044_00240_v0__160210_121223_8582'
]
no_routing = [
#'vlimant_BPH-RunIISummer15GS-00030_00212_v0__160129_135314_9755',
#'pdmvserv_TOP-RunIIWinter15GS-00074_00187_v0__160207_162312_1992',
]
stay_within_site_whitelist = False
specific_task=None
if specific and ":" in specific:
specific,specific_task = specific.split(':')
if specific:
wfs = session.query(Workflow).filter(Workflow.name.contains(specific)).all()
else:
wfs = session.query(Workflow).filter(Workflow.status == 'away').all()
random.shuffle( wfs )
for wfo in wfs:
if wfo.name in no_routing and not options.augment:
continue
if specific and not specific in wfo.name:
continue
if specific:
wfi = workflowInfo(url, wfo.name)
else:
cached = filter(lambda d : d['RequestName']==wfo.name, workflows)
if not cached : continue
wfi = workflowInfo(url, wfo.name, request = cached[0])
## only running should get re-routed
if not wfi.request['RequestStatus'] in ['running-open','running-closed'] and not specific: continue
tasks_and_campaigns = []
for task in wfi.getWorkTasks():
tasks_and_campaigns.append( (task, getcampaign(task) ) )
_,_,_,sec = wfi.getIO()
## check needs override
needs_overide = False
if not needs_overide and options.augment: needs_overide=True
def overide_from_agent( wfi, needs_overide):
bad_agents = []#'http://cmssrv219.fnal.gov:5984']
if not bad_agents: return needs_overide
if needs_overide: return True
agents = wfi.getAgents()
wqss = ['Running','Acquired']
if any([agent in agents.get(wqs,{}).keys() for wqs,agent in itertools.product( wqss, bad_agents)]):
print "overriding the need for bad agent"
needs_overide = True
return needs_overide
## now parse this for action
for i_task,(task,campaign) in enumerate(tasks_and_campaigns):
if options.augment:
print task.pathName
print campaign
### rule to avoid the issue of taskchain secondary jobs being stuck at sites processing the initial step
if campaign in LHE_overflow:
if task.taskType in ['Processing']:
needs, task_name, running, idled = needs_action(wfi, task)
needs_overide = overide_from_agent( wfi, needs_overide)
extend_to = copy.deepcopy( LHE_overflow[campaign] )
if stay_within_site_whitelist:
extend_to = list(set(extend_to) & set(wfi.request['SiteWhitelist'])) ## restrict to stupid-site-whitelist
if extend_to and needs or needs_overide:
print "\t",task_name,"of",wfo.name,"running",running,"and pending",idled,"taking action : ReplaceSiteWhitelist"
modifications[wfo.name][task.pathName] = { "ReplaceSiteWhitelist" : copy.deepcopy( LHE_overflow[campaign] ) ,"Running" : running, "Pending" : idled, "Priority" : wfi.request['RequestPriority']}
#print json.dumps( modifications[wfo.name][task.pathName]['ReplaceSiteWhitelist']
altered_tasks.add( task.pathName )
else:
print task_name,"of",wfo.name,"running",running,"and pending",idled
### overflow the 76 digi-reco to the site holding the pileup
if campaign in PU_overflow:
force = PU_overflow[campaign]['force'] if 'force' in PU_overflow[campaign] else False
secondary_locations = set(SI.sites_ready)
for s in sec:
if not s in PU_locations:
presence = getDatasetPresence( url, s)
#one_secondary_locations = [site for (site,(there,frac)) in presence.items() if there]
one_secondary_locations = [site for (site,(there,frac)) in presence.items() if frac>98.]
PU_locations[s] = one_secondary_locations
print "secondary is at",sorted(PU_locations[s])
secondary_locations = set(PU_locations[s]) & secondary_locations
## we should add all sites that hold the secondary input if any
secondary_locations = list(set(PU_overflow[campaign]['sites']) & set( SI.sites_ready ))
if any([task.pathName.endswith(finish) for finish in ['_0','StepOneProc','Production']]) :
needs, task_name, running, idled = needs_action(wfi, task)
## removing the ones in the site whitelist already since they encode the primary input location
if stay_within_site_whitelist:
original_site_in_use = set(wfi.request['SiteWhitelist'])
else:
original_site_in_use = set(secondary_locations)
## remove the sites that have already running jobs
gmon = wfi.getGlideMon()
if gmon and task_name in gmon and 'Sites' in gmon[task_name]:
site_in_use = set(gmon[task_name]['Sites'])
## that determines where you want to run in addition
#augment_by = list((set(secondary_locations)- site_in_use))
augment_by = list((set(secondary_locations)- site_in_use) & original_site_in_use) ## restrict to stupid-site-whitelist
else:
augment_by = list(original_site_in_use)
needs_overide = overide_from_agent( wfi, needs_overide)
if augment_by and (needs or needs_overide or force) and PU_overflow[campaign]['pending'] < PU_overflow[campaign]['max']:
PU_overflow[campaign]['pending'] += idled
print "raising overflow to",PU_overflow[campaign]['pending'],"for",PU_overflow[campaign]['max']
## the step with an input ought to be the digi part : make this one go anywhere
modifications[wfo.name][task.pathName] = { "AddWhitelist" : augment_by , "Running" : running, "Pending" : idled, "Priority" : wfi.request['RequestPriority']}
altered_tasks.add( task.pathName )
print "\t",task_name,"of",wfo.name,"running",running,"and pending",idled,"taking action : AddWhitelist"
#print json.dumps( augment_by, indent=2 )
else:
print task_name,"of",wfo.name,"running",running,"and pending",idled
### overflow the skims back to multi-core
if campaign in ['Run2015D','Run2015C_25ns'] and task.taskType =='Skim':
original_swl = wfi.request['SiteWhitelist']
needs, task_name, running, idled = needs_action(wfi, task)
if (needs or needs_overide):
modifications[wfo.name][task.pathName] = { 'AddWhitelist' : original_swl,
"Running" : running, "Pending" : idled, "Priority" : wfi.request['RequestPriority']}
altered_tasks.add( task.pathName )
print "\t",task_name,"of",wfo.name,"running",running,"and pending",idled,"taking action : AddWhitelist"
if options.augment:
print sorted(wfi.request['SiteWhitelist']),i_task,use_HLT
### add the HLT at partner of CERN
if 'T2_CH_CERN' in wfi.request['SiteWhitelist'] and i_task==0 and use_HLT:
needs, task_name, running, idled = needs_action(wfi, task)
if options.augment: needs=True
needs = True
##needs = random.random()<0.40 remove the random, just add up to a limit
if (needs or needs_overide) and pending_HLT < max_HLT:
pending_HLT += idled
if task.pathName in modifications[wfo.name] and 'AddWhitelist' in modifications[wfo.name][task.pathName]:
modifications[wfo.name][task.pathName]["AddWhitelist"].append( "T2_CH_CERN_HLT" )
print "\t",wfo.name,"adding addHLT up to",pending_HLT,"for",max_HLT
print task.pathName
## this Replace does not work at all for HLT
#elif task.pathName in modifications[wfo.name] and 'ReplaceSiteWhitelist' in modifications[wfo.name][task.pathName]:
#modifications[wfo.name][task.pathName]["ReplaceSiteWhitelist"].append( "T2_CH_CERN_HLT" )
#print "\t",wfo.name,"adding replace HLT up to",pending_HLT,"for",max_HLT
else:
modifications[wfo.name][task.pathName] = { "AddWhitelist" : ["T2_CH_CERN_HLT"],
"Priority" : wfi.request['RequestPriority'],
"Running" : running,
"Pending" : idled}
print "\t",wfo.name,"adding HLT up to",pending_HLT,"for",max_HLT
print task.pathName
if i_task==0 and not sec and use_T0:
needs, task_name, running, idled = needs_action(wfi, task)
if options.augment: needs=True
#needs = True
#if not (wfo.name in t0_special) and not options.augment: needs = False
if not wfi.request['RequestType'] in ['MonteCarlo','MonteCarloFromGEN'] and not options.augment: needs = False
##needs = random.random()<0.40 remove the random, just add up to a limit
if (needs or needs_overide):
pending_T0 += idled
if task.pathName in modifications[wfo.name] and 'AddWhitelist' in modifications[wfo.name][task.pathName]:
modifications[wfo.name][task.pathName]["AddWhitelist"].append( "T0_CH_CERN" )
print "\t",wfo.name,"adding addT0 up to",pending_T0,"for",max_T0
print task.pathName
elif task.pathName in modifications[wfo.name] and 'ReplaceSiteWhitelist' in modifications[wfo.name][task.pathName]:
modifications[wfo.name][task.pathName]["ReplaceSiteWhitelist"].append( "T0_CH_CERN" )
print "\t",wfo.name,"adding replace T0 up to",pending_T0,"for",max_T0
else:
modifications[wfo.name][task.pathName] = { "AddWhitelist" : ["T0_CH_CERN"],
"Priority" : wfi.request['RequestPriority'],
"Running" : running,
"Pending" : idled}
print "\t",wfo.name,"adding T0 up to",pending_T0,"for",max_T0
print task.pathName
interface['modifications'].update( modifications )
## temporary core managing
interface['cores']={'T2_CH_CERN_HLT': {'min':4,'max':16}, 'default': {'min':1, 'max':4}}
#interface['max_cores']={'T2_CH_CERN_HLT': 16, 'default': 4}
#interface['min_cores']={'T2_CH_CERN_HLT': 4, 'default': 1}
#interface['resize_subtasks'] = 'RunIISpring16DR80'
interface['resizes'] = ['RunIISpring16DR80','NotACampaign']
## close and save
close( interface )
def assignor(url, specific=None, talk=True, options=None):
if userLock():
return
if duplicateLock():
return
if not componentInfo().check():
return
UC = unifiedConfiguration()
CI = campaignInfo()
SI = siteInfo()
NLI = newLockInfo()
n_assigned = 0
n_stalled = 0
wfos = []
if specific or options.early:
wfos.extend(session.query(Workflow).filter(Workflow.status == "considered").all())
wfos.extend(session.query(Workflow).filter(Workflow.status == "staging").all())
if specific:
wfos.extend(session.query(Workflow).filter(Workflow.status == "considered-tried").all())
wfos.extend(session.query(Workflow).filter(Workflow.status == "staged").all())
dataset_endpoints = json.loads(open("%s/dataset_endpoints.json" % monitor_dir).read())
max_per_round = UC.get("max_per_round").get("assignor", None)
max_cpuh_block = UC.get("max_cpuh_block")
random.shuffle(wfos)
for wfo in wfos:
if options.limit and (n_stalled + n_assigned) > options.limit:
break
if max_per_round and (n_stalled + n_assigned) > max_per_round:
break
if specific:
if not any(map(lambda sp: sp in wfo.name, specific.split(","))):
continue
# if not specific in wfo.name: continue
print "\n\n"
wfh = workflowInfo(url, wfo.name)
wfh.sendLog("assignor", "%s to be assigned" % wfo.name)
## the site whitelist takes into account siteInfo, campaignInfo, memory and cores
(lheinput, primary, parent, secondary, sites_allowed) = wfh.getSiteWhiteList()
## check if by configuration we gave it a GO
no_go = False
if not wfh.go(log=True) and not options.go:
n_stalled += 1
no_go = True
allowed_secondary = set()
for campaign in wfh.getCampaigns():
if campaign in CI.campaigns and "secondaries" in CI.campaigns[campaign]:
allowed_secondary.update(CI.campaigns[campaign]["secondaries"])
if (secondary and allowed_secondary) and (set(secondary) & allowed_secondary != set(secondary)):
wfh.sendLog("assignor", "%s is not an allowed secondary" % (", ".join(set(secondary) - allowed_secondary)))
# sendEmail('secondary not allowed','%s is not an allowed secondary'%( ', '.join(set(secondary)-allowed_secondary)))
sendLog(
"assignor",
"%s is not an allowed secondary" % (", ".join(set(secondary) - allowed_secondary)),
level="critical",
)
if not options.go:
n_stalled += 1
no_go = True
if no_go:
continue
## check on current status for by-passed assignment
if wfh.request["RequestStatus"] != "assignment-approved":
if not options.test:
wfh.sendLog("assignor", "setting %s away and skipping" % wfo.name)
## the module picking up from away will do what is necessary of it
wfo.wm_status = wfh.request["RequestStatus"]
wfo.status = "away"
session.commit()
continue
else:
print wfo.name, wfh.request["RequestStatus"]
## retrieve from the schema, dbs and reqMgr what should be the next version
version = wfh.getNextVersion()
if not version:
if options and options.ProcessingVersion:
version = options.ProcessingVersion
else:
wfh.sendLog("assignor", "cannot decide on version number")
n_stalled += 1
wfo.status = "trouble"
session.commit()
continue
original_sites_allowed = copy.deepcopy(sites_allowed)
wfh.sendLog("assignor", "Site white list %s" % sorted(sites_allowed))
override_sec_location = CI.get(wfh.request["Campaign"], "SecondaryLocation", [])
blocks = []
if "BlockWhitelist" in wfh.request:
blocks = wfh.request["BlockWhitelist"]
if "RunWhitelist" in wfh.request and wfh.request["RunWhitelist"]:
## augment with run white list
for dataset in primary:
blocks = list(set(blocks + getDatasetBlocks(dataset, runs=wfh.request["RunWhitelist"])))
wfh.sendLog("assignor", "Allowed %s" % sorted(sites_allowed))
secondary_locations = None
primary_aaa = options.primary_aaa
if (
"Campaign" in wfh.request
and wfh.request["Campaign"] in CI.campaigns
and "primary_AAA" in CI.campaigns[wfh.request["Campaign"]]
):
primary_aaa = primary_aaa or CI.campaigns[wfh.request["Campaign"]]["primary_AAA"]
secondary_aaa = options.secondary_aaa
if (
"Campaign" in wfh.request
and wfh.request["Campaign"] in CI.campaigns
and "secondary_AAA" in CI.campaigns[wfh.request["Campaign"]]
):
secondary_aaa = secondary_aaa or CI.campaigns[wfh.request["Campaign"]]["secondary_AAA"]
for sec in list(secondary):
if override_sec_location:
print "We don't care where the secondary is"
print "Cannot pass for now"
sendEmail("tempting to pass sec location check", "but we cannot yet IMO")
# pass
if secondary_aaa:
# just continue without checking
continue
presence = getDatasetPresence(url, sec)
print sec
print json.dumps(presence, indent=2)
one_secondary_locations = [site for (site, (there, frac)) in presence.items() if frac > 98.0]
# one_secondary_locations = [site for (site,(there,frac)) in presence.items() if there]
if secondary_locations == None:
secondary_locations = one_secondary_locations
else:
secondary_locations = list(set(secondary_locations) & set(one_secondary_locations))
## reduce the site white list to site with secondary only
# sites_allowed = [site for site in sites_allowed if any([osite.startswith(site) for osite in one_secondary_locations])]
sites_allowed = [site for site in sites_allowed if SI.CE_to_SE(site) in one_secondary_locations]
wfh.sendLog("assignor", "From secondary requirement, now Allowed%s" % sorted(sites_allowed))
initial_sites_allowed = copy.deepcopy(
sites_allowed
) ## keep track of this, after secondary input location restriction : that's how you want to operate it
sites_all_data = copy.deepcopy(sites_allowed)
sites_with_data = copy.deepcopy(sites_allowed)
sites_with_any_data = copy.deepcopy(sites_allowed)
primary_locations = None
available_fractions = {}
set_lfn = "/store/mc" ## by default
endpoints = set()
for prim in list(primary):
if prim in dataset_endpoints:
print "endpoints from stagor", dataset_endpoints[prim]
endpoints.update(dataset_endpoints[prim])
set_lfn = getLFNbase(prim)
presence = getDatasetPresence(url, prim, only_blocks=blocks)
if talk:
print prim
print json.dumps(presence, indent=2)
available_fractions[prim] = getDatasetBlocksFraction(
url, prim, sites=[SI.CE_to_SE(site) for site in sites_allowed], only_blocks=blocks
)
# sites_all_data = [site for site in sites_with_data if any([osite.startswith(site) for osite in [psite for (psite,(there,frac)) in presence.items() if there]])]
# sites_with_data = [site for site in sites_with_data if any([osite.startswith(site) for osite in [psite for (psite,frac) in presence.items() if frac[1]>90.]])]
sites_all_data = [
site
for site in sites_with_data
if SI.CE_to_SE(site) in [psite for (psite, (there, frac)) in presence.items() if there]
]
sites_with_data = [
site
for site in sites_with_data
if SI.CE_to_SE(site) in [psite for (psite, frac) in presence.items() if frac[1] > 90.0]
]
sites_with_any_data = [site for site in sites_with_any_data if SI.CE_to_SE(site) in presence.keys()]
wfh.sendLog(
"assignor",
"Holding the data but not allowed %s"
% sorted(
list(set([se_site for se_site in presence.keys() if not SI.SE_to_CE(se_site) in sites_allowed]))
),
)
if primary_locations == None:
primary_locations = presence.keys()
else:
primary_locations = list(set(primary_locations) & set(presence.keys()))
sites_with_data = list(set(sites_with_data))
sites_with_any_data = list(set(sites_with_any_data))
opportunistic_sites = []
down_time = False
## opportunistic running where any piece of data is available
if secondary_locations or primary_locations:
## intersection of both any pieces of the primary and good IO
# opportunistic_sites = [SI.SE_to_CE(site) for site in list((set(secondary_locations) & set(primary_locations) & set(SI.sites_with_goodIO)) - set(sites_allowed))]
if secondary_locations and primary_locations:
opportunistic_sites = [
SI.SE_to_CE(site)
for site in list(
(set(secondary_locations) & set(primary_locations))
- set([SI.CE_to_SE(site) for site in sites_allowed])
)
]
elif primary_locations:
opportunistic_sites = [
SI.SE_to_CE(site)
for site in list(set(primary_locations) - set([SI.CE_to_SE(site) for site in sites_allowed]))
]
else:
opportunistic_sites = []
wfh.sendLog("assignor", "We could be running in addition at %s" % sorted(opportunistic_sites))
if any([osite in SI.sites_not_ready for osite in opportunistic_sites]):
wfh.sendLog(
"assignor",
"One of the usable site is in downtime %s"
% ([osite in SI.sites_not_ready for osite in opportunistic_sites]),
)
down_time = True
## should this be send back to considered ?
## should be 2 but for the time-being let's lower it to get things going
copies_wanted, cpuh = wfh.getNCopies()
wfh.sendLog("assignor", "we need %s CPUh" % cpuh)
if cpuh > max_cpuh_block and not options.go:
# sendEmail('large workflow','that wf %s has a large number of CPUh %s, not assigning, please check the logs'%(wfo.name, cpuh))#,destination=['*****@*****.**'])
sendLog(
"assignor",
"%s requires a large numbr of CPUh %s , not assigning, please check with requester" % (wfo.name, cpuh),
level="critical",
)
wfh.sendLog("assignor", "Requiring a large number of CPUh %s, not assigning" % cpuh)
continue
if (
"Campaign" in wfh.request
and wfh.request["Campaign"] in CI.campaigns
and "maxcopies" in CI.campaigns[wfh.request["Campaign"]]
):
copies_needed_from_campaign = CI.campaigns[wfh.request["Campaign"]]["maxcopies"]
copies_wanted = min(copies_needed_from_campaign, copies_wanted)
if not options.early:
less_copies_than_requested = UC.get("less_copies_than_requested")
copies_wanted = max(1, copies_wanted - less_copies_than_requested) # take one out for the efficiency
else:
## find out whether there is a site in the whitelist, that is lacking jobs and reduce to 1 copy needed to get things going
pass
wfh.sendLog("assignor", "needed availability fraction %s" % copies_wanted)
## should also check on number of sources, if large enough, we should be able to overflow most, efficiently
if available_fractions and not all([available >= copies_wanted for available in available_fractions.values()]):
not_even_once = not all([available >= 1.0 for available in available_fractions.values()])
wfh.sendLog(
"assignor",
"The input dataset is not available %s times, only %s" % (copies_wanted, available_fractions.values()),
)
if down_time and not options.go and not options.early:
wfo.status = "considered"
session.commit()
wfh.sendLog("assignor", "sending back to considered because of site downtime, instead of waiting")
# sendEmail( "cannot be assigned due to downtime","%s is not sufficiently available, due to down time of a site in the whitelist. check the assignor logs. sending back to considered."% wfo.name)
sendLog(
"assignor",
"%s is not sufficiently available, due to down time of a site in the whitelist. sending back to considered."
% (wfo.name),
level="delay",
)
continue
# pass
print json.dumps(available_fractions)
if (options.go and not_even_once) or not options.go:
known = []
try:
known = json.loads(open("cannot_assign.json").read())
except:
pass
if (
not wfo.name in known
and not options.limit
and not options.go
and not options.early
and not options.partial
):
wfh.sendLog(
"assignor",
"cannot be assigned, %s is not sufficiently available.\n %s"
% (wfo.name, json.dumps(available_fractions)),
)
sendEmail(
"cannot be assigned",
"%s is not sufficiently available.\n %s" % (wfo.name, json.dumps(available_fractions)),
)
known.append(wfo.name)
open("cannot_assign.json", "w").write(json.dumps(known, indent=2))
n_stalled += 1
if options.early:
if wfo.status == "considered":
wfh.sendLog("assignor", "setting considered-tried")
wfo.status = "considered-tried"
session.commit()
else:
print "tried but status is", wfo.status
if options.partial:
print "Will move on with partial locations"
else:
continue
## default back to white list to original white list with any data
print "Allowed", sorted(sites_allowed)
if primary_aaa:
sites_allowed = initial_sites_allowed
options.TrustSitelists = True
wfh.sendLog("assignor", "Selected to read primary through xrootd %s" % sorted(sites_allowed))
else:
sites_allowed = sites_with_any_data
wfh.sendLog("assignor", "Selected for any data %s" % sorted(sites_allowed))
if secondary_aaa:
options.TrustPUSitelists = True
wfh.sendLog("assignor", "Reading secondary through xrootd from %s" % sorted(sites_allowed))
### check on endpoints for on-going transfers
if endpoints and options.partial:
sites_allowed = list(set(sites_allowed + [SI.SE_to_CE(s) for s in endpoints]))
print "with added endpoints", sorted(sites_allowed)
if not len(sites_allowed):
wfh.sendLog("assignor", "cannot be assign with no matched sites")
sendLog("assignor", "%s has no whitelist" % wfo.name, level="critical")
n_stalled += 1
continue
t1_only = [ce for ce in sites_allowed if ce.startswith("T1")]
if t1_only:
# try to pick from T1 only first
sites_out = [SI.pick_dSE([SI.CE_to_SE(ce) for ce in t1_only])]
else:
# then pick any otherwise
sites_out = [SI.pick_dSE([SI.CE_to_SE(ce) for ce in sites_allowed])]
wfh.sendLog("assignor", "Placing the output on %s" % sites_out)
parameters = {
"SiteWhitelist": sites_allowed,
"NonCustodialSites": sites_out,
"AutoApproveSubscriptionSites": list(set(sites_out)),
"AcquisitionEra": wfh.acquisitionEra(),
"ProcessingString": wfh.processingString(),
"MergedLFNBase": set_lfn,
"ProcessingVersion": version,
}
## plain assignment here
team = "production"
if os.getenv("UNIFIED_TEAM"):
team = os.getenv("UNIFIED_TEAM")
if options and options.team:
team = options.team
if False and "T2_CH_CERN" in parameters["SiteWhitelist"]:
## add some check on
### the amount pending to HLT
### the size of the request
### the priority of the request (maybe not if we decide to overflow during runs)
parameters["SiteWhitelist"] = ["T2_CH_CERN_HLT"]
team = "hlt"
## reduce the splitting by factor of 4, regardless of type of splitting
sendEmail("sending work to HLT", "%s was assigned to HLT" % wfo.name)
##parse options entered in command line if any
if options:
for key in reqMgrClient.assignWorkflow.keys:
v = getattr(options, key)
if v != None:
if type(v) == str and "," in v:
parameters[key] = filter(None, v.split(","))
else:
parameters[key] = v
if lheinput:
## throttle reading LHE article
wfh.sendLog("assignor", "Setting the number of events per job to 500k max")
parameters["EventsPerJob"] = 500000
## pick up campaign specific assignment parameters
parameters.update(CI.parameters(wfh.request["Campaign"]))
if not options.test:
parameters["execute"] = True
split_check = wfh.checkWorkflowSplitting()
if split_check != True:
parameters.update(split_check)
if "EventBased" in split_check.values():
wfh.sendLog("assignor", "Falling back to event splitting.")
# sendEmail("Fallback to EventBased","the workflow %s is too heavy to be processed as it is. Fallback to EventBased splitting"%wfo.name)
sendLog(
"assignor",
"the workflow %s is too heavy to be processed as it is. Fallback to EventBased splitting"
% wfo.name,
level="critical",
)
elif "EventsPerJob" in split_check.values():
wfh.sendLog("assignor", "Modifying the number of job per event")
# sendEmail("Modifying the job per events","the workflow %s is too heavy in number of jobs explosion"%wfo.name)
sendLog(
"assignor", "the workflow %s is too heavy in number of jobs explosion" % wfo.name, level="critical"
)
# Handle run-dependent MC
pstring = wfh.processingString()
if "PU_RD" in pstring:
numEvents = wfh.getRequestNumEvents()
eventsPerLumi = [getDatasetEventsPerLumi(prim) for prim in primary]
eventsPerLumi = sum(eventsPerLumi) / float(len(eventsPerLumi))
reqJobs = 500
if "PU_RD2" in pstring:
reqJobs = 2000
eventsPerJob = int(numEvents / (reqJobs * 1.4))
lumisPerJob = int(eventsPerJob / eventsPerLumi)
if lumisPerJob == 0:
# sendEmail("issue with event splitting for run-dependent MC","%s needs to be split by event with %s per job"%(wfo.name, eventsPerJob))
sendLog(
"assignor",
"%s needs to be split by event with %s per job" % (wfo.name, eventsPerJob),
level="critical",
)
wfh.sendLog("assignor", "%s needs to be split by event with %s per job" % (wfo.name, eventsPerJob))
parameters["EventsPerJob"] = eventsPerJob
else:
spl = wfh.getSplittings()[0]
eventsPerJobEstimated = spl["events_per_job"] if "events_per_job" in spl else None
eventsPerJobEstimated = spl["avg_events_per_job"] if "avg_events_per_job" in spl else None
if eventsPerJobEstimated and eventsPerJobEstimated > eventsPerJob:
# sendEmail("setting lumi splitting for run-dependent MC","%s was assigned with %s lumis/job"%( wfo.name, lumisPerJob))
sendLog(
"assignor", "%s was assigned with %s lumis/job" % (wfo.name, lumisPerJob), level="critical"
)
wfh.sendLog("assignor", "%s was assigned with %s lumis/job" % (wfo.name, lumisPerJob))
parameters["LumisPerJob"] = lumisPerJob
else:
# sendEmail("leaving splitting untouched for PU_RD*","please check on "+wfo.name)
sendLog(
"assignor",
"leaving splitting untouched for %s, please check on %s" % (pstring, wfo.name),
level="critical",
)
wfh.sendLog("assignor", "leaving splitting untouched for PU_RD*, please check.")
result = reqMgrClient.assignWorkflow(url, wfo.name, team, parameters)
# set status
if not options.test:
if result:
wfo.status = "away"
session.commit()
n_assigned += 1
wfh.sendLog("assignor", "Properly assigned\n%s" % (json.dumps(parameters, indent=2)))
try:
## refetch information and lock output
new_wfi = workflowInfo(url, wfo.name)
(_, prim, _, sec) = new_wfi.getIO()
for secure in list(prim) + list(sec) + new_wfi.request["OutputDatasets"]:
## lock all outputs flat
NLI.lock(secure)
# for site in [SI.CE_to_SE(site) for site in sites_allowed]:
# for output in new_wfi.request['OutputDatasets']:
# LI.lock( output, site, 'dataset in production')
# for primary in prim:
# LI.lock( primary, site, 'dataset used in input')
# for secondary in sec:
# LI.lock( secondary, site, 'required for mixing' )
except Exception as e:
print "fail in locking output"
print str(e)
sendEmail("failed locking of output", str(e))
else:
print "ERROR could not assign", wfo.name
else:
pass
print "Assignment summary:"
sendLog("assignor", "Assigned %d Stalled %s" % (n_assigned, n_stalled))
def transferor(url ,specific = None, talk=True, options=None):
if userLock(): return
if duplicateLock(): return
use_mcm = True
up = componentInfo(mcm=use_mcm, soft=['mcm'])
if not up.check(): return
use_mcm = up.status['mcm']
if options and options.test:
execute = False
else:
execute = True
SI = siteInfo()
CI = campaignInfo()
NLI = newLockInfo()
mcm = McMClient(dev=False)
dss = DSS()
#allowed_secondary = UC.get('')
print "counting all being handled..."
being_handled = len(session.query(Workflow).filter(Workflow.status == 'away').all())
being_handled += len(session.query(Workflow).filter(Workflow.status.startswith('stag')).all())
being_transfered = len(session.query(Workflow).filter(Workflow.status == 'staging').all())
being_handled += len(session.query(Workflow).filter(Workflow.status.startswith('assistance-')).all())
max_to_handle = options.maxworkflows
max_to_transfer = options.maxstaging
allowed_to_handle = max(0,max_to_handle - being_handled)
allowed_to_transfer = max(0,max_to_transfer - being_transfered)
wf_buffer = 5
if allowed_to_handle<=wf_buffer: ## buffer for having several wf per transfer
print "Not allowed to run more than",max_to_handle,"at a time. Currently",being_handled,"and",wf_buffer,"buffer"
else:
print being_handled,"already being handled",max_to_handle,"max allowed,",allowed_to_handle,"remaining","and",wf_buffer,"buffer"
if allowed_to_transfer <= wf_buffer:
print "Not allowed to transfer more than",max_to_transfer,"at a time. Currently",being_transfered,"and",wf_buffer,"buffer"
else:
print being_transfered,"already being transfered",max_to_transfer,"max allowed,",allowed_to_transfer,"remaining","and",wf_buffer,"buffer"
print "... done"
all_transfers=defaultdict(list)
workflow_dependencies = defaultdict(set) ## list of wf.id per input dataset
wfs_and_wfh=[]
print "getting all wf to consider ..."
cache = getWorkflows(url, 'assignment-approved', details=True)
for wfo in session.query(Workflow).filter(Workflow.status.startswith('considered')).all():
print "\t",wfo.name
if specific and not specific in wfo.name: continue
cache_r =filter(lambda d:d['RequestName']==wfo.name, cache)
if len(cache_r):
wfs_and_wfh.append( (wfo, workflowInfo( url, wfo.name, spec=False, request = cache_r[0]) ) )
else:
wfs_and_wfh.append( (wfo, workflowInfo( url, wfo.name, spec=False) ) )
print "... done"
transfers_per_sites = defaultdict(int)
input_sizes = {}
ignored_input_sizes = {}
input_cput = {}
input_st = {}
## list the size of those in transfer already
in_transfer_priority=None
min_transfer_priority=None
print "getting all wf in staging ..."
stucks = json.loads(open('%s/stuck_transfers.json'%monitor_dir).read())
for wfo in session.query(Workflow).filter(Workflow.status=='staging').all():
wfh = workflowInfo( url, wfo.name, spec=False)
#(lheinput,primary,parent,secondary) = wfh.getIO()
#sites_allowed = getSiteWhiteList( (lheinput,primary,parent,secondary) )
(lheinput,primary,parent,secondary,sites_allowed) = wfh.getSiteWhiteList()
for site in sites_allowed: ## we should get the actual transfer destination instead of the full white list
transfers_per_sites[site] += 1
#input_cput[wfo.name] = wfh.getComputingTime()
#input_st[wfo.name] = wfh.getSystemTime()
for prim in primary:
ds_s = dss.get( prim )
if prim in stucks:
sendLog('transferor', "%s appears stuck, so not counting it %s [GB]"%( prim, ds_s), wfi=wfh)
ignored_input_sizes[prim] = ds_s
else:
input_sizes[prim] = ds_s
sendLog('transferor', "%s needs %s [GB]"%( wfo.name, ds_s), wfi=wfh)
if in_transfer_priority==None:
in_transfer_priority = int(wfh.request['RequestPriority'])
else:
in_transfer_priority = max(in_transfer_priority, int(wfh.request['RequestPriority']))
if min_transfer_priority==None:
min_transfer_priority = int(wfh.request['RequestPriority'])
else:
min_transfer_priority = min(min_transfer_priority, int(wfh.request['RequestPriority']))
if min_transfer_priority==None or in_transfer_priority ==None:
print "nothing is lining up for transfer"
sendEmail("no request in staging","no request in staging")
return
pass
try:
print "Ignored input sizes"
ignored_values = list(ignored_input_sizes.items())
ignored_values.sort( key = lambda i : i[1] )
print "\n".join( map(str, ignored_values ) )
print "Considered input sizes"
considered_values = list(input_sizes.items())
considered_values.sort( key = lambda i : i[1] )
print "\n".join( map(str, considered_values) )
except Exception as e:
print "trying to print the summary of input size"
print str(e)
print "... done"
print "Max priority in transfer already",in_transfer_priority
print "Min priority in transfer already",min_transfer_priority
print "transfers per sites"
print json.dumps( transfers_per_sites, indent=2)
in_transfer_already = sum(input_sizes.values())
cput_in_transfer_already = sum(input_cput.values())
st_in_transfer_already = sum(input_st.values())
## list the size of all inputs
primary_input_per_workflow_gb = defaultdict(float)
print "getting all input sizes ..."
for (wfo,wfh) in wfs_and_wfh:
(_,primary,_,_) = wfh.getIO()
#input_cput[wfo.name] = wfh.getComputingTime()
#input_st[wfo.name] = wfh.getSystemTime()
for prim in primary:
## do not count it if it appears stalled !
prim_size = dss.get( prim )
input_sizes[prim] = prim_size
primary_input_per_workflow_gb[wfo.name] += prim_size
print "... done"
# shuffle first by name
random.shuffle( wfs_and_wfh )
# Sort smallest transfers first; allows us to transfer as many as possible workflows.
def prio_and_size( i, j):
if int(i[1].request['RequestPriority']) == int(j[1].request['RequestPriority']):
return cmp(int(primary_input_per_workflow_gb.get(j[0].name, 0)), int(primary_input_per_workflow_gb.get(i[0].name, 0)) )
else:
return cmp(int(i[1].request['RequestPriority']),int(j[1].request['RequestPriority']))
#wfs_and_wfh.sort(cmp = prio_and_size, reverse=True)
#wfs_and_wfh.sort(cmp = lambda i,j : cmp(int(primary_input_per_workflow_gb.get(i[0].name, 0)), int(primary_input_per_workflow_gb.get(j[0].name, 0)) ))
#sort by priority higher first
wfs_and_wfh.sort(cmp = lambda i,j : cmp(int(i[1].request['RequestPriority']),int(j[1].request['RequestPriority']) ), reverse=True)
cput_grand_total = sum(input_cput.values())
cput_to_transfer = cput_grand_total - cput_in_transfer_already
st_grand_total = sum(input_st.values())
st_to_transfer = st_grand_total - st_in_transfer_already
print "%15.4f [CPU h] worth already in transfer"%cput_in_transfer_already
print "%15.4f [CPU h] worth is the current requested transfer load"%cput_to_transfer
print "%15.4f [h] worth of absolute system time in transfer"%( cput_in_transfer_already / SI.availableSlots())
print "%15.4f [h] worth of absolute system time is the current requested transfer load"%( cput_to_transfer / SI.availableSlots())
print "%15.4f [h] worth of theoritical system time in transfer"%( st_in_transfer_already )
print "%15.4f [h] worth of theoritical system time is the current requested transfer load"%( st_to_transfer )
grand_total = sum(input_sizes.values())
to_transfer = grand_total - in_transfer_already
grand_transfer_limit = options.maxtransfer
#grand_transfer_limit = SI.total_disk()*0.25*1024## half of the free sapce in TB->GB
transfer_limit = grand_transfer_limit - in_transfer_already
print "%15.4f GB already being transfered"%in_transfer_already
print "%15.4f GB is the current requested transfer load"%to_transfer
print "%15.4f GB is the global transfer limit"%grand_transfer_limit
print "%15.4f GB is the available limit"%transfer_limit
max_staging_per_site = options.maxstagingpersite
# the max priority value per dataset.
max_priority = defaultdict(int)
needs_transfer=0 ## so that we can count'em
passing_along = 0
transfer_sizes={}
went_over_budget=False
destination_cache = {}
no_goes = set()
max_per_round = UC.get('max_per_round').get('transferor',None)
if max_per_round and not spec:
wfs_and_wfh = wfs_and_wfh[:max_per_round]
for (wfo,wfh) in wfs_and_wfh:
print wfo.name,"to be transfered with priority",wfh.request['RequestPriority']
if wfh.request['RequestStatus']!='assignment-approved':
if wfh.request['RequestStatus'] in ['aborted','rejected','rejected-archived','aborted-archived']:
wfo.status = 'trouble' ## so that we look or a replacement
else:
wfo.status = 'away'
wfh.sendLog('transferor', '%s in status %s, setting %s'%( wfo.name,wfh.request['RequestStatus'],wfo.status))
continue
(_,primary,_,_) = wfh.getIO()
this_load=sum([input_sizes[prim] for prim in primary])
no_budget = False
if ( this_load and (sum(transfer_sizes.values())+this_load > transfer_limit or went_over_budget ) ):
if went_over_budget:
wfh.sendLog('transferor', "Transfer has gone over bubget.")
else:
wfh.sendLog('transferor', "Transfer will go over bubget.")
wfh.sendLog('transferor', "%15.4f GB this load, %15.4f GB already this round, %15.4f GB is the available limit"%(this_load, sum(transfer_sizes.values()), transfer_limit))
#if sum(transfer_sizes.values()) > transfer_limit:
went_over_budget = True
if in_transfer_priority!=None and min_transfer_priority!=None:
if int(wfh.request['RequestPriority']) >= in_transfer_priority and min_transfer_priority!=in_transfer_priority:
wfh.sendLog('transferor',"Higher priority sample %s >= %s go-on over budget"%( wfh.request['RequestPriority'], in_transfer_priority))
else:
if not options.go:
wfh.sendLog('transferor',"%s minimum priority %s < %s : stop"%( min_transfer_priority,wfh.request['RequestPriority'],in_transfer_priority))
no_budget = True
## throtlle by campaign go
no_go = False
if not wfh.go(log=True) and not options.go:
no_go = True
no_goes.add( wfo.name )
allowed_secondary = set()
for campaign in wfh.getCampaigns():
if campaign in CI.campaigns and 'secondaries' in CI.campaigns[campaign]:
allowed_secondary.update( CI.campaigns[campaign]['secondaries'] )
if secondary:
if (secondary and allowed_secondary) and (set(secondary)&allowed_secondary!=set(secondary)):
wfh.sendLog('assignor','%s is not an allowed secondary'%(', '.join(set(secondary)-allowed_secondary)))
no_go = True
if no_go:
continue
## check if the batch is announced
def check_mcm(wfn):
announced=False
is_real=False
if not wfn.startswith('pdmvserv'):
is_real = True
try:
for b in mcm.getA('batches',query='contains=%s'% wfo.name):
is_real = True
if b['status']=='announced':
announced=True
break
except:
try:
for b in mcm.getA('batches',query='contains=%s'% wfo.name):
is_real = True
if b['status']=='announced':
announced=True
break
except:
print "could not get mcm batch announcement, assuming not real"
return announced,is_real
if not use_mcm:
announced,is_real = False,True
else:
if wfh.request['RequestType'] in ['ReReco']:
announced,is_real = True,True
else:
announced,is_real = check_mcm( wfo.name )
if not announced:
wfh.sendLog('transferor', "does not look announced.")
if not is_real:
wfh.sendLog('transferor', "does not appear to be genuine.")
## prevent any duplication. if the wf is not mentioned in any batch, regardless of status
continue
## check on a grace period
injection_time = time.mktime(time.strptime('.'.join(map(str,wfh.request['RequestDate'])),"%Y.%m.%d.%H.%M.%S")) / (60.*60.)
now = time.mktime(time.gmtime()) / (60.*60.)
if float(now - injection_time) < 4.:
if not options.go and not announced:
wfh.sendLog('transferor', "It is too soon to start transfer: %3.2fH remaining"%(now - injection_time))
continue
if passing_along >= allowed_to_handle:
#if int(wfh.request['RequestPriority']) >= in_transfer_priority and min_transfer_priority!=in_transfer_priority:
if in_transfer_priority!=None and min_transfer_priority!=None:
if int(wfh.request['RequestPriority']) >= in_transfer_priority and int(wfh.request['RequestPriority']) !=min_transfer_priority:
## higher priority, and not only this priority being transfered
wfh.sendLog('transferor',"Higher priority sample %s >= %s go-on over %s"%( wfh.request['RequestPriority'], in_transfer_priority, max_to_handle))
else:
wfh.sendLog('transferor'," Not allowed to pass more than %s at a time. Currently %s handled, and adding %s"%( max_to_handle, being_handled, passing_along))
if not options.go:
## should not allow to jump that fence
break
if this_load and needs_transfer >= allowed_to_transfer:
if in_transfer_priority!=None and min_transfer_priority!=None:
if int(wfh.request['RequestPriority']) >= in_transfer_priority and int(wfh.request['RequestPriority']) !=min_transfer_priority:
## higher priority, and not only this priority being transfered
wfh.sendLog('transferor',"Higher priority sample %s >= %s go-on over %s"%(wfh.request['RequestPriority'], in_transfer_priority,max_to_transfer))
else:
wfh.sendLog('transferor',"Not allowed to transfer more than %s at a time. Currently %s transfering, and adding %s"%( max_to_transfer, being_transfered, needs_transfer))
if not options.go:
no_budget = True
if no_budget:
continue
## the site white list considers site, campaign, memory and core information
(lheinput,primary,parent,secondary,sites_allowed) = wfh.getSiteWhiteList()
if options and options.tosites:
sites_allowed = options.tosites.split(',')
for dataset in list(primary)+list(parent)+list(secondary):
## lock everything flat
NLI.lock( dataset )
if not sites_allowed:
wfh.sendLog('transferor',"not possible site to run at")
#sendEmail("no possible sites","%s has no possible sites to run at"%( wfo.name ))
sendLog('transferor',"%s has no possible sites to run at"%( wfo.name ),level='critical')
continue
blocks = []
if 'BlockWhitelist' in wfh.request and wfh.request['BlockWhitelist']:
blocks = wfh.request['BlockWhitelist']
if 'RunWhitelist' in wfh.request and wfh.request['RunWhitelist']:
## augment with run white list
for dataset in primary:
blocks = list(set( blocks + getDatasetBlocks( dataset, runs=wfh.request['RunWhitelist'] ) ))
if 'LumiList' in wfh.request and wfh.request['LumiList']:
## augment with the lumi white list
blocks = list(set( blocks + getDatasetBlocks( dataset, lumis= wfh.request['LumiList'] ) ))
if blocks:
print "Reading",len(blocks),"in block whitelist"
can_go = True
staging=False
allowed=True
primary_destinations = set()
if primary:
copies_needed_from_CPUh,CPUh = wfh.getNCopies()
if talk:
print wfo.name,'reads',', '.join(primary),'in primary'
## chope the primary dataset
for prim in primary:
## keep track of what needs what
workflow_dependencies[prim].add( wfo.id )
max_priority[prim] = max(max_priority[prim],int(wfh.request['RequestPriority']))
wfh.sendLog('transferor',"Would make %s from cpu requirement %s"%( copies_needed_from_CPUh, CPUh))
copies_needed = copies_needed_from_CPUh
if 'Campaign' in wfh.request and wfh.request['Campaign'] in CI.campaigns and 'maxcopies' in CI.campaigns[wfh.request['Campaign']]:
copies_needed_from_campaign = CI.campaigns[wfh.request['Campaign']]['maxcopies']
copies_needed = min(copies_needed_from_campaign, copies_needed)
wfh.sendLog('transferor',"Maxed to %s by campaign configuration %s"%( copies_needed, wfh.request['Campaign']))
### new ways of making the whole thing
destinations,all_block_names = getDatasetDestinations(url, prim, within_sites = [SI.CE_to_SE(site) for site in sites_allowed], only_blocks=blocks )
print json.dumps(destinations, indent=2)
## get where the dataset is in full and completed
prim_location = [site for (site,info) in destinations.items() if info['completion']==100 and info['data_fraction']==1]
## the rest is places it is going to be
prim_destination = [site for site in destinations.keys() if not site in prim_location]
if len(prim_location) >= copies_needed:
wfh.sendLog('transferor',"The input is all fully in place at %s sites %s"%( len(prim_location), sorted(prim_location)))
continue
copies_needed = max(0,copies_needed - len(prim_location))
wfh.sendLog('transferor',"not counting existing copies ; now need %s"% copies_needed)
copies_being_made = [ sum([info['blocks'].keys().count(block) for site,info in destinations.items() if site in prim_destination]) for block in all_block_names]
latching_on_transfers = set()
[latching_on_transfers.update(info['blocks'].values()) for site,info in destinations.items() if site in prim_destination]
latching_on_transfers = list(latching_on_transfers)
#print latching_on_transfers
## figure out where all this is going to go
prim_to_distribute = [site for site in sites_allowed if not SI.CE_to_SE(site) in prim_location]
prim_to_distribute = [site for site in prim_to_distribute if not SI.CE_to_SE(site) in prim_destination]
## take out the ones that cannot receive transfers
prim_to_distribute = [site for site in prim_to_distribute if not any([osite.startswith(site) for osite in SI.sites_veto_transfer])]
wfh.sendLog('transferor',"Could be going to: %s"% sorted( prim_to_distribute))
if not prim_to_distribute or any([transfers_per_sites[site] < max_staging_per_site for site in prim_to_distribute]):
## means there is openings let me go
print "There are transfer slots available:",[(site,transfers_per_sites[site]) for site in prim_to_distribute]
#for site in sites_allowed:
# #increment accross the board, regardless of real destination: could be changed
# transfers_per_sites[site] += 1
else:
if int(wfh.request['RequestPriority']) >= in_transfer_priority and min_transfer_priority!=in_transfer_priority:
wfh.sendLog('transferor', "Higher priority sample %s >= %s go-on over transfer slots available"%(wfh.request['RequestPriority'], in_transfer_priority))
else:
wfh.sendLog('transferor',"Not allowed to transfer more than %s per site at a time. Going overboard for %s"%( max_staging_per_site, sorted([site for site in prim_to_distribute if transfers_per_sites[site]>=max_staging_per_site])))
if not options.go:
allowed = False
break
for latching in latching_on_transfers:
tfo = session.query(Transfer).filter(Transfer.phedexid == int(latching)).first()
if not tfo:
tfo = session.query(Transfer).filter(Transfer.phedexid == -int(latching)).first()
if not tfo:
tfo = Transfer( phedexid = latching)
tfo.workflows_id = []
session.add(tfo)
else:
tfo.phedexid = latching ## make it positive ever
if not wfo.id in tfo.workflows_id:
print "adding",wfo.id,"to",tfo.id,"with phedexid",latching
l = copy.deepcopy( tfo.workflows_id )
l.append( wfo.id )
tfo.workflows_id = l
if not options.test:
session.commit()
else:
session.flush() ## regardless of commit later on, we need to let the next wf feeding on this transfer to see it in query
can_go = False
transfer_sizes[prim] = input_sizes[prim]
staging = True
# reduce the number of copies required by the on-going full transfer : how do we bootstrap on waiting for them ??
#copies_needed = max(0,copies_needed - len(prim_destination))
copies_needed = max(0,copies_needed - min(copies_being_made))
wfh.sendLog('transferor', "Not counting the copies being made ; then need %s"% copies_needed)
if copies_needed == 0:
wfh.sendLog('transferor', "The output is either fully in place or getting in full somewhere with %s"% latching_on_transfers)
can_go = True
continue
elif len(prim_to_distribute)==0:
wfh.sendLog('transferor', "We are going to need extra copies, but no destinations seems available")
prim_to_distribute = [site for site in sites_allowed if not SI.CE_to_SE(site) in prim_location]
prim_to_distribute = [site for site in prim_to_distribute if not any([osite.startswith(site) for osite in SI.sites_veto_transfer])]
if len(prim_to_distribute)>0: ## maybe that a parameter we can play with to limit the
if not options or options.chop:
### hard include the tape disk andpoint ?
#tapes = [site for site in getDatasetPresence( url, prim, vetos=['T0','T2','T3','Disk']) if site.endswith('MSS')]
chops,sizes = getDatasetChops(prim, chop_threshold = options.chopsize, only_blocks=blocks)
spreading = distributeToSites( chops, prim_to_distribute, n_copies = copies_needed, weights=SI.cpu_pledges, sizes=sizes)
transfer_sizes[prim] = sum(sizes)
if not spreading:
sendLog('transferor','cannot send %s to any site, it cannot fit anywhere'% prim, level='critical')
wfh.sendLog('transferor', "cannot send to any site. %s cannot seem to fit anywhere"%(prim))
staging=False
can_go = False
else:
spreading = {}
for site in prim_to_distribute:
if blocks:
spreading[site]=blocks
else:
spreading[site]=[prim]
transfer_sizes[prim] = input_sizes[prim] ## this is approximate if blocks are specified
can_go = False
wfh.sendLog('transferor', "selected CE destinations %s"%(sorted( spreading.keys())))
for (site,items) in spreading.items():
all_transfers[site].extend( items )
transfers_per_sites[site] += 1
primary_destinations.add( site )
if not allowed:
wfh.sendLog('transferor', "Not allowed to move on with")
continue
if secondary:
override_sec_destination = []
if 'SecondaryLocation' in CI.campaigns[wfh.request['Campaign']]:
override_sec_destination = CI.campaigns[wfh.request['Campaign']]['SecondaryLocation']
print wfo.name,'reads',', '.join(secondary),'in secondary'
for sec in secondary:
workflow_dependencies[sec].add( wfo.id )
if True:
## new style, failing on minbias
if not sec in destination_cache:
## this is barbbaric, and does not show the correct picture on workflow by workflow with different whitelist
destination_cache[sec],_ = getDatasetDestinations(url, sec) ## NO SITE WHITE LIST ADDED
#destination_cache[sec],_ = getDatasetDestinations(url, sec, within_sites = [SI.CE_to_SE(site) for site in sites_allowed])
## limit to the site whitelist NOW
se_allowed = [SI.CE_to_SE(site) for site in sites_allowed]
destinations = dict([(k,v) for (k,v) in destination_cache[sec].items() if site in se_allowed])
## truncate location/destination to those making up for >90% of the dataset
bad_destinations = [destinations.pop(site) for (site,info) in destinations.items() if info['data_fraction']<0.9]
sec_location = [site for (site,info) in destinations.items() if info['completion']>=95]
sec_destination = [site for site in destinations.keys() if not site in sec_location]
else:
## old style
presence = getDatasetPresence( url, sec )
sec_location = [site for site,pres in presence.items() if pres[1]>90.] ## more than 90% of the minbias at sites
subscriptions = listSubscriptions( url ,sec )
sec_destination = [site for site in subscriptions]
sec_to_distribute = [site for site in sites_allowed if not any([osite.startswith(site) for osite in sec_location])]
sec_to_distribute = [site for site in sec_to_distribute if not any([osite.startswith(site) for osite in sec_destination])]
sec_to_distribute = [site for site in sec_to_distribute if not any([osite.startswith(site) for osite in SI.sites_veto_transfer])]
if override_sec_destination:
## intersect with where we want the PU to be
not_needed_anymore = list(set(sec_to_distribute) - set(override_sec_destination))
#sendEmail("secondary superfluous","the dataset %s could be removed from %s"%( sec, not_needed_anymore ))
sendLog('transferor', "the dataset %s could be removed from %s"%( sec, not_needed_anymore ))
sec_to_distribute = list(set(sec_to_distribute) & set(override_sec_destination))
if len( sec_to_distribute )>0:
print "secondary could go to",sorted(sec_to_distribute)
sec_size = dss.get( sec )
for site in sec_to_distribute:
site_se =SI.CE_to_SE(site)
if (SI.disk[site_se]*1024.) > sec_size:
all_transfers[site].append( sec )
can_go = False
else:
print "could not send the secondary input to",site_se,"because it is too big for the available disk",SI.disk[site_se]*1024,"GB need",sec_size
if primary_destinations and site in primary_destinations:
#sendEmail('secondary input too big','%s is too big (%s) for %s (%s)'%( sec, sec_size, site_se, SI.disk[site_se]*1024))
sendLog('transferor', '%s is too big (%s) for %s (%s)'%( sec, sec_size, site_se, SI.disk[site_se]*1024), level='critical')
else:
print "the secondary input does not have to be send to site"
## is that possible to do something more
if can_go:
## no explicit transfer required this time
if staging:
## but using existing ones
wfh.sendLog('transferor', "latches on existing transfers, and nothing else, settin staging")
wfo.status = 'staging'
needs_transfer+=1
else:
wfh.sendLog('transferor', "should just be assigned now to %s"%sorted(sites_allowed))
wfo.status = 'staged'
passing_along+=1
wfh.sendLog('transferor', "setting status to %s"%wfo.status)
session.commit()
continue
else:
## there is an explicit transfer required
if staging:
## and also using an existing one
wfh.sendLog('transferor', "latches on existing transfers")
if not options.test:
wfo.status = 'staging'
wfh.sendLog('transferor', "setting status to %s"%wfo.status)
session.commit()
wfh.sendLog('transferor',"needs a transfer")
needs_transfer+=1
passing_along+=1
if no_goes:
#sendEmail("no go for managing","No go for \n"+"\n".join( no_goes ))
sendLog('transferor', "No go for \n"+"\n".join( no_goes ), level='critical')
print "accumulated transfers"
print json.dumps(all_transfers, indent=2)
fake_id=-1
wf_id_in_prestaging=set()
for (site,items_to_transfer) in all_transfers.iteritems():
items_to_transfer = list(set(items_to_transfer))
## convert to storage element
site_se = SI.CE_to_SE(site)
## site that do not want input datasets
if site in SI.sites_veto_transfer:
print site,"does not want transfers"
continue
## throttle the transfer size to T2s ? we'd be screwed by a noPU sample properly configured.
## massage a bit the items
blocks = [it for it in items_to_transfer if '#' in it]
block_datasets = list(set([it.split('#')[0] for it in blocks]))
datasets = [it for it in items_to_transfer if not '#' in it]
details_text = "Making a replica to %s (CE) %s (SE) for"%( site, site_se)
#print "\t",len(blocks),"blocks"
## remove blocks if full dataset is send out
blocks = [block for block in blocks if not block.split('#')[0] in datasets]
#print "\t",len(blocks),"needed blocks for",list(set([block.split('#')[0] for block in blocks]))
#print "\t",len(datasets),"datasets"
#print "\t",datasets
details_text += '\n\t%d blocks'%len(blocks)
details_text += '\n\t%d needed blocks for %s'%( len(blocks), sorted(list(set([block.split('#')[0] for block in blocks]))))
details_text += '\n\t%d datasets'% len(datasets)
details_text += '\n\t%s'%sorted(datasets)
items_to_transfer = blocks + datasets
if execute:
sendLog('transferor', details_text)
else:
print "Would make a replica to",site,"(CE)",site_se,"(SE) for"
print details_text
## operate the transfer
if options and options.stop:
## ask to move-on
answer = raw_input('Continue with that ?')
if not answer.lower() in ['y','yes','go']:
continue
if execute:
priority = 'normal'
cds = [ds for ds in datasets+block_datasets if ds in max_priority]
if cds and False: ## I don't think this is working. subscription should be updated on the fly and regularly for raising the priority if needed
## decide on an overall priority : that's a bit too large though
if any([max_priority[ds]>=90000 for ds in cds]):
priority = 'high'
elif all([max_priority[ds]<80000 for ds in cds]):
priority = 'low'
result = makeReplicaRequest(url, site_se, items_to_transfer, 'prestaging', priority=priority)
else:
result= {'phedex':{'request_created' : []}}
fake_id-=1
if not result:
print "ERROR Could not make a replica request for",site,items_to_transfer,"pre-staging"
continue
for phedexid in [o['id'] for o in result['phedex']['request_created']]:
new_transfer = session.query(Transfer).filter(Transfer.phedexid == int(phedexid)).first()
if not new_transfer:
new_transfer = session.query(Transfer).filter(Transfer.phedexid == -int(phedexid)).first()
print phedexid,"transfer created"
if not new_transfer:
new_transfer = Transfer( phedexid = phedexid)
session.add( new_transfer )
else:
new_transfer.phedexid = phedexid ## make it positive again
new_transfer.workflows_id = set()
for transfering in list(set(map(lambda it : it.split('#')[0], items_to_transfer))):
new_transfer.workflows_id.update( workflow_dependencies[transfering] )
new_transfer.workflows_id = list(new_transfer.workflows_id)
wf_id_in_prestaging.update(new_transfer.workflows_id)
session.commit()
## auto approve it
if execute:
approved = approveSubscription(url, phedexid, [site_se])
for wfid in wf_id_in_prestaging:
tr_wf = session.query(Workflow).get(wfid)
if tr_wf and tr_wf.status!='staging':
if execute:
tr_wf.status = 'staging'
if talk:
print "setting",tr_wf.name,"to staging"
session.commit()
def outcleanor(url, options):
if options.approve:
for user in ['*Vlimant']:#,'*Cremonesi']:
deletes = listDelete( url , user = user)
for (site,who,tid) in deletes:
if 'MSS' in site: continue### ever
print site,who,tid
print "approving deletion"
print approveSubscription(url, tid, nodes = [site], comments = 'Production cleaning by data ops')
return
sites_and_datasets = defaultdict(list)
our_copies = defaultdict(list)
wf_cleaned = {}
wfs = []
for fetch in options.fetch.split(','):
wfs.extend(session.query(Workflow).filter(Workflow.status==fetch).all())
random.shuffle( wfs )
last_answer = None
for wfo in wfs :
if options.number and len(wf_cleaned)>= options.number:
print "Reached",options.number,"cleaned"
break
print '-'*100
wfi = workflowInfo(url, wfo.name)
goes = {} # boolean per output
for dataset in wfi.request['OutputDatasets']:
goes[dataset] = False
keep_one_out = True
status = getDatasetStatus( dataset )
print "\n\tLooking at",dataset,status,"\n"
vetoes = None
if status == 'INVALID':
vetoes = ['Export','Buffer'] ## can take themselves out
keep_one_out = False # just wipe clean
elif status == None:
print dataset,"actually does not exist. skip"
goes[dataset] = True
continue
elif status in ['PRODUCTION','VALID'] and wfo.status in ['forget','trouble']:
print dataset,"should probably be invalidated. (",wfo.status,") skip"
keep_one_out = False # just wipe clean
continue ## you are not sure. just skip it for the time being
elif status == 'PRODUCTION' and wfo.status in ['clean']:
print dataset,"should probably be set valid .skip"
continue ## you are not sure. just skip it for the time being
if status == 'VALID' and dataset.startswith('/MinBias'):
print "This is a /MinBias. skip"
continue
if '/DQM' in dataset:
keep_one_out = False
total_size = getDatasetSize( dataset )
our_presence = getDatasetPresence(url, dataset, complete=None, group="DataOps", vetoes=vetoes)
also_our_presence = getDatasetPresence(url, dataset, complete=None, group="", vetoes=vetoes)
## merge in one unique dict
for site in also_our_presence:
if site in our_presence:
there,frac = our_presence[site]
other,ofrac = also_our_presence[site]
our_presence[site] = (max(there,other),max(frac,ofrac))
else:
our_presence[site] = also_our_presence[site]
if our_presence: print our_presence
## analysis ops copies need to be taken into account
anaops_presence = getDatasetPresence(url, dataset, complete=None, group="AnalysisOps")
own_by_anaops = anaops_presence.keys()
## all our copies
to_be_cleaned = our_presence.keys()
if not len(to_be_cleaned):
print "nowhere to be found of ours,",len(own_by_anaops),"in analysi ops pool"
goes[dataset] = True
continue
print "Where we own bits of dataset"
print to_be_cleaned
if len(own_by_anaops):
## remove site with the anaops copies
to_be_cleaned = list(set(to_be_cleaned) - set(own_by_anaops))
keep_one_out = False ## in that case, just remove our copies
print "Own by anaops (therefore not keep a copy of ours)"
print own_by_anaops
else:
## we should not be looking at anything that was not passed to DDM, otherwise we'll be cutting the grass under our feet
using_the_same = getWorkflowByInput(url, dataset, details=True)
conflict = False
for other in using_the_same:
if other['RequestName'] == wfo.name: continue
if other['RequestType'] == 'Resubmission': continue
if not other['RequestStatus'] in ['announced','normal-archived','aborted','rejected','aborted-archived','rejected-archived','closed-out','None',None]:
print other['RequestName'],'is in status',other['RequestStatus'],'preventing from cleaning',dataset
conflict=True
break
if conflict:
continue
## not being used. a bit less dangerous to clean-out
## keep one full copy out there
full_copies = [site for (site,(there,fract)) in our_presence.items() if there]
if keep_one_out:
if not len(full_copies):
print "we do not own a full copy of",dataset,status,wfo.status,".skip"
continue
stay_there = random.choice( full_copies ) #at a place own by ops
print "Where we keep a full copy", stay_there
to_be_cleaned.remove( stay_there )
our_copies[stay_there].append( dataset )
else:
print "We do not want to keep a copy of ",dataset,status,wfo.status
if len(to_be_cleaned):
print "Where we can clean"
print to_be_cleaned
for site in to_be_cleaned:
sites_and_datasets[site].append( (dataset, total_size*our_presence[site][1]/100., status) )
goes[dataset] = True
else:
print "no cleaning to be done"
goes[dataset] = True
print wfo.name,"scrutinized"
if all(goes.values()):
print "\t",wfo.name,"can toggle -out"
def ask():
global last_answer
last_answer = raw_input('go on ?')
return last_answer
if options.auto or ask() in ['y','']:
if all(goes.values()):
wfo.status = wfo.status+'-out'
wf_cleaned[wfo.name] = wfo.status
continue
elif last_answer in ['q','n']:
break
else:
return
if options.auto:
pass
elif last_answer in ['q']:
return
print "Potential cleanups"
for (site,items) in sites_and_datasets.items():
cleanup = sum([size for (_,size,_) in items])
print "\n\t potential cleanup of","%8.4f"%cleanup,"GB at ",site
print "\n".join([ds+" "+st for ds,_,st in items])
datasets = [ ds for ds,_,st in items]
print "Copies and bits we are going to delete"
print json.dumps( sites_and_datasets, indent=2)
print "Copies we are keeping"
print json.dumps( our_copies, indent=2 )
print "Workflows cleaned for output"
print json.dumps( wf_cleaned, indent=2 )
stamp = time.strftime("%Y%m%d%H%M%S", time.localtime())
open('outcleaning_%s.json'%stamp,'w').write( json.dumps( sites_and_datasets, indent=2))
open('keepcopies_%s.json'%stamp,'w').write( json.dumps( our_copies, indent=2))
open('wfcleanout_%s.json'%stamp,'w').write( json.dumps( wf_cleaned, indent=2))
if (not options.test) and (options.auto or raw_input("Satisfied ? (y will trigger status change and deletion requests)") in ['y']):
for (site,items) in sites_and_datasets.items():
datasets = [ ds for ds,_,st in items]
print "making deletion to",site
result = makeDeleteRequest(url, site, datasets, "Cleanup output after production. DataOps will take care of approving it.")
print result
## approve it right away ?
if 'MSS' in site: continue
if 'Export' in site: continue
if 'Buffer' in site: continue
for did in [item['id'] for item in result['phedex']['request_created']]:
print "auto-approve disabled, but ready"
#approveSubscription(url, did, nodes = [site], comments = 'Auto-approving production cleaning deletion')
pass
session.commit()
else:
print "Not making the deletion and changing statuses"
def equalizor(url, specific=None, options=None):
up = componentInfo(mcm=False, soft=['mcm'])
if not up.check(): return
if not specific:
workflows = getWorkflows(url, status='running-closed', details=True)
workflows.extend(getWorkflows(url, status='running-open',
details=True))
## start from scratch
modifications = defaultdict(dict)
## define regionality site => fallback allowed. feed on an ssb metric ??
mapping = defaultdict(list)
reversed_mapping = defaultdict(list)
regions = defaultdict(list)
SI = siteInfo()
for site in SI.sites_ready:
region = site.split('_')[1]
if not region in ['US', 'DE', 'IT']: continue
regions[region] = [region]
def site_in_depletion(s):
return True
if s in SI.sites_pressure:
(m, r, pressure) = SI.sites_pressure[s]
if float(m) < float(r):
print s, m, r, "lacking pressure"
return True
else:
print s, m, r, "pressure"
pass
return False
for site in SI.sites_ready:
region = site.split('_')[1]
## fallback to the region, to site with on-going low pressure
mapping[site] = [
fb for fb in SI.sites_ready
if any([('_%s_' %
(reg) in fb and fb != site and site_in_depletion(fb))
for reg in regions[region]])
]
use_T0 = False
if options.augment: use_T0 = True
use_HLT = False
if options.augment: use_HLT = True
if use_HLT:
mapping['T2_CH_CERN'].append('T2_CH_CERN_HLT')
#mapping['T2_IT_Legnaro'].append('T1_IT_CNAF')
for reg in ['IT', 'DE', 'UK']:
mapping['T2_CH_CERN'].extend(
[fb for fb in SI.sites_ready if '_%s_' % reg in fb])
for site, fallbacks in mapping.items():
for fb in fallbacks:
reversed_mapping[fb].append(site)
## this is the fallback mapping
print json.dumps(mapping, indent=2)
#print json.dumps( reversed_mapping, indent=2)
altered_tasks = set()
def running_idle(wfi, task_name):
gmon = wfi.getGlideMon()
#print gmon
if not gmon: return (0, 0)
if not task_name in gmon: return (0, 0)
return (gmon[task_name]['Running'], gmon[task_name]['Idle'])
def needs_action(wfi, task, min_idled=100, pressure=0.2):
task_name = task.pathName.split('/')[-1]
running, idled = running_idle(wfi, task_name)
go = True
if not idled and not running:
go = False
if idled < 100:
go = False
if (not running and idled) or (running and
(idled / float(running) > pressure)):
go = True
else:
go = False
return go, task_name, running, idled
def getcampaign(task):
taskname = task.pathName.split('/')[-1]
if hasattr(task, 'prepID'):
return task.prepID.split('-')[1]
elif taskname.count('-') >= 1:
return taskname.split('-')[1]
else:
return None
def close(interface):
open('%s/equalizor.json.new' % monitor_dir,
'w').write(json.dumps(interface, indent=2))
os.system('mv %s/equalizor.json.new %s/equalizor.json' %
(monitor_dir, monitor_dir))
os.system('cp %s/equalizor.json %s/logs/equalizor/equalizor.%s.json' %
(monitor_dir, monitor_dir, time.mktime(time.gmtime())))
interface = {'reversed_mapping': reversed_mapping, 'modifications': {}}
if options.augment or options.remove:
interface['modifications'] = json.loads(
open('%s/equalizor.json' % monitor_dir).read())['modifications']
if options.remove:
if specific in interface['modifications']:
print "poping", specific
interface['modifications'].pop(specific)
close(interface)
return
PU_locations = {}
PU_overflow = {
#'RunIISpring15PrePremix' : {
# 'sites' : ["T1_US_FNAL", "T1_DE_KIT" , "T1_IT_CNAF", "T1_RU_JINR" ,"T2_CH_CERN"],
# 'max' : 20000,
# 'pending' : 0
# },
'RunIIFall15DR76': {
'sites': [
'T1_ES_PIC', 'T2_US_Purdue', 'T2_UK_SGrid_RALPP', 'T2_BE_IIHE',
'T2_DE_DESY', 'T2_IT_Legnaro', 'T2_US_Caltech', 'T1_DE_KIT',
'T2_UK_London_Brunel', 'T2_IT_Pisa', 'T1_US_FNAL',
'T2_IT_Rome', 'T2_US_Florida', 'T1_IT_CNAF', 'T1_RU_JINR',
'T2_UK_London_IC', 'T2_US_Nebraska', 'T2_FR_CCIN2P3',
'T2_US_UCSD', 'T2_ES_CIEMAT', 'T1_FR_CCIN2P3',
'T2_US_Wisconsin', 'T2_US_MIT', 'T2_DE_RWTH', 'T1_UK_RAL',
'T2_US_Vanderbilt', 'T2_CH_CERN'
],
'max':
20000,
'pending':
0
},
'RunIISpring16DR80': {
'sites': [
'T1_ES_PIC', 'T2_US_Purdue', 'T2_UK_SGrid_RALPP', 'T2_BE_IIHE',
'T2_DE_DESY', 'T2_IT_Legnaro', 'T2_US_Caltech', 'T1_DE_KIT',
'T2_UK_London_Brunel', 'T2_IT_Pisa', 'T1_US_FNAL',
'T2_IT_Rome', 'T2_US_Florida', 'T1_IT_CNAF', 'T1_RU_JINR',
'T2_UK_London_IC', 'T2_US_Nebraska', 'T2_FR_CCIN2P3',
'T2_US_UCSD', 'T2_ES_CIEMAT', 'T1_FR_CCIN2P3',
'T2_US_Wisconsin', 'T2_US_MIT', 'T2_DE_RWTH', 'T1_UK_RAL',
'T2_US_Vanderbilt', 'T2_CH_CERN'
],
'max':
20000,
'pending':
0,
'force':
True
},
'RunIISpring15DR74': {
'sites': [
'T1_ES_PIC', 'T1_DE_KIT', 'T1_US_FNAL', 'T1_IT_CNAF',
'T1_RU_JINR', 'T1_FR_CCIN2P3', 'T1_UK_RAL', 'T2_CH_CERN'
],
'max':
20000,
'pending':
0
}
}
set_to = SI.sites_AAA
LHE_overflow = {
'RunIIWinter15GS': set_to,
'RunIISummer15GS': set_to,
'Summer12': set_to,
'Summer11Leg': set_to
#'RunIIFall15MiniAODv2' : set_to,
}
pending_HLT = 0
max_HLT = 60000
pending_T0 = 0
max_T0 = 60000
try:
gmon = json.loads(
os.popen(
'curl -s http://cms-gwmsmon.cern.ch/prodview/json/T2_CH_CERN_HLT'
).read())
pending_HLT += gmon["Running"]
pending_HLT += gmon["MatchingIdle"]
except:
pass
t0_special = [
'vlimant_BPH-RunIISummer15GS-00030_00212_v0__160129_135314_9755',
'pdmvserv_TSG-RunIISummer15GS-00044_00240_v0__160210_121223_8582'
]
no_routing = [
#'vlimant_BPH-RunIISummer15GS-00030_00212_v0__160129_135314_9755',
#'pdmvserv_TOP-RunIIWinter15GS-00074_00187_v0__160207_162312_1992',
]
stay_within_site_whitelist = False
specific_task = None
if specific and ":" in specific:
specific, specific_task = specific.split(':')
if specific:
wfs = session.query(Workflow).filter(
Workflow.name.contains(specific)).all()
else:
wfs = session.query(Workflow).filter(Workflow.status == 'away').all()
random.shuffle(wfs)
for wfo in wfs:
if wfo.name in no_routing and not options.augment:
continue
if specific and not specific in wfo.name:
continue
if specific:
wfi = workflowInfo(url, wfo.name)
else:
cached = filter(lambda d: d['RequestName'] == wfo.name, workflows)
if not cached: continue
wfi = workflowInfo(url, wfo.name, request=cached[0])
## only running should get re-routed
if not wfi.request['RequestStatus'] in [
'running-open', 'running-closed'
] and not specific:
continue
tasks_and_campaigns = []
for task in wfi.getWorkTasks():
tasks_and_campaigns.append((task, getcampaign(task)))
_, _, _, sec = wfi.getIO()
## check needs override
needs_overide = False
if not needs_overide and options.augment: needs_overide = True
def overide_from_agent(wfi, needs_overide):
bad_agents = [] #'http://cmssrv219.fnal.gov:5984']
if not bad_agents: return needs_overide
if needs_overide: return True
agents = wfi.getAgents()
wqss = ['Running', 'Acquired']
if any([
agent in agents.get(wqs, {}).keys()
for wqs, agent in itertools.product(wqss, bad_agents)
]):
print "overriding the need for bad agent"
needs_overide = True
return needs_overide
## now parse this for action
for i_task, (task, campaign) in enumerate(tasks_and_campaigns):
if options.augment:
print task.pathName
print campaign
### rule to avoid the issue of taskchain secondary jobs being stuck at sites processing the initial step
if campaign in LHE_overflow:
if task.taskType in ['Processing']:
needs, task_name, running, idled = needs_action(wfi, task)
needs_overide = overide_from_agent(wfi, needs_overide)
extend_to = copy.deepcopy(LHE_overflow[campaign])
if stay_within_site_whitelist:
extend_to = list(
set(extend_to) & set(wfi.request['SiteWhitelist'])
) ## restrict to stupid-site-whitelist
if extend_to and needs or needs_overide:
print "\t", task_name, "of", wfo.name, "running", running, "and pending", idled, "taking action : ReplaceSiteWhitelist"
modifications[wfo.name][task.pathName] = {
"ReplaceSiteWhitelist":
copy.deepcopy(LHE_overflow[campaign]),
"Running":
running,
"Pending":
idled,
"Priority":
wfi.request['RequestPriority']
}
#print json.dumps( modifications[wfo.name][task.pathName]['ReplaceSiteWhitelist']
altered_tasks.add(task.pathName)
else:
print task_name, "of", wfo.name, "running", running, "and pending", idled
### overflow the 76 digi-reco to the site holding the pileup
if campaign in PU_overflow:
force = PU_overflow[campaign][
'force'] if 'force' in PU_overflow[campaign] else False
secondary_locations = set(SI.sites_ready)
for s in sec:
if not s in PU_locations:
presence = getDatasetPresence(url, s)
#one_secondary_locations = [site for (site,(there,frac)) in presence.items() if there]
one_secondary_locations = [
site for (site, (there, frac)) in presence.items()
if frac > 98.
]
PU_locations[s] = one_secondary_locations
print "secondary is at", sorted(PU_locations[s])
secondary_locations = set(
PU_locations[s]) & secondary_locations
## we should add all sites that hold the secondary input if any
secondary_locations = list(
set(PU_overflow[campaign]['sites']) & set(SI.sites_ready))
if any([
task.pathName.endswith(finish)
for finish in ['_0', 'StepOneProc', 'Production']
]):
needs, task_name, running, idled = needs_action(wfi, task)
## removing the ones in the site whitelist already since they encode the primary input location
if stay_within_site_whitelist:
original_site_in_use = set(
wfi.request['SiteWhitelist'])
else:
original_site_in_use = set(secondary_locations)
## remove the sites that have already running jobs
gmon = wfi.getGlideMon()
if gmon and task_name in gmon and 'Sites' in gmon[
task_name]:
site_in_use = set(gmon[task_name]['Sites'])
## that determines where you want to run in addition
#augment_by = list((set(secondary_locations)- site_in_use))
augment_by = list(
(set(secondary_locations) - site_in_use)
& original_site_in_use
) ## restrict to stupid-site-whitelist
else:
augment_by = list(original_site_in_use)
needs_overide = overide_from_agent(wfi, needs_overide)
if augment_by and (
needs or needs_overide
or force) and PU_overflow[campaign][
'pending'] < PU_overflow[campaign]['max']:
PU_overflow[campaign]['pending'] += idled
print "raising overflow to", PU_overflow[campaign][
'pending'], "for", PU_overflow[campaign]['max']
## the step with an input ought to be the digi part : make this one go anywhere
modifications[wfo.name][task.pathName] = {
"AddWhitelist": augment_by,
"Running": running,
"Pending": idled,
"Priority": wfi.request['RequestPriority']
}
altered_tasks.add(task.pathName)
print "\t", task_name, "of", wfo.name, "running", running, "and pending", idled, "taking action : AddWhitelist"
#print json.dumps( augment_by, indent=2 )
else:
print task_name, "of", wfo.name, "running", running, "and pending", idled
### overflow the skims back to multi-core
if campaign in ['Run2015D', 'Run2015C_25ns'
] and task.taskType == 'Skim':
original_swl = wfi.request['SiteWhitelist']
needs, task_name, running, idled = needs_action(wfi, task)
if (needs or needs_overide):
modifications[wfo.name][task.pathName] = {
'AddWhitelist': original_swl,
"Running": running,
"Pending": idled,
"Priority": wfi.request['RequestPriority']
}
altered_tasks.add(task.pathName)
print "\t", task_name, "of", wfo.name, "running", running, "and pending", idled, "taking action : AddWhitelist"
if options.augment:
print sorted(wfi.request['SiteWhitelist']), i_task, use_HLT
### add the HLT at partner of CERN
if 'T2_CH_CERN' in wfi.request[
'SiteWhitelist'] and i_task == 0 and use_HLT:
needs, task_name, running, idled = needs_action(wfi, task)
if options.augment: needs = True
needs = True
##needs = random.random()<0.40 remove the random, just add up to a limit
if (needs or needs_overide) and pending_HLT < max_HLT:
pending_HLT += idled
if task.pathName in modifications[
wfo.name] and 'AddWhitelist' in modifications[
wfo.name][task.pathName]:
modifications[wfo.name][task.pathName][
"AddWhitelist"].append("T2_CH_CERN_HLT")
print "\t", wfo.name, "adding addHLT up to", pending_HLT, "for", max_HLT
print task.pathName
## this Replace does not work at all for HLT
#elif task.pathName in modifications[wfo.name] and 'ReplaceSiteWhitelist' in modifications[wfo.name][task.pathName]:
#modifications[wfo.name][task.pathName]["ReplaceSiteWhitelist"].append( "T2_CH_CERN_HLT" )
#print "\t",wfo.name,"adding replace HLT up to",pending_HLT,"for",max_HLT
else:
modifications[wfo.name][task.pathName] = {
"AddWhitelist": ["T2_CH_CERN_HLT"],
"Priority": wfi.request['RequestPriority'],
"Running": running,
"Pending": idled
}
print "\t", wfo.name, "adding HLT up to", pending_HLT, "for", max_HLT
print task.pathName
if i_task == 0 and not sec and use_T0:
needs, task_name, running, idled = needs_action(wfi, task)
if options.augment: needs = True
#needs = True
#if not (wfo.name in t0_special) and not options.augment: needs = False
if not wfi.request['RequestType'] in [
'MonteCarlo', 'MonteCarloFromGEN'
] and not options.augment:
needs = False
##needs = random.random()<0.40 remove the random, just add up to a limit
if (needs or needs_overide):
pending_T0 += idled
if task.pathName in modifications[
wfo.name] and 'AddWhitelist' in modifications[
wfo.name][task.pathName]:
modifications[wfo.name][
task.pathName]["AddWhitelist"].append("T0_CH_CERN")
print "\t", wfo.name, "adding addT0 up to", pending_T0, "for", max_T0
print task.pathName
elif task.pathName in modifications[
wfo.
name] and 'ReplaceSiteWhitelist' in modifications[
wfo.name][task.pathName]:
modifications[wfo.name][task.pathName][
"ReplaceSiteWhitelist"].append("T0_CH_CERN")
print "\t", wfo.name, "adding replace T0 up to", pending_T0, "for", max_T0
else:
modifications[wfo.name][task.pathName] = {
"AddWhitelist": ["T0_CH_CERN"],
"Priority": wfi.request['RequestPriority'],
"Running": running,
"Pending": idled
}
print "\t", wfo.name, "adding T0 up to", pending_T0, "for", max_T0
print task.pathName
interface['modifications'].update(modifications)
## temporary core managing
interface['cores'] = {
'T2_CH_CERN_HLT': {
'min': 4,
'max': 16
},
'default': {
'min': 1,
'max': 4
}
}
#interface['max_cores']={'T2_CH_CERN_HLT': 16, 'default': 4}
#interface['min_cores']={'T2_CH_CERN_HLT': 4, 'default': 1}
#interface['resize_subtasks'] = 'RunIISpring16DR80'
interface['resizes'] = ['RunIISpring16DR80', 'NotACampaign']
## close and save
close(interface)
def checkor(url, spec=None, options=None):
if userLock(): return
if duplicateLock() and not options.go: return
fDB = closeoutInfo()
UC = unifiedConfiguration()
use_mcm = True
up = componentInfo(mcm=use_mcm, soft=['mcm'])
if not up.check(): return
use_mcm = up.status['mcm']
def time_point(label="",sub_lap=False):
now = time.mktime(time.gmtime())
nows = time.asctime(time.gmtime())
print "Time check (%s) point at : %s"%(label, nows)
print "Since start: %s [s]"% ( now - time_point.start)
if sub_lap:
print "Sub Lap : %s [s]"% ( now - time_point.sub_lap )
time_point.sub_lap = now
else:
print "Lap : %s [s]"% ( now - time_point.lap )
time_point.lap = now
time_point.sub_lap = now
time_point.sub_lap = time_point.lap = time_point.start = time.mktime(time.gmtime())
runnings = session.query(Workflow).filter(Workflow.status == 'away').all()
standings = session.query(Workflow).filter(Workflow.status.startswith('assistance')).all()
## intersect with what is actually in completed status in request manager now
all_completed = set(getWorkflows(url, 'completed' ))
wfs=[]
if options.strict:
## the one which were running and now have completed
print "strict option is on: checking workflows that freshly completed"
wfs.extend( filter(lambda wfo: wfo.name in all_completed , runnings))
if options.update:
print "update option is on: checking workflows that have not completed yet"
wfs.extend( filter(lambda wfo: not wfo.name in all_completed , runnings))
if options.clear:
print "clear option is on: checking workflows that are ready to toggle closed-out"
wfs.extend( filter(lambda wfo: 'custodial' in wfo.status, standings))
if options.review:
print "review option is on: checking the workflows that needed intervention"
wfs.extend( filter(lambda wfo: not 'custodial' in wfo.status, standings))
## what is left out are the wf which were running and ended up aborted/failed/...
custodials = defaultdict(list) #sites : dataset list
transfers = defaultdict(list) #sites : dataset list
invalidations = [] #a list of files
SI = siteInfo()
CI = campaignInfo()
mcm = McMClient(dev=False) if use_mcm else None
def get_campaign(output, wfi):
## this should be a perfect matching of output->task->campaign
campaign = None
era = None
wf_campaign = None
if 'Campaign' in wfi.request: wf_campaign = wfi.request['Campaign']
try:
era = output.split('/')[2].split('-')[0]
except:
era = None
if wfi.isRelval():
campaign = wf_campaign
else:
campaign = era if era else wf_campaign
return campaign
## retrieve bypass and onhold configuration
bypasses = []
forcings = []
overrides = getForceCompletes()
holdings = []
actors = UC.get('allowed_bypass')
for bypassor,email in actors:
bypass_file = '/afs/cern.ch/user/%s/%s/public/ops/bypass.json'%(bypassor[0],bypassor)
if not os.path.isfile(bypass_file):
#sendLog('checkor','no file %s',bypass_file)
continue
try:
bypasses.extend( json.loads(open(bypass_file).read()))
except:
sendLog('checkor',"cannot get by-passes from %s for %s"%(bypass_file ,bypassor))
sendEmail("malformated by-pass information","%s is not json readable"%(bypass_file), destination=[email])
holding_file = '/afs/cern.ch/user/%s/%s/public/ops/onhold.json'%(bypassor[0],bypassor)
if not os.path.isfile(holding_file):
#sendLog('checkor',"no file %s"%holding_file)
continue
try:
extending = json.loads(open(holding_file).read())
print bypassor,"is holding",extending
holdings.extend( extending )
except:
sendLog('checkor',"cannot get holdings from %s for %s"%(holding_file, bypassor))
sendEmail("malformated by-pass information","%s is not json readable"%(holding_file), destination=[email])
## once this was force-completed, you want to bypass
for rider,email in actors:
rider_file = '/afs/cern.ch/user/%s/%s/public/ops/forcecomplete.json'%(rider[0],rider)
if not os.path.isfile(rider_file):
print "no file",rider_file
#sendLog('checkor',"no file %s"%rider_file)
continue
try:
bypasses.extend( json.loads(open( rider_file ).read() ) )
except:
sendLog('checkor',"cannot get force complete list from %s"%rider)
sendEmail("malformated force complet file","%s is not json readable"%rider_file, destination=[email])
if use_mcm:
forcings = mcm.get('/restapi/requests/forcecomplete')
#if forcings:
# sendEmail('force completing mechanism','please check what checkor is doing with %s'%( ','.join(forcings)))
pattern_fraction_pass = UC.get('pattern_fraction_pass')
total_running_time = 5.*60.
sleep_time = 1
if len(wfs):
sleep_time = min(max(0.5, total_running_time / len(wfs)), 10)
random.shuffle( wfs )
in_manual = 0
## now you have a record of what file was invalidated globally from TT
TMDB_invalid = dataCache.get('file_invalidation')
#try:
# TMDB_invalid = set([row[3] for row in csv.reader( os.popen('curl -s "https://docs.google.com/spreadsheets/d/11fFsDOTLTtRcI4Q3gXw0GNj4ZS8IoXMoQDC3CbOo_2o/export?format=csv"'))])
# TMDB_invalid = map(lambda e : e.split(':')[-1], TMDB_invalid)
# print len(TMDB_invalid),"globally invalidated files"
#except Exception as e:
# print "TMDB not fetched"
# print str(e)
# TMDB_invalid = []
print len(wfs),"to consider, pausing for",sleep_time
max_per_round = UC.get('max_per_round').get('checkor',None)
if options.limit: max_per_round=options.limit
if max_per_round and not spec: wfs = wfs[:max_per_round]
for wfo in wfs:
if spec and not (spec in wfo.name): continue
time.sleep( sleep_time )
time_point("Starting with %s"% wfo.name)
## get info
wfi = workflowInfo(url, wfo.name)
wfi.sendLog('checkor',"checking on %s %s"%( wfo.name,wfo.status))
## make sure the wm status is up to date.
# and send things back/forward if necessary.
wfo.wm_status = wfi.request['RequestStatus']
if wfo.wm_status == 'closed-out':
## manually closed-out
wfi.sendLog('checkor',"%s is already %s, setting close"%( wfo.name , wfo.wm_status))
wfo.status = 'close'
session.commit()
continue
elif wfo.wm_status in ['failed','aborted','aborted-archived','rejected','rejected-archived','aborted-completed']:
## went into trouble
wfo.status = 'trouble'
wfi.sendLog('checkor',"%s is in trouble %s"%(wfo.name, wfo.wm_status))
session.commit()
continue
elif wfo.wm_status in ['assigned','acquired']:
## not worth checking yet
wfi.sendLog('checkor',"%s is not running yet"%wfo.name)
session.commit()
continue
if '-onhold' in wfo.status:
if wfo.name in holdings and wfo.name not in bypasses:
wfi.sendLog('checkor',"%s is on hold"%wfo.name)
continue
if wfo.wm_status != 'completed': #and not wfo.name in bypasses:
## for sure move on with closeout check if in completed
wfi.sendLog('checkor',"no need to check on %s in status %s"%(wfo.name, wfo.wm_status))
session.commit()
continue
if wfo.name in holdings and wfo.name not in bypasses:
wfo.status = 'assistance-onhold'
wfi.sendLog('checkor',"setting %s on hold"%wfo.name)
session.commit()
continue
session.commit()
#sub_assistance="" # if that string is filled, there will be need for manual assistance
existing_assistance_tags = set(wfo.status.split('-')[1:]) #[0] should be assistance
assistance_tags = set()
is_closing = True
## get it from somewhere
bypass_checks = False
for bypass in bypasses:
if bypass in wfo.name:
wfi.sendLog('checkor',"we can bypass checks on %s because of keyword %s "%( wfo.name, bypass))
bypass_checks = True
break
pids = wfi.getPrepIDs()
force_by_mcm = False
force_by_user = False
for force in forcings:
if force in pids:
wfi.sendLog('checkor',"we can bypass checks and force complete %s because of prepid %s "%( wfo.name, force))
bypass_checks = True
force_by_mcm = True
break
for user in overrides:
for force in overrides[user]:
if force in wfo.name:
wfi.sendLog('checkor',"we can bypass checks and force complete %s because of keyword %s of user %s"%( wfo.name, force, user))
bypass_checks = True
force_by_user = True
break
tiers_with_no_check = copy.deepcopy(UC.get('tiers_with_no_check')) # dqm*
vetoed_custodial_tier = copy.deepcopy(UC.get('tiers_with_no_custodial')) #dqm*, reco
to_ddm_tier = copy.deepcopy(UC.get('tiers_to_DDM'))
campaigns = {} ## this mapping of campaign per output dataset assumes era==campaing, which is not true for relval
expected_outputs = copy.deepcopy( wfi.request['OutputDatasets'] )
for out in wfi.request['OutputDatasets']:
c = get_campaign(out, wfi)
campaigns[out] = c
if c in CI.campaigns and 'custodial_override' in CI.campaigns[c]:
vetoed_custodial_tier = list(set(vetoed_custodial_tier) - set(CI.campaigns[c]['custodial_override']))
## add those that we need to check for custodial copy
tiers_with_no_check = list(set(tiers_with_no_check) - set(CI.campaigns[c]['custodial_override'])) ## would remove DQM from the vetoed check
check_output_text = "Initial outputs:"+",".join(sorted(wfi.request['OutputDatasets'] ))
wfi.request['OutputDatasets'] = [ out for out in wfi.request['OutputDatasets'] if not any([out.split('/')[-1] == veto_tier for veto_tier in tiers_with_no_check])]
check_output_text += "\nWill check on:"+",".join(sorted(wfi.request['OutputDatasets'] ))
check_output_text += "\ntiers out:"+",".join( sorted(tiers_with_no_check ))
check_output_text += "\ntiers no custodial:"+",".join( sorted(vetoed_custodial_tier) )
wfi.sendLog('checkor', check_output_text )
## anything running on acdc : getting the real prepid is not worth it
familly = getWorkflowById(url, wfi.request['PrepID'], details=True)
acdc = []
acdc_inactive = []
forced_already=False
acdc_bads = []
true_familly = []
for member in familly:
if member['RequestType'] != 'Resubmission': continue
if member['RequestName'] == wfo.name: continue
if member['RequestDate'] < wfi.request['RequestDate']: continue
if member['PrepID'] != wfi.request['PrepID'] : continue
#if 'OriginalRequestName' in member and (not 'ACDC' in member['OriginalRequestName']) and member['OriginalRequestName'] != wfo.name: continue
if member['RequestStatus'] == None: continue
if not set(member['OutputDatasets']).issubset( set(expected_outputs)):
if not member['RequestStatus'] in ['rejected-archived','rejected','aborted','aborted-archived']:
##this is not good at all
wfi.sendLog('checkor','inconsistent ACDC %s'%member['RequestName'] )
#sendLog('checkor','inconsistent ACDC %s'%member['RequestName'], level='critical')
acdc_bads.append( member['RequestName'] )
is_closing = False
assistance_tags.add('manual')
continue
true_familly.append( member['RequestName'] )
#try:
# parse_one(url, member['RequestName'])
#except:
# print "Could not make error report for",member['RequestName']
if member['RequestStatus'] in ['running-open','running-closed','assigned','acquired']:
print wfo.name,"still has an ACDC running",member['RequestName']
acdc.append( member['RequestName'] )
## cannot be bypassed!
is_closing = False
assistance_tags.add('recovering')
if (force_by_mcm or force_by_user) and not forced_already:
wfi.sendLog('checkor','%s is being forced completed while recovering'%wfo.name)
wfi.notifyRequestor("The workflow %s was force completed"% wfo.name, do_batch=False)
forceComplete(url, wfi)
forced_already=True
else:
acdc_inactive.append( member['RequestName'] )
assistance_tags.add('recovered')
if acdc_bads:
#sendEmail('inconsistent ACDC','for %s, ACDC %s is inconsistent, preventing from closing'%( wfo.name, ','.join(acdc_bads) ))
sendLog('checkor','For %s, ACDC %s is inconsistent, preventing from closing or will create a mess.'%( wfo.name, ','.join(acdc_bads) ), level='critical')
time_point("checked workflow familly", sub_lap=True)
## completion check
percent_completions = {}
if not 'TotalInputEvents' in wfi.request:
event_expected,lumi_expected = 0,0
if not 'recovery' in wfo.status:
#sendEmail("missing member of the request","TotalInputEvents is missing from the workload of %s"% wfo.name, destination=['*****@*****.**'])
sendLog('checkor',"TotalInputEvents is missing from the workload of %s"% wfo.name, level='critical')
else:
event_expected,lumi_expected = wfi.request['TotalInputEvents'],wfi.request['TotalInputLumis']
if 'RequestNumEvents' in wfi.request and int(wfi.request['RequestNumEvents']):
event_expected = int(wfi.request['RequestNumEvents'])
elif 'Task1' in wfi.request and 'RequestNumEvents' in wfi.request['Task1']:
event_expected = wfi.request['Task1']['RequestNumEvents']
for i in range(1,20):
if 'Task%d'%i in wfi.request:
## this is wrong ibsolute
if 'FilterEfficiency' in wfi.request['Task%d'%i]:
event_expected *= float(wfi.request['Task%d'%i]['FilterEfficiency'])
event_expected = int(event_expected)
fractions_pass = {}
events_per_lumi = {}
over_100_pass = False
(lhe,prim,_,_) = wfi.getIO()
if lhe or prim: over_100_pass = False
time_point("execpted statistics", sub_lap=True)
for output in wfi.request['OutputDatasets']:
event_count,lumi_count = getDatasetEventsAndLumis(dataset=output)
events_per_lumi[output] = event_count/float(lumi_count) if lumi_count else 100
percent_completions[output] = 0.
if lumi_expected:
percent_completions[output] = lumi_count / float( lumi_expected )
if event_expected:
wfi.sendLog('checkor', "event completion real %s expected %s"%(event_count, event_expected ))
percent_completions[output] = max(percent_completions[output], float(event_count) / float( event_expected ) )
default_pass = UC.get('default_fraction_pass')
fractions_pass[output] = default_pass
c = campaigns[output]
if c in CI.campaigns and 'fractionpass' in CI.campaigns[c]:
if type(CI.campaigns[c]['fractionpass']) == dict:
tier = output.split('/')[-1]
priority = str(wfi.request['RequestPriority'])
## defined per tier
fractions_pass[output] = CI.campaigns[c]['fractionpass'].get('all', default_pass)
if tier in CI.campaigns[c]['fractionpass']:
fractions_pass[output] = CI.campaigns[c]['fractionpass'][tier]
if priority in CI.campaigns[c]['fractionpass']:
fractions_pass[output] = CI.campaigns[c]['fractionpass'][priority]
else:
fractions_pass[output] = CI.campaigns[c]['fractionpass']
wfi.sendLog('checkor', "overriding fraction to %s for %s by campaign requirement"%( fractions_pass[output], output))
if options.fractionpass:
fractions_pass[output] = options.fractionpass
print "overriding fraction to",fractions_pass[output],"by command line for",output
for key in pattern_fraction_pass:
if key in output:
fractions_pass[output] = pattern_fraction_pass[key]
print "overriding fraction to",fractions_pass[output],"by dataset key",key
if not all([percent_completions[out] >= fractions_pass[out] for out in fractions_pass]):
possible_recoveries = wfi.getRecoveryDoc()
if possible_recoveries == []:
wfi.sendLog('checkor','%s has missing statistics \n%s \n%s, but nothing is recoverable. passing through to annoucement'%(
wfo.name, json.dumps(percent_completions, indent=2), json.dumps(fractions_pass, indent=2) ))
sendLog('checkor','%s is not completed, but has nothing to be recovered, passing along ?'%wfo.name, level='critical')
#sendEmail('nothing is recoverable','%s is not completed, but has nothing to be recovered, passing along ?'%wfo.name)#,destination=['*****@*****.**'])
## do not bypass for now, until Alan understands why we are loosing ACDC docs
bypass_checks = True
else:
wfi.sendLog('checkor','%s is not completed \n%s \n%s'%(
wfo.name, json.dumps(percent_completions, indent=2), json.dumps(fractions_pass, indent=2) ))
## hook for creating automatically ACDC ?
if not bypass_checks:
assistance_tags.add('recovery')
is_closing = False
if over_100_pass and any([percent_completions[out] >100 for out in fractions_pass]):
print wfo.name,"is over completed"
print json.dumps(percent_completions, indent=2)
if not bypass_checks:
assistance_tags.add('over100')
is_closing = False
time_point("checked output size", sub_lap=True)
## correct lumi < 300 event per lumi
#for output in wfi.request['OutputDatasets']:
#events_per_lumi[output] = getDatasetEventsPerLumi( output )
lumi_upper_limit = {}
for output in wfi.request['OutputDatasets']:
upper_limit = 301.
campaign = campaigns[output]
if campaign in CI.campaigns and 'lumisize' in CI.campaigns[campaign]:
upper_limit = CI.campaigns[campaign]['lumisize']
print "overriding the upper lumi size to",upper_limit,"for",campaign
if options.lumisize:
upper_limit = options.lumisize
print "overriding the upper lumi size to",upper_limit,"by command line"
lumi_upper_limit[output] = upper_limit
if wfi.request['RequestType'] in ['ReDigi','ReReco']: lumi_upper_limit[output] = -1
if any([ (lumi_upper_limit[out]>0 and events_per_lumi[out] >= lumi_upper_limit[out]) for out in events_per_lumi]):
print wfo.name,"has big lumisections"
print json.dumps(events_per_lumi, indent=2)
## hook for rejecting the request ?
if not bypass_checks:
assistance_tags.add('biglumi')
is_closing = False
any_presence = {}
for output in wfi.request['OutputDatasets']:
any_presence[output] = getDatasetPresence(url, output, vetoes=[])
time_point("checked dataset presence", sub_lap=True)
## custodial copy
custodial_locations = {}
custodial_presences = {}
for output in wfi.request['OutputDatasets']:
custodial_presences[output] = [s for s in any_presence[output] if 'MSS' in s]
custodial_locations[output] = phedexClient.getCustodialSubscriptionRequestSite(output)
if not custodial_locations[output]:
custodial_locations[output] = []
time_point("checked custodiality", sub_lap=True)
## presence in phedex
phedex_presence ={}
for output in wfi.request['OutputDatasets']:
phedex_presence[output] = phedexClient.getFileCountDataset(url, output )
time_point("checked phedex count", sub_lap=True)
out_worth_checking = [out for out in custodial_locations.keys() if out.split('/')[-1] not in vetoed_custodial_tier]
size_worth_checking = sum([getDatasetSize(out)/1023. for out in out_worth_checking ]) ## size in TBs of all outputs
size_worht_going_to_ddm = sum([getDatasetSize(out)/1023. for out in out_worth_checking if out.split('/')[-1] in to_ddm_tier ]) ## size in TBs of all outputs
if not all(map( lambda sites : len(sites)!=0, [custodial_locations[out] for out in out_worth_checking])):
print wfo.name,"has not all custodial location"
print json.dumps(custodial_locations, indent=2)
##########
## hook for making a custodial replica ?
custodial = None
## get from other outputs
for output in out_worth_checking:
if len(custodial_locations[output]):
custodial = custodial_locations[output][0]
if custodial and float(SI.storage[custodial]) < size_worth_checking:
print "cannot use the other output custodial:",custodial,"because of limited space"
custodial = None
## try to get it from campaign configuration
if not custodial:
for output in out_worth_checking:
campaign = campaigns[output]
if campaign in CI.campaigns and 'custodial' in CI.campaigns[campaign]:
custodial = CI.campaigns[campaign]['custodial']
print "Setting custodial to",custodial,"from campaign configuration"
group = None
if campaign in CI.campaigns and 'phedex_group' in CI.campaigns[campaign]:
group = CI.campaigns[campaign]['phedex_group']
print "using group",group,"for replica"
if custodial and float(SI.storage[custodial]) < size_worth_checking:
print "cannot use the campaign configuration custodial:",custodial,"because of limited space"
custodial = None
## get from the parent
pick_custodial = True
use_parent_custodial = UC.get('use_parent_custodial')
tape_size_limit = options.tape_size_limit if options.tape_size_limit else UC.get("tape_size_limit")
_,prim,_,_ = wfi.getIO()
if not custodial and prim and use_parent_custodial:
parent_dataset = prim.pop()
## this is terribly dangerous to assume only
parents_custodial = phedexClient.getCustodialSubscriptionRequestSite( parent_dataset )
###parents_custodial = findCustodialLocation(url, parent_dataset)
if not parents_custodial:
parents_custodial = []
if len(parents_custodial):
custodial = parents_custodial[0]
else:
print "the input dataset",parent_dataset,"does not have custodial in the first place. abort"
#sendEmail( "dataset has no custodial location", "Please take a look at %s in the logs of checkor"%parent_dataset)
## does not work for RAWOADSIM
sendLog('checkor',"Please take a look at %s for missing custodial location"% parent_dataset)
## cannot be bypassed, this is an issue to fix
is_closing = False
pick_custodial = False
assistance_tags.add('parentcustodial')
if custodial and float(SI.storage[custodial]) < size_worth_checking:
print "cannot use the parent custodial:",custodial,"because of limited space"
custodial = None
if not custodial and pick_custodial:
## pick one at random
custodial = SI.pick_SE(size=size_worth_checking)
if custodial and size_worht_going_to_ddm > tape_size_limit:
print wfi.sendLog('checkor',"The total output size (%s TB) is too large for the limit set (%s TB)"%( size_worth_checking, tape_size_limit))
custodial = None
if not custodial:
print "cannot find a custodial for",wfo.name
wfi.sendLog('checkor',"cannot find a custodial for %s probably because of the total output size %d"%( wfo.name, size_worth_checking))
sendLog('checkor',"cannot find a custodial for %s probably because of the total output size %d"%( wfo.name, size_worth_checking), level='critical')
if custodial and (is_closing or bypass_checks):
print "picked",custodial,"for tape copy"
## remember how much you added this round already ; this stays locally
SI.storage[custodial] -= size_worth_checking
## register the custodial request, if there are no other big issues
for output in out_worth_checking:
if not len(custodial_locations[output]):
if phedex_presence[output]>=1:
wfi.sendLog('checkor','Using %s as a tape destination for %s'%(custodial, output))
custodials[custodial].append( output )
if group: custodials[custodial][-1]+='@%s'%group
## let's wait and see if that's needed
assistance_tags.add('custodial')
else:
print "no file in phedex for",output," not good to add to custodial requests"
#cannot be bypassed
is_closing = False
time_point("determined tape location", sub_lap=True)
## disk copy
disk_copies = {}
for output in wfi.request['OutputDatasets']:
disk_copies[output] = [s for s in any_presence[output] if (not 'MSS' in s) and (not 'Buffer' in s)]
if not all(map( lambda sites : len(sites)!=0, disk_copies.values())):
print wfo.name,"has not all output on disk"
print json.dumps(disk_copies, indent=2)
## presence in dbs
dbs_presence = {}
dbs_invalid = {}
for output in wfi.request['OutputDatasets']:
dbs_presence[output] = dbs3Client.getFileCountDataset( output )
dbs_invalid[output] = dbs3Client.getFileCountDataset( output, onlyInvalid=True)
time_point("dbs file count", sub_lap=True)
if not all([dbs_presence[out] == (dbs_invalid[out]+phedex_presence[out]) for out in wfi.request['OutputDatasets']]) and not options.ignorefiles:
mismatch_notice = wfo.name+" has a dbs,phedex mismatch\n"
mismatch_notice += "in dbs\n"+json.dumps(dbs_presence, indent=2) +"\n"
mismatch_notice += "invalide in dbs\n"+json.dumps(dbs_invalid, indent=2) +"\n"
mismatch_notice += "in phedex\n"+json.dumps(phedex_presence, indent=2) +"\n"
wfi.sendLog('checkor',mismatch_notice)
if not 'recovering' in assistance_tags:
assistance_tags.add('filemismatch')
#print this for show and tell if no recovery on-going
for out in dbs_presence:
_,_,missing_phedex,missing_dbs = getDatasetFiles(url, out)
if missing_phedex:
wfi.sendLog('checkor',"These %d files are missing in phedex\n%s"%(len(missing_phedex),
"\n".join( missing_phedex )))
were_invalidated = sorted(set(missing_phedex) & set(TMDB_invalid ))
if were_invalidated:
wfi.sendLog('checkor',"These %d files were invalidated globally\n%s"%(len(were_invalidated),
"\n".join(were_invalidated)))
sendLog('checkor',"These %d files were invalidated globally\n%s\nand are invalidated in dbs"%(len(were_invalidated),
"\n".join(were_invalidated)), level='critical')
dbs3Client.setFileStatus( were_invalidated, newstatus=0 )
if missing_dbs:
wfi.sendLog('checkor',"These %d files are missing in dbs\n%s"%(len(missing_dbs),
"\n".join( missing_dbs )))
were_invalidated = sorted(set(missing_dbs) & set(TMDB_invalid ))
if were_invalidated:
wfi.sendLog('checkor',"These %d files were invalidated globally\n%s"%(len(were_invalidated),
"\n".join(were_invalidated)))
#if not bypass_checks:
## I don't think we can by pass this
is_closing = False
time_point("checked file count", sub_lap=True)
fraction_invalid = 0.20
if not all([(dbs_invalid[out] <= int(fraction_invalid*dbs_presence[out])) for out in wfi.request['OutputDatasets']]) and not options.ignoreinvalid:
print wfo.name,"has a dbs invalid file level too high"
print json.dumps(dbs_presence, indent=2)
print json.dumps(dbs_invalid, indent=2)
print json.dumps(phedex_presence, indent=2)
## need to be going and taking an eye
assistance_tags.add('invalidfiles')
if not bypass_checks:
#sub_assistance+="-invalidfiles"
is_closing = False
## put that heavy part at the end
## duplication check
duplications = {}
files_per_rl = {}
for output in wfi.request['OutputDatasets']:
duplications[output] = "skiped"
files_per_rl[output] = "skiped"
time_point("checked invalidation", sub_lap=True)
if (is_closing or bypass_checks) and (not options.ignoreduplicates):
print "starting duplicate checker for",wfo.name
for output in wfi.request['OutputDatasets']:
print "\tchecking",output
duplications[output] = True
try:
duplications[output],files_per_rl[output] = dbs3Client.duplicateRunLumiFiles( output , skipInvalid=True, verbose=True)
except:
try:
duplications[output],files_per_rl[output] = dbs3Client.duplicateRunLumiFiles( output , skipInvalid=True, verbose=True)
except Exception as e:
wfi.sendLog('checkor','Not possible to check on duplicate lumi count on %s'%(output))
sendLog('checkor','Not possible to check on duplicate lumi count on %s\n%s'%(output,str(e)),level='critical')
is_closing=False
if is_closing and any(duplications.values()) and not options.ignoreduplicates:
duplicate_notice = ""
duplicate_notice += "%s has duplicates\n"%wfo.name
duplicate_notice += json.dumps( duplications,indent=2)
duplicate_notice += '\n'
duplicate_notice += json.dumps( files_per_rl, indent=2)
wfi.sendLog('checkor',duplicate_notice)
## hook for making file invalidation ?
## it shouldn't be allowed to bypass it
assistance_tags.add('duplicates')
is_closing = False
time_point("checked duplicates", sub_lap=True)
time_point("done with %s"%wfo.name)
## for visualization later on
if not wfo.name in fDB.record:
#print "adding",wfo.name,"to close out record"
fDB.record[wfo.name] = {
'datasets' :{},
'name' : wfo.name,
'closeOutWorkflow' : None,
}
fDB.record[wfo.name]['closeOutWorkflow'] = is_closing
fDB.record[wfo.name]['priority'] = wfi.request['RequestPriority']
fDB.record[wfo.name]['prepid'] = wfi.request['PrepID']
for output in wfi.request['OutputDatasets']:
if not output in fDB.record[wfo.name]['datasets']: fDB.record[wfo.name]['datasets'][output] = {}
rec = fDB.record[wfo.name]['datasets'][output]
#rec['percentage'] = float('%.2f'%(percent_completions[output]*100))
rec['percentage'] = math.floor(percent_completions[output]*10000)/100.## round down
rec['duplicate'] = duplications[output] if output in duplications else 'N/A'
rec['phedexReqs'] = float('%.2f'%any_presence[output][custodial_presences[output][0]][1]) if len(custodial_presences[output])!=0 else 'N/A'
rec['closeOutDataset'] = is_closing
rec['transPerc'] = float('%.2f'%any_presence[output][ disk_copies[output][0]][1]) if len(disk_copies[output])!=0 else 'N/A'
rec['correctLumis'] = int(events_per_lumi[output]) if (events_per_lumi[output] > lumi_upper_limit[output]) else True
rec['missingSubs'] = False if len(custodial_locations[output])==0 else ','.join(list(set(custodial_locations[output])))
rec['dbsFiles'] = dbs_presence[output]
rec['dbsInvFiles'] = dbs_invalid[output]
rec['phedexFiles'] = phedex_presence[output]
rec['acdc'] = "%d / %d"%(len(acdc),len(acdc+acdc_inactive))
rec['familly'] = true_familly
now = time.gmtime()
rec['timestamp'] = time.mktime(now)
rec['updated'] = time.asctime(now)+' (GMT)'
## make the lumi summary
if wfi.request['RequestType'] == 'ReReco':
try:
os.system('python Unified/lumi_summary.py %s 1 > /dev/null'%(wfi.request['PrepID']))
os.system('python Unified/lumi_plot.py %s > /dev/null'%(wfi.request['PrepID']))
wfi.sendLog('checkor','Lumi summary available at %s/datalumi/lumi.%s.html'%(unified_url,wfi.request['PrepID']))
except Exception as e:
print str(e)
## make the error report
## and move on
if is_closing:
## toggle status to closed-out in request manager
wfi.sendLog('checkor',"setting %s closed-out"% wfo.name)
if not options.test:
if wfo.wm_status in ['closed-out','announced','normal-archived']:
print wfo.name,"is already",wfo.wm_status,"not trying to closed-out and assuming it does"
res = None
else:
res = reqMgrClient.closeOutWorkflowCascade(url, wfo.name)
print "close out answer",res
if not res in ["None",None]:
print "try to get the current status again"
wfi_bis = workflowInfo(url, wfo.name)
if wfi_bis.request['RequestStatus'] == 'closed-out':
print "the request did toggle to closed-out"
res = None
if not res in ["None",None]:
print "retrying to closing out"
print res
res = reqMgrClient.closeOutWorkflowCascade(url, wfo.name)
if res in [None,"None"]:
wfo.status = 'close'
session.commit()
if use_mcm and force_by_mcm:
## shoot large on all prepids, on closing the wf
for pid in pids:
mcm.delete('/restapi/requests/forcecomplete/%s'%pid)
else:
print "could not close out",wfo.name,"will try again next time"
else:
if not 'custodial' in assistance_tags or wfi.isRelval():
## do only the report for those
for member in acdc+acdc_inactive+[wfo.name]:
try:
parse_one(url, member)
except:
print "Could not make error report for",member
## full known list
#recovering # has active ACDC
##OUT #recovered #had inactive ACDC
#recovery #not over the pass bar
#over100 # over 100%
#biglumi # has a big lumiblock
#parentcustodial # the parent does not have a valid subscription yet
#custodial # has had the transfer made, is waiting for a valid custodial subscription to appear
#filemismatch # there is a dbs/phedex mismatch
#duplicates #a lumi section is there twice
## manual is not added yet, and should be so by recoveror
print wfo.name,"was tagged with :",list(assistance_tags)
if 'recovering' in assistance_tags:
## if active ACDC, being under threshold, filemismatch do not matter
assistance_tags = assistance_tags - set(['recovery','filemismatch'])
if 'recovery' in assistance_tags and 'recovered' in assistance_tags:
## should not set -recovery to anything that had ACDC already
assistance_tags = assistance_tags - set(['recovery','recovered'])
## straight to manual
assistance_tags.add('manual')
in_manual += 1
if 'recovery' in assistance_tags and 'manual' in assistance_tags:
## this is likely because something bad is happening, so leave it to manual
assistance_tags = assistance_tags - set(['recovery'])
assistance_tags.add('manual')
in_manual += 1
## that means there is something that needs to be done acdc, lumi invalidation, custodial, name it
print wfo.name,"needs assistance with",",".join( assistance_tags )
print wfo.name,"existing conditions",",".join( existing_assistance_tags )
#########################################
##### notification to requester #########
go_notify=False
if assistance_tags and not 'manual' in existing_assistance_tags and existing_assistance_tags != assistance_tags:
go_notify=True
if go_notify:
#if wfo.name in already_notified:
# print "double notification"
# sendEmail('double notification','please take a look at %s'%(wfo.name))
#else:
# already_notified.append( wfo.name )
###detailslink = 'https://cmsweb.cern.ch/reqmgr/view/details/%s'
#detailslink = 'https://cmsweb.cern.ch/reqmgr2/fetch?rid=%s'%(wfo.name)
###perflink = 'https://cmsweb.cern.ch/couchdb/workloadsummary/_design/WorkloadSummary/_show/histogramByWorkflow/%s'%(wfo.name)
perflink = '%s/report/%s'%(unified_url,wfo.name)
splitlink = 'https://cmsweb.cern.ch/reqmgr/view/splitting/%s'%(wfo.name)
## notify templates
messages= {
'recovery': 'Samples completed with missing statistics:\n%s\n%s '%( '\n'.join(['%.2f %% complete for %s'%(percent_completions[output]*100, output) for output in wfi.request['OutputDatasets'] ] ), perflink ),
'biglumi': 'Samples completed with large luminosity blocks:\n%s\n%s '%('\n'.join(['%d > %d for %s'%(events_per_lumi[output], lumi_upper_limit[output], output) for output in wfi.request['OutputDatasets'] if (events_per_lumi[output] > lumi_upper_limit[output])]), splitlink),
'duplicates': 'Samples completed with duplicated luminosity blocks:\n%s\n'%( '\n'.join(['%s'%output for output in wfi.request['OutputDatasets'] if output in duplications and duplications[output] ] ) ),
'filemismatch': 'Samples completed with inconsistency in DBS/Phedex',
#'manual' : 'Workflow completed and requires manual checks by Ops',
}
content = "The request PREPID (WORKFLOW) is facing issue in production.\n"
motive = False
for case in messages:
if case in assistance_tags:
content+= "\n"+messages[case]+"\n"
motive = True
content += "You are invited to check, while this is being taken care of by Comp-Ops.\n"
content += "This is an automated message from Comp-Ops.\n"
items_notified = set()
if use_mcm and motive:
wfi.notifyRequestor( content , mcm = mcm)
#########################################
## logic to set the status further
if assistance_tags:
new_status = 'assistance-'+'-'.join(sorted(assistance_tags) )
else:
new_status = 'assistance'
## case where the workflow was in manual from recoveror
if not 'manual' in wfo.status or new_status!='assistance-recovery':
wfo.status = new_status
if not options.test:
wfi.sendLog('checkor','setting %s to %s'%(wfo.name, wfo.status))
session.commit()
else:
print "current status is",wfo.status,"not changing to anything"
#open('already_notifified.json','w').write( json.dumps( already_notified , indent=2))
fDB.html()
if not spec and in_manual!=0:
sendEmail("fresh assistance status available","Fresh status are available at %s/assistance.html"%unified_url,destination=['*****@*****.**'])
#it's a bit annoying
pass
## custodial requests
print "Custodials"
print json.dumps(custodials, indent=2)
for site in custodials:
items_at = defaultdict(set)
for i in custodials[site]:
item, group = i.split('@') if '@' in i else (i,'DataOps')
items_at[group].add( item )
for group,items in items_at.items():
print ','.join(items),'=>',site,'@',group
if not options.test:
result = makeReplicaRequest(url, site, sorted(items) ,"custodial copy at production close-out",custodial='y',priority='low', approve = (site in SI.sites_auto_approve) , group=group)
print result
print "File Invalidation"
print invalidations
def checkor(url, spec=None, options=None):
fDB = closeoutInfo()
use_mcm = True
up = componentInfo(mcm=use_mcm, soft=['mcm'])
if not up.check(): return
use_mcm = up.status['mcm']
wfs=[]
if options.fetch:
## get all in running and check
wfs.extend( session.query(Workflow).filter(Workflow.status == 'away').all() )
wfs.extend( session.query(Workflow).filter(Workflow.status== 'assistance').all() )
if options.nofetch:
## than get all in need for assistance
wfs.extend( session.query(Workflow).filter(Workflow.status.startswith('assistance-')).all() )
custodials = defaultdict(list) #sites : dataset list
transfers = defaultdict(list) #sites : dataset list
invalidations = [] #a list of files
SI = siteInfo()
CI = campaignInfo()
mcm = McMClient(dev=False)
def get_campaign(output, wfi):
campaign = None
try:
campaign = output.split('/')[2].split('-')[0]
except:
if 'Campaign' in wfi.request:
campaign = wfi.request['Campaign']
return campaign
by_passes = []
holdings = []
for bypassor,email in [('jbadillo','*****@*****.**'),('vlimant','*****@*****.**'),('jen_a','*****@*****.**')]:
bypass_file = '/afs/cern.ch/user/%s/%s/public/ops/bypass.json'%(bypassor[0],bypassor)
if not os.path.isfile(bypass_file):
print "no file",bypass_file
continue
try:
by_passes.extend( json.loads(open(bypass_file).read()))
except:
print "cannot get by-passes from",bypass_file,"for",bypassor
sendEmail("malformated by-pass information","%s is not json readable"%(bypass_file), destination=[email])
holding_file = '/afs/cern.ch/user/%s/%s/public/ops/onhold.json'%(bypassor[0],bypassor)
if not os.path.isfile(holding_file):
print "no file",holding_file
continue
try:
holdings.extend( json.loads(open(holding_file).read()))
except:
print "cannot get holdings from",holding_file,"for",bypassor
sendEmail("malformated by-pass information","%s is not json readable"%(holding_file), destination=[email])
total_running_time = 5.*60.
sleep_time = max(0.5, total_running_time / len(wfs))
for wfo in wfs:
if spec and not (spec in wfo.name): continue
time.sleep( sleep_time )
print "checking on",wfo.name
## get info
wfi = workflowInfo(url, wfo.name)
## make sure the wm status is up to date.
# and send things back/forward if necessary.
wfo.wm_status = wfi.request['RequestStatus']
if wfo.wm_status == 'closed-out':
## manually closed-out
print wfo.name,"is already",wfo.wm_status
wfo.status = 'close'
session.commit()
continue
elif wfo.wm_status in ['failed','aborted','aborted-archived','rejected','rejected-archived','aborted-completed']:
## went into trouble
wfo.status = 'trouble'
print wfo.name,"is in trouble",wfo.wm_status
session.commit()
continue
elif wfo.wm_status in ['assigned','acquired']:
## not worth checking yet
print wfo.name,"not running yet"
session.commit()
continue
if '-onhold' in wfo.status:
if wfo.name in holdings:
print wfo.name,"on hold"
continue
if wfo.name in holdings:
wfo.status = 'assistance-onhold'
print "setting",wfo.name,"on hold"
session.commit()
continue
if wfo.wm_status != 'completed':
## for sure move on with closeout check if in completed
print "no need to check on",wfo.name,"in status",wfo.wm_status
session.commit()
continue
session.commit()
sub_assistance="" # if that string is filled, there will be need for manual assistance
is_closing = True
## do the closed-out checks one by one
## get it from somewhere
by_pass_checks = False
if wfo.name in by_passes:
print "we can bypass checks on",wfo.name
by_pass_checks = True
for bypass in by_passes:
if bypass in wfo.name:
print "we can bypass",wfo.name,"because of keyword",bypass
by_pass_checks = True
break
# tuck out DQMIO/DQM
wfi.request['OutputDatasets'] = [ out for out in wfi.request['OutputDatasets'] if not '/DQM' in out]
## anything running on acdc
familly = getWorkflowById(url, wfi.request['PrepID'], details=True)
acdc = []
acdc_inactive = []
has_recovery_going=False
had_any_recovery = False
for member in familly:
if member['RequestType'] != 'Resubmission': continue
if member['RequestName'] == wfo.name: continue
if member['RequestDate'] < wfi.request['RequestDate']: continue
if member['RequestStatus'] in ['running-open','running-closed','assignment-approved','assigned','acquired']:
print wfo.name,"still has an ACDC running",member['RequestName']
acdc.append( member['RequestName'] )
#print json.dumps(member,indent=2)
## hook for just waiting ...
is_closing = False
has_recovery_going=True
elif member['RequestStatus']==None:
print member['RequestName'],"is not real"
pass
else:
acdc_inactive.append( member['RequestName'] )
had_any_recovery = True
## completion check
percent_completions = {}
# print "let's see who is crashing", wfo.name
# print wfi.request['TotalInputEvents'],wfi.request['TotalInputLumis']
if not 'TotalInputEvents' in wfi.request:
event_expected,lumi_expected = 0,0
if not 'recovery' in wfo.status:
sendEmail("missing member of the request","TotalInputEvents is missing from the workload of %s"% wfo.name, destination=['*****@*****.**'])
else:
event_expected,lumi_expected = wfi.request['TotalInputEvents'],wfi.request['TotalInputLumis']
fractions_pass = {}
for output in wfi.request['OutputDatasets']:
event_count,lumi_count = getDatasetEventsAndLumis(dataset=output)
percent_completions[output] = 0.
if lumi_expected:
percent_completions[output] = lumi_count / float( lumi_expected )
fractions_pass[output] = 0.95
c = get_campaign(output, wfi)
if c in CI.campaigns and 'fractionpass' in CI.campaigns[c]:
fractions_pass[output] = CI.campaigns[c]['fractionpass']
print "overriding fraction to",fractions_pass[output],"for",output
if options.fractionpass:
fractions_pass[output] = options.fractionpass
print "overriding fraction to",fractions_pass[output],"by command line for",output
if not all([percent_completions[out] >= fractions_pass[out] for out in fractions_pass]):
print wfo.name,"is not completed"
print json.dumps(percent_completions, indent=2)
print json.dumps(fractions_pass, indent=2)
## hook for creating automatically ACDC ?
if has_recovery_going:
sub_assistance+='-recovering'
elif had_any_recovery:
## we want to have this looked at
sub_assistance+='-manual'
else:
sub_assistance+='-recovery'
is_closing = False
## correct lumi < 300 event per lumi
events_per_lumi = {}
for output in wfi.request['OutputDatasets']:
events_per_lumi[output] = getDatasetEventsPerLumi( output )
lumi_upper_limit = {}
for output in wfi.request['OutputDatasets']:
upper_limit = 301.
campaign = get_campaign(output, wfi)
#if 'EventsPerLumi' in wfi.request and 'FilterEfficiency' in wfi.request:
# upper_limit = 1.5*wfi.request['EventsPerLumi']*wfi.request['FilterEfficiency']
# print "setting the upper limit of lumisize to",upper_limit,"by request configuration"
if campaign in CI.campaigns and 'lumisize' in CI.campaigns[campaign]:
upper_limit = CI.campaigns[campaign]['lumisize']
print "overriding the upper lumi size to",upper_limit,"for",campaign
if options.lumisize:
upper_limit = options.lumisize
print "overriding the upper lumi size to",upper_limit,"by command line"
lumi_upper_limit[output] = upper_limit
if any([ events_per_lumi[out] >= lumi_upper_limit[out] for out in events_per_lumi]):
print wfo.name,"has big lumisections"
print json.dumps(events_per_lumi, indent=2)
## hook for rejecting the request ?
sub_assistance+='-biglumi'
is_closing = False
any_presence = {}
for output in wfi.request['OutputDatasets']:
any_presence[output] = getDatasetPresence(url, output, vetoes=[])
## custodial copy
custodial_locations = {}
custodial_presences = {}
for output in wfi.request['OutputDatasets']:
custodial_presences[output] = [s for s in any_presence[output] if 'MSS' in s]
custodial_locations[output] = phedexClient.getCustodialSubscriptionRequestSite(output)
if not custodial_locations[output]:
custodial_locations[output] = []
## presence in phedex
phedex_presence ={}
for output in wfi.request['OutputDatasets']:
phedex_presence[output] = phedexClient.getFileCountDataset(url, output )
vetoed_custodial_tier = ['MINIAODSIM']
out_worth_checking = [out for out in custodial_locations.keys() if out.split('/')[-1] not in vetoed_custodial_tier]
if not all(map( lambda sites : len(sites)!=0, [custodial_locations[out] for out in out_worth_checking])):
print wfo.name,"has not all custodial location"
print json.dumps(custodial_locations, indent=2)
##########
## hook for making a custodial replica ?
custodial = None
## get from other outputs
for output in out_worth_checking:
if len(custodial_locations[output]):
custodial = custodial_locations[output][0]
## try to get it from campaign configuration
if not custodial:
for output in out_worth_checking:
campaign = get_campaign(output, wfi)
if campaign in CI.campaigns and 'custodial' in CI.campaigns[campaign]:
custodial = CI.campaigns[campaign]['custodial']
print "Setting custodial to",custodial,"from campaign configuration"
break
## get from the parent
pick_custodial = True
if not custodial and 'InputDataset' in wfi.request:
## this is terribly dangerous to assume only
parents_custodial = phedexClient.getCustodialSubscriptionRequestSite( wfi.request['InputDataset'])
###parents_custodial = findCustodialLocation(url, wfi.request['InputDataset'])
if not parents_custodial:
parents_custodial = []
if len(parents_custodial):
custodial = parents_custodial[0]
else:
print "the input dataset",wfi.request['InputDataset'],"does not have custodial in the first place. abort"
sendEmail( "dataset has no custodial location", "Please take a look at %s in the logs of checkor"%wfi.request['InputDataset'])
is_closing = False
pick_custodial = False
if not custodial and pick_custodial:
## pick one at random
custodial = SI.pick_SE()
if custodial and ((not sub_assistance and not acdc) or by_pass_checks):
## register the custodial request, if there are no other big issues
for output in out_worth_checking:
if not len(custodial_locations[output]):
if phedex_presence[output]>=1:
custodials[custodial].append( output )
else:
print "no file in phedex for",output," not good to add to custodial requests"
else:
print "cannot find a custodial for",wfo.name
is_closing = False
## disk copy
disk_copies = {}
for output in wfi.request['OutputDatasets']:
disk_copies[output] = [s for s in any_presence[output] if (not 'MSS' in s) and (not 'Buffer' in s)]
if not all(map( lambda sites : len(sites)!=0, disk_copies.values())):
print wfo.name,"has not all output on disk"
print json.dumps(disk_copies, indent=2)
## presence in dbs
dbs_presence = {}
dbs_invalid = {}
for output in wfi.request['OutputDatasets']:
dbs_presence[output] = dbs3Client.getFileCountDataset( output )
dbs_invalid[output] = dbs3Client.getFileCountDataset( output, onlyInvalid=True)
fraction_invalid = 0.01
if not all([dbs_presence[out] == (dbs_invalid[out]+phedex_presence[out]) for out in wfi.request['OutputDatasets']]) and not options.ignorefiles:
print wfo.name,"has a dbs,phedex mismatch"
print json.dumps(dbs_presence, indent=2)
print json.dumps(dbs_invalid, indent=2)
print json.dumps(phedex_presence, indent=2)
## hook for just waiting ...
is_closing = False
if not all([(dbs_invalid[out] <= int(fraction_invalid*dbs_presence[out])) for out in wfi.request['OutputDatasets']]) and not options.ignorefiles:
print wfo.name,"has a dbs invalid file level too high"
print json.dumps(dbs_presence, indent=2)
print json.dumps(dbs_invalid, indent=2)
print json.dumps(phedex_presence, indent=2)
## need to be going and taking an eye
sub_assistance+="-invalidfiles"
is_closing = False
## put that heavy part at the end
## duplication check
duplications = {}
if is_closing:
print "starting duplicate checker for",wfo.name
for output in wfi.request['OutputDatasets']:
print "\tchecking",output
duplications[output] = True
try:
duplications[output] = dbs3Client.duplicateRunLumi( output )
except:
try:
duplications[output] = dbs3Client.duplicateRunLumi( output )
except:
print "was not possible to get the duplicate count for",output
is_closing=False
if any(duplications.values()) and not options.ignoreduplicates:
print wfo.name,"has duplicates"
print json.dumps(duplications,indent=2)
## hook for making file invalidation ?
sub_assistance+='-duplicates'
is_closing = False
## for visualization later on
if not wfo.name in fDB.record:
#print "adding",wfo.name,"to close out record"
fDB.record[wfo.name] = {
'datasets' :{},
'name' : wfo.name,
'closeOutWorkflow' : None,
}
fDB.record[wfo.name]['closeOutWorkflow'] = is_closing
for output in wfi.request['OutputDatasets']:
if not output in fDB.record[wfo.name]['datasets']: fDB.record[wfo.name]['datasets'][output] = {}
rec = fDB.record[wfo.name]['datasets'][output]
rec['percentage'] = float('%.2f'%(percent_completions[output]*100))
rec['duplicate'] = duplications[output] if output in duplications else 'N/A'
rec['phedexReqs'] = float('%.2f'%any_presence[output][custodial_presences[output][0]][1]) if len(custodial_presences[output])!=0 else 'N/A'
rec['closeOutDataset'] = is_closing
rec['transPerc'] = float('%.2f'%any_presence[output][ disk_copies[output][0]][1]) if len(disk_copies[output])!=0 else 'N/A'
rec['correctLumis'] = int(events_per_lumi[output]) if (events_per_lumi[output] > lumi_upper_limit[output]) else True
rec['missingSubs'] = False if len(custodial_locations[output])==0 else ','.join(list(set(custodial_locations[output])))
rec['dbsFiles'] = dbs_presence[output]
rec['dbsInvFiles'] = dbs_invalid[output]
rec['phedexFiles'] = phedex_presence[output]
rec['acdc'] = "%d / %d"%(len(acdc),len(acdc+acdc_inactive))
if by_pass_checks:
## force closing
is_closing = True
## and move on
if is_closing:
## toggle status to closed-out in request manager
print "setting",wfo.name,"closed-out"
if not options.test:
res = reqMgrClient.closeOutWorkflowCascade(url, wfo.name)
print "close out answer",res
if not res in ["None",None]:
print "retrying to closing out"
print res
res = reqMgrClient.closeOutWorkflowCascade(url, wfo.name)
if res in [None,"None"]:
wfo.status = 'close'
session.commit()
else:
print "could not close out",wfo.name,"will try again next time"
else:
## that means there is something that needs to be done acdc, lumi invalidation, custodial, name it
new_status = 'assistance'+sub_assistance
print wfo.name,"needs assistance with",new_status
if sub_assistance and wfo.status != new_status and 'PrepID' in wfi.request and not 'manual' in wfo.status:
pid = wfi.getPrepIDs()[0].replace('task_','')
#pid = wfi.request['PrepID'].replace('task_','')
## notify
messages= {
'recovery' : 'Samples completed with missing lumi count:\n%s '%( '\n'.join(['%.2f %% complete for %s'%(percent_completions[output]*100, output) for output in wfi.request['OutputDatasets'] ] ) ),
'biglumi' : 'Samples completed with large luminosity blocks:\n%s '%('\n'.join(['%d > %d for %s'%(events_per_lumi[output], lumi_upper_limit[output], output) for output in wfi.request['OutputDatasets'] if (events_per_lumi[output] > lumi_upper_limit[output])])),
'duplicate' : 'Samples completed with duplicated luminosity blocks:\n%s'%( '\n'.join(['%s'%output for output in wfi.request['OutputDatasets'] if output in duplications and duplications[output] ] ) ),
}
text ="The request %s (%s) is facing issue in production.\n" %( pid, wfo.name )
content = ""
for case in messages:
if case in new_status:
content+= "\n"+messages[case]+"\n"
text += content
text += "You are invited to check, while this is being taken care of by Ops.\n"
text += "This is an automated message."
if use_mcm and content:
print "Sending notification back to requestor"
print text
batches = mcm.getA('batches',query='contains=%s&status=announced'%pid)
if len(batches):
## go notify the batch
bid = batches[-1]['prepid']
print "batch nofication to",bid
mcm.put('/restapi/batches/notify', { "notes" : text, "prepid" : bid})
## go notify the request
print "request notification to",pid
mcm.put('/restapi/requests/notify',{ "message" : text, "prepids" : [pid] })
## case where the workflow was in manual from recoveror
if not 'manual' in wfo.status or new_status!='assistance-recovery':
wfo.status = new_status
if not options.test:
print "setting",wfo.name,"to",wfo.status
session.commit()
else:
print "current status is",wfo.status,"not changing to anything"
fDB.html()
## custodial requests
print "Custodials"
print json.dumps(custodials, indent=2)
for site in custodials:
print ','.join(custodials[site]),'=>',site
if not options.test:
result = makeReplicaRequest(url, site, list(set(custodials[site])),"custodial copy at production close-out",custodial='y',priority='low', approve = (site in SI.sites_auto_approve) )
print result
print "Transfers"
print json.dumps(transfers, indent=2)
## replicas requests
for site in transfers:
print ','.join(transfers[site]),'=>',site
if not options.test:
result = None
#result = makeReplicaRequest(url, site, list(set(transfers[site])),"copy to disk at production close-out")
print result
print "File Invalidation"
print invalidations
def transferor(url ,specific = None, talk=True, options=None):
if userLock(): return
if duplicateLock(): return
use_mcm = True
up = componentInfo(mcm=use_mcm, soft=['mcm'])
if not up.check(): return
use_mcm = up.status['mcm']
if options and options.test:
execute = False
else:
execute = True
SI = siteInfo()
CI = campaignInfo()
LI = lockInfo()
NLI = newLockInfo()
mcm = McMClient(dev=False)
dss = DSS()
print "counting all being handled..."
being_handled = len(session.query(Workflow).filter(Workflow.status == 'away').all())
being_handled += len(session.query(Workflow).filter(Workflow.status.startswith('stag')).all())
being_transfered = len(session.query(Workflow).filter(Workflow.status == 'staging').all())
being_handled += len(session.query(Workflow).filter(Workflow.status.startswith('assistance-')).all())
max_to_handle = options.maxworkflows
max_to_transfer = options.maxstaging
allowed_to_handle = max(0,max_to_handle - being_handled)
allowed_to_transfer = max(0,max_to_transfer - being_transfered)
wf_buffer = 5
if allowed_to_handle<=wf_buffer: ## buffer for having several wf per transfer
print "Not allowed to run more than",max_to_handle,"at a time. Currently",being_handled,"and",wf_buffer,"buffer"
else:
print being_handled,"already being handled",max_to_handle,"max allowed,",allowed_to_handle,"remaining","and",wf_buffer,"buffer"
if allowed_to_transfer <= wf_buffer:
print "Not allowed to transfer more than",max_to_transfer,"at a time. Currently",being_transfered,"and",wf_buffer,"buffer"
else:
print being_transfered,"already being transfered",max_to_transfer,"max allowed,",allowed_to_transfer,"remaining","and",wf_buffer,"buffer"
print "... done"
all_transfers=defaultdict(list)
needing_locks=defaultdict(list)
workflow_dependencies = defaultdict(set) ## list of wf.id per input dataset
wfs_and_wfh=[]
print "getting all wf to consider ..."
cache = getWorkflows(url, 'assignment-approved', details=True)
for wfo in session.query(Workflow).filter(Workflow.status=='considered').all():
print "\t",wfo.name
if specific and not specific in wfo.name: continue
cache_r =filter(lambda d:d['RequestName']==wfo.name, cache)
if len(cache_r):
wfs_and_wfh.append( (wfo, workflowInfo( url, wfo.name, spec=False, request = cache_r[0]) ) )
else:
wfs_and_wfh.append( (wfo, workflowInfo( url, wfo.name, spec=False) ) )
print "... done"
transfers_per_sites = defaultdict(int)
input_sizes = {}
input_cput = {}
input_st = {}
## list the size of those in transfer already
in_transfer_priority=0
min_transfer_priority=100000000
print "getting all wf in staging ..."
for wfo in session.query(Workflow).filter(Workflow.status=='staging').all():
wfh = workflowInfo( url, wfo.name, spec=False)
(lheinput,primary,parent,secondary) = wfh.getIO()
sites_allowed = getSiteWhiteList( (lheinput,primary,parent,secondary) )
for site in sites_allowed: ## we should get the actual transfer destination instead of the full white list
transfers_per_sites[site] += 1
#input_cput[wfo.name] = wfh.getComputingTime()
#input_st[wfo.name] = wfh.getSystemTime()
for prim in primary:
input_sizes[prim] = dss.get( prim )
print "\t",wfo.name,"needs",input_sizes[prim],"GB"
in_transfer_priority = max(in_transfer_priority, int(wfh.request['RequestPriority']))
min_transfer_priority = min(min_transfer_priority, int(wfh.request['RequestPriority']))
print "... done"
print "Max priority in transfer already",in_transfer_priority
print "Min priority in transfer already",min_transfer_priority
print "transfers per sites"
print json.dumps( transfers_per_sites, indent=2)
in_transfer_already = sum(input_sizes.values())
cput_in_transfer_already = sum(input_cput.values())
st_in_transfer_already = sum(input_st.values())
# shuffle first by name
random.shuffle( wfs_and_wfh )
#sort by priority higher first
wfs_and_wfh.sort(cmp = lambda i,j : cmp(int(i[1].request['RequestPriority']),int(j[1].request['RequestPriority']) ), reverse=True)
## list the size of all inputs
print "getting all input sizes ..."
for (wfo,wfh) in wfs_and_wfh:
(_,primary,_,_) = wfh.getIO()
#input_cput[wfo.name] = wfh.getComputingTime()
#input_st[wfo.name] = wfh.getSystemTime()
for prim in primary:
input_sizes[prim] = dss.get( prim )
print "... done"
cput_grand_total = sum(input_cput.values())
cput_to_transfer = cput_grand_total - cput_in_transfer_already
st_grand_total = sum(input_st.values())
st_to_transfer = st_grand_total - st_in_transfer_already
print "%15.4f [CPU h] worth already in transfer"%cput_in_transfer_already
print "%15.4f [CPU h] worth is the current requested transfer load"%cput_to_transfer
print "%15.4f [h] worth of absolute system time in transfer"%( cput_in_transfer_already / SI.availableSlots())
print "%15.4f [h] worth of absolute system time is the current requested transfer load"%( cput_to_transfer / SI.availableSlots())
print "%15.4f [h] worth of theoritical system time in transfer"%( st_in_transfer_already )
print "%15.4f [h] worth of theoritical system time is the current requested transfer load"%( st_to_transfer )
grand_total = sum(input_sizes.values())
to_transfer = grand_total - in_transfer_already
grand_transfer_limit = options.maxtransfer
transfer_limit = grand_transfer_limit - in_transfer_already
print "%15.4f GB already being transfered"%in_transfer_already
print "%15.4f GB is the current requested transfer load"%to_transfer
print "%15.4f GB is the global transfer limit"%grand_transfer_limit
print "%15.4f GB is the available limit"%transfer_limit
max_staging_per_site = options.maxstagingpersite
# the max priority value per dataset.
max_priority = defaultdict(int)
needs_transfer=0 ## so that we can count'em
passing_along = 0
transfer_sizes={}
went_over_budget=False
destination_cache = {}
for (wfo,wfh) in wfs_and_wfh:
print wfh.request['RequestPriority']
print wfo.name,"to be transfered"
#wfh = workflowInfo( url, wfo.name)
(_,primary,_,_) = wfh.getIO()
this_load=sum([input_sizes[prim] for prim in primary])
if ( this_load and (sum(transfer_sizes.values())+this_load > transfer_limit or went_over_budget ) ):
if went_over_budget:
print "Transfer has gone over bubget."
else:
print "Transfer will go over bubget."
print "%15.4f GB this load"%this_load
print "%15.4f GB already this round"%sum(transfer_sizes.values())
print "%15.4f GB is the available limit"%transfer_limit
went_over_budget=True
if int(wfh.request['RequestPriority']) >= in_transfer_priority and min_transfer_priority!=in_transfer_priority:
print "Higher priority sample",wfh.request['RequestPriority'],">=",in_transfer_priority,"go-on over budget"
else:
if not options.go:
print min_transfer_priority,"minimum priority",wfh.request['RequestPriority'],"<",in_transfer_priority,"stop"
continue
## throtlle by campaign go
if not CI.go( wfh.request['Campaign'] ):
print "No go for",wfh.request['Campaign']
if not options.go:
sendEmail("no go for managing","No go for "+wfh.request['Campaign'])
continue
## check if the batch is announced
def check_mcm(wfn):
announced=False
is_real=False
if not wfn.startswith('pdmvserv'):
is_real = True
try:
for b in mcm.getA('batches',query='contains=%s'% wfo.name):
is_real = True
if b['status']=='announced':
announced=True
break
except:
try:
for b in mcm.getA('batches',query='contains=%s'% wfo.name):
is_real = True
if b['status']=='announced':
announced=True
break
except:
print "could not get mcm batch announcement, assuming not real"
return announced,is_real
if not use_mcm:
announced,is_real = False,True
else:
announced,is_real = check_mcm( wfo.name )
if not announced:
print wfo.name,"does not look announced."# skipping?, rejecting?, reporting?"
if not is_real:
print wfo.name,"does not appear to be genuine."
## prevent any duplication. if the wf is not mentioned in any batch, regardless of status
continue
## check on a grace period
injection_time = time.mktime(time.strptime('.'.join(map(str,wfh.request['RequestDate'])),"%Y.%m.%d.%H.%M.%S")) / (60.*60.)
now = time.mktime(time.gmtime()) / (60.*60.)
if float(now - injection_time) < 4.:
if not options.go and not announced:
print "It is too soon to start transfer: %3.2fH remaining"%(now - injection_time)
continue
if passing_along >= allowed_to_handle:
#if int(wfh.request['RequestPriority']) >= in_transfer_priority and min_transfer_priority!=in_transfer_priority:
if int(wfh.request['RequestPriority']) >= in_transfer_priority and int(wfh.request['RequestPriority']) !=min_transfer_priority:
## higher priority, and not only this priority being transfered
print "Higher priority sample",wfh.request['RequestPriority'],">=",in_transfer_priority,"go-on over",max_to_handle
else:
print "Not allowed to pass more than",max_to_handle,"at a time. Currently",being_handled,"handled, and adding",passing_along
if not options.go: break
if this_load and needs_transfer >= allowed_to_transfer:
#if int(wfh.request['RequestPriority']) >= in_transfer_priority and min_transfer_priority!=in_transfer_priority:
if int(wfh.request['RequestPriority']) >= in_transfer_priority and int(wfh.request['RequestPriority']) !=min_transfer_priority:
## higher priority, and not only this priority being transfered
print "Higher priority sample",wfh.request['RequestPriority'],">=",in_transfer_priority,"go-on over",max_to_transfer
else:
print "Not allowed to transfer more than",max_to_transfer,"at a time. Currently",being_transfered,"transfering, and adding",needs_transfer
if not options.go: continue
(lheinput,primary,parent,secondary) = wfh.getIO()
for dataset in list(primary)+list(parent)+list(secondary):
## lock everything flat
NLI.lock( dataset )
if options and options.tosites:
sites_allowed = options.tosites.split(',')
else:
sites_allowed = getSiteWhiteList( (lheinput,primary,parent,secondary) )
if 'SiteWhitelist' in CI.parameters(wfh.request['Campaign']):
sites_allowed = CI.parameters(wfh.request['Campaign'])['SiteWhitelist']
if 'SiteBlacklist' in CI.parameters(wfh.request['Campaign']):
sites_allowed = list(set(sites_allowed) - set(CI.parameters(wfh.request['Campaign'])['SiteBlacklist']))
## reduce right away to sites in case of memory limitation
memory_allowed = SI.sitesByMemory( wfh.request['Memory'] )
if memory_allowed!=None:
print "sites allowing", wfh.request['Memory'],"are",memory_allowed
sites_allowed = list(set(sites_allowed) & set(memory_allowed))
if not sites_allowed:
print wfo.name,"has no possible sites to run at"
print "available for",wfh.request['Memory'],"are",memory_allowed
sendEmail("no possible sites","%s has no possible sites to run at"%( wfo.name ))
continue
blocks = []
if 'BlockWhitelist' in wfh.request and wfh.request['BlockWhitelist']:
blocks = wfh.request['BlockWhitelist']
if 'RunWhitelist' in wfh.request and wfh.request['RunWhitelist']:
## should make the block selection here
pass
if 'LumiList' in wfh.request and wfh.request['LumiList']:
## same, we could be doing the white list here too
pass
if blocks:
print "Reading",len(blocks),"in whitelist"
can_go = True
staging=False
allowed=True
if primary:
copies_needed_from_CPUh,CPUh = wfh.getNCopies()
if talk:
print wfo.name,'reads',', '.join(primary),'in primary'
## chope the primary dataset
for prim in primary:
max_priority[prim] = max(max_priority[prim],int(wfh.request['RequestPriority']))
sites_allowed = [site for site in sites_allowed if not any([osite.startswith(site) for osite in SI.sites_veto_transfer])]
print "Sites allowed minus the vetoed transfer"
print sorted(sites_allowed)
copies_needed_from_site = int(0.35*len(sites_allowed))+1 ## should just go for a fixed number based if the white list grows that big
print "Would make",copies_needed_from_site,"copies from site white list"
copies_needed = copies_needed_from_site
print "Would make",copies_needed_from_CPUh,"from cpu requirement",CPUh
copies_needed = copies_needed_from_CPUh
if options.maxcopy>0:
## stop maxing things out ??
#copies_needed = min(options.maxcopy,copies_needed)
#print "Maxed to",copies_needed
if copies_needed_from_CPUh > options.maxcopy:
sendEmail('An example of more than three copies','for %s it could have been beneficial to make %s copies'%( wfo.name, copies_needed_from_CPUh))
if 'Campaign' in wfh.request and wfh.request['Campaign'] in CI.campaigns and 'maxcopies' in CI.campaigns[wfh.request['Campaign']]:
copies_needed_from_campaign = CI.campaigns[wfh.request['Campaign']]['maxcopies']
copies_needed = min(copies_needed_from_campaign,copies_needed_from_site)
print "Maxed to",copies_needed,"by campaign configuration",wfh.request['Campaign']
## remove the sites that do not want transfers
workflow_dependencies[prim].add( wfo.id )
#####################################
###### JR 3/8/15 #### deprecating this
"""
presence = getDatasetPresence( url, prim , within_sites = [SI.CE_to_SE(site) for site in sites_allowed])
prim_location = [site for site,pres in presence.items() if pres[0]==True]
prim_parts = [site for site,pres in presence.items() if pres[0]==False]
if len(prim_location) >= copies_needed:
print "The output is all fully in place at",len(prim_location),"sites"
continue
# reduce the number of copies required by existing full copies
copies_needed = max(0,copies_needed - len(prim_location))
print "now need",copies_needed
subscriptions = listSubscriptions( url , prim , sites_allowed )
prim_destination = list(set([site for (site,(tid,decision)) in subscriptions.items() if decision and not any([site.endswith(veto) for veto in ['MSS','Export','Buffer']])]))
## remove the subscription where the dataset is in parts at
#prim_destination = list(set([site for (site,(tid,decision)) in subscriptions.items() if decision and not any([site.endswith(veto) for veto in ['MSS','Export','Buffer']]) and not site in prim_parts]))
## need to reject from that list the ones with a full copy already: i.e the transfer corresponds to the copy in place
prim_destination = [site for site in prim_destination if not site in prim_location]
## add transfer dependencies
latching_on_transfers = list(set([ tid for (site,(tid,decision)) in subscriptions.items() if decision and site in prim_destination and not any([site.endswith(veto) for veto in ['MSS','Export','Buffer']])]))
print latching_on_transfers
"""
###### JR 3/8/15 #### deprecating this
#####################################
### new ways of making the whole thing
destinations,all_block_names = getDatasetDestinations(url, prim, within_sites = [SI.CE_to_SE(site) for site in sites_allowed], only_blocks=blocks )
#destinations,all_block_names = getDatasetDestinations(url, prim, within_sites = [SI.CE_to_SE(site) for site in sites_allowed], only_blocks=blocks, group='DataOps')
#anaops_destinations,anaops_all_block_names = getDatasetDestinations(url, prim, within_sites = [SI.CE_to_SE(site) for site in sites_allowed], only_blocks=blocks, group='AnalysisOps' )
print json.dumps(destinations, indent=2)
## get where the dataset is in full and completed
prim_location = [site for (site,info) in destinations.items() if info['completion']==100 and info['data_fraction']==1]
## the rest is places it is going to be
prim_destination = [site for site in destinations.keys() if not site in prim_location]
## need to take out the transfer veto
prim_destination = [site for site in prim_destination if not any([osite.startswith(site) for osite in SI.sites_veto_transfer])]
for dsite in prim_destination:
needing_locks[dsite].append( prim )
if len(prim_location) >= copies_needed:
print "The output is all fully in place at",len(prim_location),"sites",prim_location
continue
copies_needed = max(0,copies_needed - len(prim_location))
print "now need",copies_needed
copies_being_made = [ sum([info['blocks'].keys().count(block) for site,info in destinations.items() if site in prim_destination]) for block in all_block_names]
latching_on_transfers = set()
[latching_on_transfers.update(info['blocks'].values()) for site,info in destinations.items() if site in prim_destination]
latching_on_transfers = list(latching_on_transfers)
#print latching_on_transfers
## figure out where all this is going to go
prim_to_distribute = [site for site in sites_allowed if not SI.CE_to_SE(site) in prim_location]
prim_to_distribute = [site for site in prim_to_distribute if not SI.CE_to_SE(site) in prim_destination]
## take out the ones that cannot receive transfers
prim_to_distribute = [site for site in prim_to_distribute if not any([osite.startswith(site) for osite in SI.sites_veto_transfer])]
if any([transfers_per_sites[site] < max_staging_per_site for site in prim_to_distribute]):
## means there is openings let me go
print "There are transfer slots available:",[(site,transfers_per_sites[site]) for site in prim_to_distribute]
for site in sites_allowed:
#increment accross the board, regardless of real destination: could be changed
transfers_per_sites[site] += 1
else:
if int(wfh.request['RequestPriority']) >= in_transfer_priority and min_transfer_priority!=in_transfer_priority:
print "Higher priority sample",wfh.request['RequestPriority'],">=",in_transfer_priority,"go-on over transfer slots available"
else:
print "Not allowed to transfer more than",max_staging_per_site," per site at a time. Going overboard for",[site for site in prim_to_distribute if transfers_per_sites[site]>=max_staging_per_site]
if not options.go:
allowed = False
break
for latching in latching_on_transfers:
tfo = session.query(Transfer).filter(Transfer.phedexid == latching).first()
if not tfo:
tfo = Transfer( phedexid = latching)
tfo.workflows_id = []
session.add(tfo)
if not wfo.id in tfo.workflows_id:
print "adding",wfo.id,"to",tfo.id,"with phedexid",latching
l = copy.deepcopy( tfo.workflows_id )
l.append( wfo.id )
tfo.workflows_id = l
if not options.test:
session.commit()
else:
session.flush() ## regardless of commit later on, we need to let the next wf feeding on this transfer to see it in query
can_go = False
transfer_sizes[prim] = input_sizes[prim]
staging = True
# reduce the number of copies required by the on-going full transfer : how do we bootstrap on waiting for them ??
#copies_needed = max(0,copies_needed - len(prim_destination))
copies_needed = max(0,copies_needed - min(copies_being_made))
print "then need",copies_needed
if copies_needed == 0:
print "The output is either fully in place or getting in full somewhere with",latching_on_transfers
can_go = True
continue
if len(prim_to_distribute)>0: ## maybe that a parameter we can play with to limit the
if not options or options.chop:
chops,sizes = getDatasetChops(prim, chop_threshold = options.chopsize, only_blocks=blocks)
spreading = distributeToSites( chops, prim_to_distribute, n_copies = copies_needed, weights=SI.cpu_pledges, sizes=sizes)
transfer_sizes[prim] = sum(sizes)
else:
spreading = {}
for site in prim_to_distribute:
if blocks:
spreading[site]=blocks
else:
spreading[site]=[prim]
transfer_sizes[prim] = input_sizes[prim] ## this is approximate if blocks are specified
can_go = False
print "selected CE destinations",spreading.keys()
for (site,items) in spreading.items():
all_transfers[site].extend( items )
if not allowed:
print "Not allowed to move on with",wfo.name
continue
if secondary:
if talk:
print wfo.name,'reads',', '.join(secondary),'in secondary'
for sec in secondary:
workflow_dependencies[sec].add( wfo.id )
if False:
## new style, failing on minbias
if not sec in destination_cache:
## this is barbbaric, and does not show the correct picture on workflow by workflow with different whitelist
destination_cache[sec],_ = getDatasetDestinations(url, sec, within_sites = [SI.CE_to_SE(site) for site in sites_allowed])
destinations = destination_cache[sec]
## truncate location/destination to those making up for >90% of the dataset
bad_destinations = [destinations.pop(site) for (site,info) in destinations.items() if info['data_fraction']<0.9]
sec_location = [site for (site,info) in destinations.items() if info['completion']>=95]
sec_destination = [site for site in destinations.keys() if not site in sec_location]
else:
## old style
presence = getDatasetPresence( url, sec )
sec_location = [site for site,pres in presence.items() if pres[1]>90.] ## more than 90% of the minbias at sites
subscriptions = listSubscriptions( url ,sec )
sec_destination = [site for site in subscriptions]
for site in sec_location:
needing_locks[site].append( sec )
for site in sec_destination:
needing_locks[site].append( sec )
sec_to_distribute = [site for site in sites_allowed if not any([osite.startswith(site) for osite in sec_location])]
sec_to_distribute = [site for site in sec_to_distribute if not any([osite.startswith(site) for osite in sec_destination])]
sec_to_distribute = [site for site in sec_to_distribute if not any([osite.startswith(site) for osite in SI.sites_veto_transfer])]
if len( sec_to_distribute )>0:
sec_size = dss.get( sec )
for site in sec_to_distribute:
site_se =SI.CE_to_SE(site)
if (SI.disk[site_se]*1024.) > sec_size:
all_transfers[site].append( sec )
can_go = False
else:
print "could not send the secondary input to",site_se,"because it is too big for the available disk",SI.disk[site_se]*1024,"GB need",sec_size
#sendEmail('secondary input too big','%s is too big (%s) for %s (%s)'%( sec, sec_size, site_se, SI.disk[site_se]*1024))
## is that possible to do something more
if can_go:
## no explicit transfer required this time
if staging:
## but using existing ones
print wfo.name,"latches on existing transfers, and nothing else"
wfo.status = 'staging'
needs_transfer+=1
else:
print wfo.name,"should just be assigned NOW to",sites_allowed
wfo.status = 'staged'
passing_along+=1
print "setting status to",wfo.status
session.commit()
continue
else:
## there is an explicit transfer required
if staging:
## and also using an existing one
print wfo.name,"latches on existing transfers"
if not options.test:
wfo.status = 'staging'
print "setting status to",wfo.status
session.commit()
print wfo.name,"needs a transfer"
needs_transfer+=1
passing_along+=1
print "accumulated locks of dataset in place"
print json.dumps(needing_locks, indent=2)
for site,items in needing_locks.items():
for item in items:
LI.lock( item, SI.CE_to_SE(site), 'usable input')
print "accumulated transfers"
print json.dumps(all_transfers, indent=2)
fake_id=-1
wf_id_in_prestaging=set()
for (site,items_to_transfer) in all_transfers.iteritems():
items_to_transfer = list(set(items_to_transfer))
## convert to storage element
site_se = SI.CE_to_SE(site)
## site that do not want input datasets
if site in SI.sites_veto_transfer:
print site,"does not want transfers"
continue
## throttle the transfer size to T2s ? we'd be screwed by a noPU sample properly configured.
## massage a bit the items
blocks = [it for it in items_to_transfer if '#' in it]
datasets = [it for it in items_to_transfer if not '#' in it]
if execute:
print "Making a replica to",site,"(CE)",site_se,"(SE) for"
else:
print "Would make a replica to",site,"(CE)",site_se,"(SE) for"
print "\t",len(blocks),"blocks"
## remove blocks if full dataset is send out
blocks = [block for block in blocks if not block.split('#')[0] in datasets]
print "\t",len(blocks),"needed blocks for",list(set([block.split('#')[0] for block in blocks]))
print "\t",len(datasets),"datasets"
print "\t",datasets
items_to_transfer = blocks + datasets
## operate the transfer
if options and options.stop:
## ask to move-on
answer = raw_input('Continue with that ?')
if not answer.lower() in ['y','yes','go']:
continue
if execute:
result = makeReplicaRequest(url, site_se, items_to_transfer, 'prestaging', priority='normal')
## make use of max_priority dataset:priority to set the subscriptions priority
"""
## does not function
once = True
for item in items_to_transfer:
bds = item.split('#')[0]
if max_priority[bds] >= 90000:
if once:
w=10
print "waiting",w,"s before raising priority"
time.sleep(w)
once=False
## raise it to high priority
print item,"subscription priority raised to high at",site_se
#print "This does not work yet properly it seems"
print updateSubscription(url, site_se, item, priority='high')
"""
#for item in list(set([it.split('#')[0] for it in items_to_transfer])):
for item in items_to_transfer:
LI.lock( item, site_se, 'pre-staging')
else:
#result= {'phedex':{'request_created' : [{'id' : fake_id}]}}
result= {'phedex':{'request_created' : []}}
fake_id-=1
if not result:
print "ERROR Could not make a replica request for",site,items_to_transfer,"pre-staging"
continue
for phedexid in [o['id'] for o in result['phedex']['request_created']]:
new_transfer = session.query(Transfer).filter(Transfer.phedexid == phedexid).first()
print phedexid,"transfer created"
if not new_transfer:
new_transfer = Transfer( phedexid = phedexid)
session.add( new_transfer )
new_transfer.workflows_id = set()
for transfering in list(set(map(lambda it : it.split('#')[0], items_to_transfer))):
new_transfer.workflows_id.update( workflow_dependencies[transfering] )
new_transfer.workflows_id = list(new_transfer.workflows_id)
wf_id_in_prestaging.update(new_transfer.workflows_id)
session.commit()
## auto approve it
if execute:
approved = approveSubscription(url, phedexid, [site_se])
for wfid in wf_id_in_prestaging:
tr_wf = session.query(Workflow).get(wfid)
if tr_wf and tr_wf.status!='staging':
if execute:
tr_wf.status = 'staging'
if talk:
print "setting",tr_wf.name,"to staging"
session.commit()
def equalizor(url , specific = None, options=None):
up = componentInfo(mcm=False, soft=['mcm'])
if not up.check(): return
if not specific:
workflows = getWorkflows(url, status='running-closed', details=True)
workflows.extend(getWorkflows(url, status='running-open', details=True))
## start from scratch
modifications = defaultdict(dict)
## define regionality site => fallback allowed. feed on an ssb metric ??
mapping = defaultdict(list)
reversed_mapping = defaultdict(list)
regions = defaultdict(list)
SI = siteInfo()
CI = campaignInfo()
UC = unifiedConfiguration()
for site in SI.sites_ready:
region = site.split('_')[1]
if not region in ['US'
,'DE','IT','FR',
'ES',
'UK' ### latest addition
]: continue
regions[region] = [region]
def site_in_depletion(s):
return True
if s in SI.sites_pressure:
(m, r, pressure) = SI.sites_pressure[s]
if float(m) < float(r):
print s,m,r,"lacking pressure"
return True
else:
print s,m,r,"pressure"
pass
return False
for site in SI.sites_ready:
region = site.split('_')[1]
## fallback to the region, to site with on-going low pressure
mapping[site] = [fb for fb in SI.sites_ready if any([('_%s_'%(reg) in fb and fb!=site and site_in_depletion(fb))for reg in regions[region]]) ]
for site in SI.sites_ready:
if site.split('_')[1] == 'US': ## to all site in the US
## add NERSC
mapping[site].append('T3_US_NERSC')
## add OSG
mapping[site].append('T3_US_OSG')
pass
#mapping['T2_IT_Rome'].append('T3_US_OSG')
#mapping['T1_US_FNAL'].append('T3_US_NERSC')
use_T0 = ('T0_CH_CERN' in UC.get("site_for_overflow"))
if options.t0: use_T0 = True
#if options.augment : use_T0 = True
use_HLT = ('T2_CH_CERN_HLT' in UC.get("site_for_overflow"))
if options.hlt: use_HLT = True
#if options.augment : use_HLT=True
if use_HLT:
mapping['T2_CH_CERN'].append('T2_CH_CERN_HLT')
if use_T0:
mapping['T2_CH_CERN'].append('T0_CH_CERN')
mapping['T1_IT_CNAF'].append('T0_CH_CERN')
mapping['T1_FR_CCIN2P3'].append('T0_CH_CERN')
mapping['T1_DE_KIT'].append('T0_CH_CERN')
## temptatively
#mapping['T0_CH_CERN'].append( 'T2_CH_CERN' )
## all europ can read from CERN
for reg in ['IT','DE','UK','FR','BE','ES']:
mapping['T2_CH_CERN'].extend([fb for fb in SI.sites_ready if '_%s_'%reg in fb])
pass
## all europ T1 among each others
europ_t1 = [site for site in SI.sites_ready if site.startswith('T1') and any([reg in site for reg in ['IT','DE','UK','FR','ES']])]
print europ_t1
for one in europ_t1:
for two in europ_t1:
if one==two: continue
mapping[one].append(two)
pass
## fnal can read from cnaf ?
#mapping['T1_IT_CNAF'].append( 'T1_US_FNAL' )
mapping['T1_IT_CNAF'].extend( [site for site in SI.sites_ready if '_US_' in site] ) ## all US can read from CNAF
mapping['T1_IT_CNAF'].append( 'T2_CH_CERN' )
mapping['T1_DE_KIT'].append( 'T2_CH_CERN' )
mapping['T2_CH_CERN'].append( 'T1_IT_CNAF' )
mapping['T2_CH_CERN'].append( 'T1_US_FNAL' )
#mapping['T2_UK_London_IC'].append( 'T2_CH_CERN' )
#mapping['T1_UK_RAL'].append( 'T2_BE_IIHE' )
mapping['T2_UK_London_IC'].append( 'T2_BE_IIHE' )
mapping['T2_UK_London_IC'].append( 'T2_FR_CCIN2P3' )
for site in SI.sites_ready:
if '_US_' in site:
mapping[site].append('T2_CH_CERN')
## make them appear as OK to use
force_sites = []
## overflow CERN to underutilized T1s
upcoming = json.loads( open('%s/GQ.json'%monitor_dir).read())
for possible in SI.sites_T1s:
if not possible in upcoming:
mapping['T2_CH_CERN'].append(possible)
pass
## remove add-hoc sites from overflow mapping
prevent_sites = []#'T2_US_Purdue']
for prevent in prevent_sites:
if prevent in mapping: mapping.pop( prevent )
for src in mapping:
for prevent in prevent_sites:
if prevent in mapping[src]:
mapping[src].remove( prevent )
## create the reverse mapping for the condor module
for site,fallbacks in mapping.items():
for fb in fallbacks:
if not site in reversed_mapping[fb]:
reversed_mapping[fb].append(site)
## this is the fallback mapping
print "Direct mapping : site => overflow"
print json.dumps( mapping, indent=2)
print "Reverse mapping : dest <= from origin"
print json.dumps( reversed_mapping, indent=2)
altered_tasks = set()
def running_idle( wfi , task_name):
gmon = wfi.getGlideMon()
#print gmon
if not gmon: return (0,0)
if not task_name in gmon: return (0,0)
return (gmon[task_name]['Running'], gmon[task_name]['Idle'])
def needs_action( wfi, task, min_idled = 100):
task_name = task.pathName.split('/')[-1]
running, idled = running_idle( wfi, task_name)
go = True
if not idled and not running :
go = False
if idled < 100:
go = False
if (not running and idled) or (running and (idled / float(running) > needs_action.pressure)):
go = True
else:
go = False
return go, task_name, running, idled
needs_action.pressure = UC.get('overflow_pressure')
def getPerf( task , stats_to_go = 200):
task = task.split('/')[1]+'/'+task.split('/')[-1]
try:
u = 'http://cms-gwmsmon.cern.ch/prodview/json/history/memoryusage720/%s'%task
print u
perf_data = json.loads(os.popen('curl -s --retry 5 %s'%u).read())
except Exception as e:
print str(e)
return (None,None)
buckets = filter(lambda i:i['key']!=0,perf_data['aggregations']["2"]['buckets'])
buckets.sort( key = lambda i:i['key'])
s=0
for bucket in buckets:
s+= bucket['doc_count']
bucket['cum'] = s
s_m = sum( bucket['key']*bucket['doc_count'] for bucket in buckets)
w_m = sum( bucket['doc_count'] for bucket in buckets)
m_m = max( bucket['key'] for bucket in buckets) if buckets else None
#90% percentile calculation
percentile_m = int(0.90 * w_m)
p_m = 0
s=0
for bucket in buckets:
p_m = bucket['key']
if bucket['cum'] > percentile_m:
break
p_m *= 1.1
print "percentile mem",p_m
max_count_m = None
max_count = 0
for bucket in buckets:
if bucket['doc_count'] > max_count:
max_count_m = bucket['key']
max_count = bucket['doc_count']
if max_count_m:
max_count_m *= 1.1
print "max count mem",max_count_m
b_m = None
if w_m > stats_to_go:
if p_m:
b_m = int(p_m)
else:
b_m = int(m_m) ## this is very bad if there are just a couple of outliers
b_m = int((s_m / float(w_m)) * 1.2)
else:
print "not enough stats for memory",w_m
try:
perf_data = json.loads(os.popen('curl -s --retry 5 http://cms-gwmsmon.cern.ch/prodview/json/history/runtime720/%s'%task).read())
except Exception as e:
print str(e)
return (b_m,None)
buckets = filter(lambda i:i['key']!=0,perf_data['aggregations']["2"]['buckets'])
buckets.sort( key = lambda i:i['key'])
s=0
for bucket in buckets:
s+= bucket['doc_count']
bucket['cum'] = s
s_t = sum( bucket['key']*bucket['doc_count'] for bucket in buckets)
w_t = sum( bucket['doc_count'] for bucket in buckets)
m_t = max( bucket['key'] for bucket in buckets) if buckets else None
percentile_t = int(0.90 * w_t)
p_t = 0
for bucket in buckets:
p_t = bucket['key']
if bucket['cum'] > percentile_t:
break
p_t *= 1.1
print "percentile time",p_t
max_count_t = None
max_count = 0
for bucket in buckets:
if bucket['doc_count'] > max_count:
max_count_t = bucket['key']
max_count = bucket['doc_count']
if max_count_t:
max_count_t *= 1.1
print "max count time",max_count_t
b_t = None
if w_t > stats_to_go:
b_t = m_t
else:
print "not enough stats for time",w_t
return (b_m,b_t)
def getcampaign( task ):
try:
taskname = task.pathName.split('/')[-1]
if hasattr( task, 'prepID'):
return task.prepID.split('-')[1]
elif taskname.count('-')>=1:
return taskname.split('-')[1]
else:
return None
except Exception as e :
print "Inconsistent prepid very likely"
print str(e)
return None
def close( interface ):
open('%s/equalizor.json.new'%monitor_pub_dir,'w').write( json.dumps( interface, indent=2))
os.system('mv %s/equalizor.json.new %s/equalizor.json'%(monitor_pub_dir,monitor_pub_dir))
os.system('cp %s/equalizor.json %s/logs/equalizor/equalizor.%s.json'%(monitor_pub_dir,monitor_dir,time.mktime(time.gmtime())))
interface = {
'mapping' : mapping,
'reversed_mapping' : reversed_mapping,
'modifications' : {},
'time' : {},
'memory' : {}
}
if options.augment or options.remove:
interface['modifications'] = json.loads( open('%s/equalizor.json'%monitor_pub_dir).read())['modifications']
interface['memory'] = json.loads( open('%s/equalizor.json'%monitor_pub_dir).read())['memory']
interface['time'] = json.loads( open('%s/equalizor.json'%monitor_pub_dir).read())['time']
if options.remove:
if specific in interface['modifications']:
print "poping",specific
interface['modifications'].pop(specific)
close( interface )
return
PU_locations = {}
PU_overflow = {}
PRIM_overflow = {}
PREMIX_overflow = {}
LHE_overflow = {}
tune_performance = []
pending_HLT = 0
max_HLT = 60000
pending_T0 = 0
max_T0 = 60000
try:
gmon = json.loads(os.popen('curl -s http://cms-gwmsmon.cern.ch/prodview/json/T2_CH_CERN_HLT').read())
pending_HLT += gmon["Running"]
pending_HLT += gmon["MatchingIdle"]
except:
pass
restricting_to_ready = ['pdmvserv_HIG-RunIISummer15wmLHEGS-00420_00157_v0__160909_001612_2018',
'pdmvserv_HIG-RunIISummer15wmLHEGS-00418_00157_v0__160909_001621_321',
'pdmvserv_HIG-RunIISummer15wmLHEGS-00419_00157_v0__160909_001621_2641'
]
remove_from = {
#'cerminar_Run2016B-v1-BTagCSV-23Sep2016_8020_160923_163224_2174' : ['T2_CH_CERN_HLT']
}
add_to = {
#'pdmvserv_EXO-RunIISpring16MiniAODv2-05060_00552_v0__161001_151813_7925' : ['T3_US_OSG'],
#'cerminar_Run2016C-v2-SingleElectron-23Sep2016_8020_160923_182146_3498' : ['T3_US_NERSC'],
#'cerminar_Run2016C-v2-Tau-23Sep2016_8020_160923_182336_5649' : ['T3_US_NERSC'],
}
stay_within_site_whitelist = False
specific_task=None
if specific and ":" in specific:
specific,specific_task = specific.split(':')
if specific:
wfs = session.query(Workflow).filter(Workflow.name.contains(specific)).all()
else:
wfs = session.query(Workflow).filter(Workflow.status == 'away').all()
performance = {}
resizing = {}
no_routing = [
]
random.shuffle( wfs )
for wfo in wfs:
if not wfo.status in ['away']: continue
if wfo.name in no_routing and not options.augment:
continue
if specific and not specific in wfo.name:
continue
if specific:
wfi = workflowInfo(url, wfo.name)
else:
cached = filter(lambda d : d['RequestName']==wfo.name, workflows)
if not cached : continue
wfi = workflowInfo(url, wfo.name, request = cached[0])
## only running-* should get re-routed, unless done by hand
if not wfi.request['RequestStatus'] in ['running-open','running-closed'] and not specific: continue
is_chain = (wfi.request['RequestType'] in ['TaskChain','StepChain'])
tasks_and_campaigns = []
for task in wfi.getWorkTasks():
tasks_and_campaigns.append( (task, getcampaign(task) ) )
lhe,prim,_,sec,sites_allowed = wfi.getSiteWhiteList()#getIO()
ncores = wfi.getMulticore()
memory_allowed = SI.sitesByMemory( float(wfi.request['Memory']) , maxCore=ncores)
if not lhe and not prim and not sec:
## no input at all: go for OSG!!!
add_to[wfo.name] = ['T3_US_OSG']
## check needs override
needs_overide = False
if not needs_overide and options.augment: needs_overide=True
def overide_from_agent( wfi, needs_overide):
bad_agents = []#'http://cmssrv219.fnal.gov:5984']
if not bad_agents: return needs_overide
if needs_overide: return True
agents = wfi.getAgents()
wqss = ['Running','Acquired']
if any([agent in agents.get(wqs,{}).keys() for wqs,agent in itertools.product( wqss, bad_agents)]):
print "overriding the need for bad agent"
needs_overide = True
return needs_overide
## now parse this for action
for i_task,(task,campaign) in enumerate(tasks_and_campaigns):
if options.augment:
print task.pathName
print campaign
resize = CI.get(campaign,'resize',{})
if resize and not is_chain:
resizing[task.pathName] = resize
tune = CI.get(campaign,'tune',options.tune)
if tune and not campaign in tune_performance:
tune_performance.append( campaign )
overflow = CI.get(campaign,'overflow',{})
if overflow:
if "PRIM" in overflow and not campaign in PRIM_overflow:
PRIM_overflow[campaign] = copy.deepcopy(overflow['PRIM'])
print "adding",campaign,"to PRIM overflow"
if "PREMIX" in overflow and not campaign in PREMIX_overflow:
PREMIX_overflow[campaign] = copy.deepcopy(overflow['PREMIX'])
print "adding",campaign,"to PREMIX overflow"
if "PU" in overflow and not campaign in PU_overflow:
PU_overflow[campaign] = copy.deepcopy(overflow['PU'])
print "adding",campaign,"to PU overflow rules"
if "LHE" in overflow and not campaign in LHE_overflow:
site_list = overflow['LHE'].get('site_list',"")
if site_list:
if type(site_list)==list:
LHE_overflow[campaign] = site_list
else:
print site_list
if hasattr(SI,site_list):
LHE_overflow[campaign] = copy.deepcopy( getattr(SI,site_list) )
else:
LHE_overflow[campaign] = site_list.split(',')
print "adding",campaign,"to light input overflow rules",LHE_overflow[campaign]
### setup the resizing
### get the task performance, for further massaging.
if campaign in tune_performance or options.tune:
print "performance",task.taskType,task.pathName
if task.taskType in ['Processing','Production']:
set_memory,set_time = getPerf( task.pathName )
#print "Performance %s GB %s min"%( set_memory,set_time)
wfi.sendLog('equalizor','Performance tuning to %s GB %s min for %s'%( set_memory,set_time,task.pathName.split('/')[-1] ))
## get values from gmwsmon
# massage the values : 95% percentile
performance[task.pathName] = {}
if set_memory:
performance[task.pathName]['memory']=min(set_memory,15000) ## max to 15GB
if set_time:
performance[task.pathName]['time'] = min(set_time, 1440) ## max to 24H
## rule to remove from the site whitelist site that do not look ready for unified (local banning)
if wfo.name in restricting_to_ready:
if task.taskType in ['Production']:
new_list = list(set(SI.sites_ready)&set(wfi.request['SiteWhitelist']))
modifications[wfo.name][task.pathName] = { "ReplaceSiteWhitelist" : new_list }
if campaign in PREMIX_overflow:
## figure out secondary location and neighbors
## figure out primary presence and neighbors
## do the intersection and add if in need.
needs, task_name, running, idled = needs_action(wfi, task)
#needs = True
## trick to be removed once all wf are passed through the agent patch
assigned_log = filter(lambda change : change["Status"] in ["assigned","acquired"],wfi.request['RequestTransition'])
if assigned_log:
then = assigned_log[0]['UpdateTime']
if then < 1479481842:
print "assigned too early"
needs = False
else:
print "assigned later enough"
else:
needs = False
if is_chain and task.pathName.endswith('_1'):
print i_task,"in chain prevents overflowing"
needs = False
if task.taskType in ['Processing','Production'] and needs:
secondary_locations = set(SI.sites_ready + force_sites)
for s in sec:
if not s in PU_locations:
presence = getDatasetPresence( url, s)
one_secondary_locations = [site for (site,(there,frac)) in presence.items() if frac>98.]
PU_locations[s] = one_secondary_locations
print "secondary is at",sorted(PU_locations[s])
secondary_locations = set([SI.SE_to_CE(site) for site in PU_locations[s]]) & secondary_locations
aaa_sec_grid = set(secondary_locations)
for site in sorted(aaa_sec_grid):
aaa_sec_grid.update( mapping.get(site, []) )
if len(prim):
dataset = list(prim)[0]
all_blocks,blocks = wfi.getActiveBlocks()
count_all = sum([len(v) for k,v in all_blocks.items()])
presence = getDatasetPresence(url, dataset, only_blocks=blocks )
aaa_prim_grid = set([SI.SE_to_CE(site) for site in presence.keys()])
for site in sorted(aaa_prim_grid):
aaa_prim_grid.update( mapping.get(site, []) )
print sorted(aaa_prim_grid),"around primary location",sorted(presence.keys())
print sorted(aaa_sec_grid),"aroudn secondary location",sorted(secondary_locations)
## intersect
aaa_grid = aaa_sec_grid & aaa_prim_grid
else:
print "premix overflow from a taskchain"
### hack hack hack
#modifications[wfo.name][task.pathName]= {"ReplaceSiteWhitelist" : ['T2_CH_CERN','T1_US_FNAL']}
aaa_grid = set(wfi.request['SiteWhitelist'])
banned_until_you_find_a_way_to_do_this = ['T3_US_OSG']
aaa_grid = filter(lambda s : not s in banned_until_you_find_a_way_to_do_this, aaa_grid)
if aaa_grid:
wfi.sendLog('equalizor','Extending site whitelist to %s'%sorted(aaa_grid))
modifications[wfo.name][task.pathName]= {"AddWhitelist" : sorted(aaa_grid)}
## rule to overflow jobs on the primary input
if campaign in PRIM_overflow:
if task.taskType in ['Processing','Production']:
if not wfi.request['TrustSitelists']:
###xrootd is OFF
dataset = list(prim)[0]
all_blocks,blocks = wfi.getActiveBlocks()
count_all = sum([len(v) for k,v in all_blocks.items()])
presence = getDatasetPresence(url, dataset, only_blocks=blocks )
in_full = [SI.SE_to_CE(site) for site,(there,_) in presence.items() if there]
aaa_grid= set()
aaa_grid_in_full = set(in_full)
for site in sorted(aaa_grid_in_full):
aaa_grid_in_full.update( mapping.get(site, []) )
## just add the neighbors to the existing whitelist. we could do more with block classAd
for site in wfi.request['SiteWhitelist']:
aaa_grid.update( mapping.get(site, []) )
aaa_grid = aaa_grid & set(sites_allowed + ['T3_US_NERSC']) ## and restrict to site that would be allowed at all (mcore, mem)
aaa_grid_in_full = aaa_grid_in_full & set(sites_allowed + ['T3_US_NERSC']) ## and restrict to site that would be allowed at all (mcore, mem)
gmon = wfi.getGlideMon()
needs, task_name, running, idled = needs_action(wfi, task)
print needs,running,idled
site_in_use = set(gmon[task_name]['Sites']) if gmon and task_name in gmon and 'Sites' in gmon[task_name] else set()
print dataset,"at",sorted(in_full),len(blocks),"/",count_all
print "running at",sorted(site_in_use)
print "set for",sorted(wfi.request['SiteWhitelist'])
print "around current whitelist" ,sorted(aaa_grid)
print "around where the data is now in full", sorted(aaa_grid_in_full)
if needs and not (site_in_use & set(in_full)) and aaa_grid_in_full:
print "we could be going for replace at that point"
wfi.sendLog('equalizor','Replaceing site whitelie to %s dynamically'% sorted(aaa_grid_in_full))
modifications[wfo.name][task.pathName] = { "ReplaceSiteWhitelist" : sorted( aaa_grid_in_full) }
else:
if aaa_grid:
print wfo.name
wfi.sendLog('equalizor','Adding in site white list %s dynamically'% sorted(aaa_grid) )
if wfo.name in modifications and task.pathName in modifications[wfo.name] and 'AddWhitelist' in modifications[wfo.name][task.pathName]:
modifications[wfo.name][task.pathName]["AddWhitelist"].extend(sorted(aaa_grid))
else:
modifications[wfo.name][task.pathName] = { "AddWhitelist" : sorted(aaa_grid) }
else:
## the request is already is in xrootd mode (either too generous, or just about right with neighbors of full data)
dataset = list(prim)[0]
all_blocks,blocks = wfi.getActiveBlocks()
count_all = sum([len(v) for k,v in all_blocks.items()])
fraction_left = float(len(blocks))/ count_all
#if fraction_left< 0.5: print '\n'.join( blocks )
presence = getDatasetPresence(url, dataset, only_blocks=blocks )
## in full is really the only place we can go to safely, since we have no job-data matching
in_full = [SI.SE_to_CE(site) for site,(there,_) in presence.items() if there]
gmon = wfi.getGlideMon()
needs, task_name, running, idled = needs_action(wfi, task)
site_in_use = set(gmon[task_name]['Sites']) if gmon and task_name in gmon and 'Sites' in gmon[task_name] else set()
print needs,running,idled
aaa_grid = set(in_full)
for site in list(aaa_grid):
aaa_grid.update( mapping.get(site, []) )
new_ones = set(in_full) - set(wfi.request['SiteWhitelist']) ## symptomatic of data have been repositionned
common = set(in_full) & set(wfi.request['SiteWhitelist'])
extra_shit = set(wfi.request['SiteWhitelist']) - aaa_grid ## symptomatic of too generous site-whitelist
aaa_grid = aaa_grid & set(sites_allowed+ ['T3_US_NERSC']) ## restrict to site that would be allowed at all (mcore, mem)
new_grid = aaa_grid - set(wfi.request['SiteWhitelist'])
print dataset,"is in full ",len(blocks),"/",count_all," at",in_full
print '\n'.join( sorted(blocks) )
print "running at",site_in_use
print "in common of the site whitelist",sorted(common)
print "site now also hosting the data",sorted(new_ones)
print "site in whitelist with no data",sorted(extra_shit)## with no data and not within aaa reach
if new_ones:
## we will be add sites
if needs and aaa_grid:
print wfo.name,"would replace for",sorted(aaa_grid)
print "but no thanks"
wfi.sendLog('equalizor','Changing the site whitelist to %s dynamically'%(sorted(aaa_grid)))
modifications[wfo.name][task.pathName] = { "ReplaceSiteWhitelist" : sorted(aaa_grid) }
elif new_grid:
print wfo.name,"would complement up to",sorted(aaa_grid)
wfi.sendLog('equalizor','Adding site white list to %s dynamically'% sorted(new_grid) )
modifications[wfo.name][task.pathName] = { "AddWhitelist" : sorted(new_grid) }
elif len(extra_shit)>5:
if aaa_grid:
print wfo.name,"would be restricting down to",sorted(aaa_grid),"because of",sorted(extra_shit)
wfi.sendLog('equalizor','Restricting the white list to %s dynamically'% sorted(aaa_grid) )
modifications[wfo.name][task.pathName] = { "ReplaceSiteWhitelist" : sorted(aaa_grid) }
else:
print wfo.name,"don't do anything"
if wfo.name in remove_from and task.taskType in ['Processing','Production']:
remove = remove_from[wfo.name]
restrict_to = set(wfi.request['SiteWhitelist'])
intersection= set(remove)&set(restrict_to)
if intersection:
print intersection,"is indeed in the original whitelist"
restrict_to = restrict_to - set(remove)
modifications[wfo.name][task.pathName] = { "ReplaceSiteWhitelist" : sorted(restrict_to) }
if wfo.name in add_to:
if task.taskType in ['Production','Processing']:
augment_to = add_to[wfo.name]
print "adding",sorted(augment_to),"to",wfo.name
if wfo.name in modifications and task.pathName in modifications[wfo.name] and 'AddWhitelist' in modifications[wfo.name][task.pathName]:
modifications[wfo.name][task.pathName]['AddWhitelist'].extend( augment_to )
else:
modifications[wfo.name][task.pathName] = { "AddWhitelist" : augment_to }
### rule to avoid the issue of taskchain secondary jobs being stuck at sites processing the initial step
if campaign in LHE_overflow:
#if not is_chain and task.taskType in ['Processing']:
if task.taskType in ['Processing']:
needs, task_name, running, idled = needs_action(wfi, task)
needs_overide = overide_from_agent( wfi, needs_overide)
extend_to = list(set(copy.deepcopy( LHE_overflow[campaign] )))
if stay_within_site_whitelist:
extend_to = list(set(extend_to) & set(wfi.request['SiteWhitelist'])) ## restrict to stupid-site-whitelist
extend_to = list(set(extend_to) & set(SI.sites_ready + force_sites))
if is_chain:
print "further restricting to initially allowed sites"
## restrict to initial allowed sites
extend_to = list(set(extend_to) & set(sites_allowed))
if not extend_to:
print "Nowhere to extend to"
continue
if extend_to and needs or needs_overide:
modifications[wfo.name][task.pathName] = { "ReplaceSiteWhitelist" : extend_to ,"Running" : running, "Pending" : idled, "Priority" : wfi.request['RequestPriority']}
wfi.sendLog('equalizor','%s of %s is running %d and pending %d, taking action : ReplaceSiteWhitelist \n %s'%( task_name,
wfo.name,
running,
idled ,
json.dumps( sorted(modifications[wfo.name][task.pathName]['ReplaceSiteWhitelist']))))
altered_tasks.add( task.pathName )
else:
wfi.sendLog('equalizor','%s of %s is running %d and pending %d'%( task_name, wfo.name, running, idled))
### overflow the 76 digi-reco to the site holding the pileup
if campaign in PU_overflow:
force = PU_overflow[campaign]['force'] if 'force' in PU_overflow[campaign] else False
secondary_locations = set(SI.sites_ready + force_sites)
for s in sec:
if not s in PU_locations:
presence = getDatasetPresence( url, s)
#one_secondary_locations = [site for (site,(there,frac)) in presence.items() if there]
one_secondary_locations = [site for (site,(there,frac)) in presence.items() if frac>98.]
PU_locations[s] = one_secondary_locations
print "secondary is at",sorted(PU_locations[s])
secondary_locations = set([SI.SE_to_CE(site) for site in PU_locations[s]]) & secondary_locations
## we should add all sites that hold the secondary input if any
### given that we have the secondary location available, it is not necessary to use the add-hoc list
##secondary_locations = list(set(PU_overflow[campaign]['sites']) & set( SI.sites_ready ))
## intersect with the sites that are allowed from the request requirement
secondary_locations = secondary_locations & set(memory_allowed)
if any([task.pathName.endswith(finish) for finish in ['_0','StepOneProc','Production']]) :
needs, task_name, running, idled = needs_action(wfi, task)
## removing the ones in the site whitelist already since they encode the primary input location
if stay_within_site_whitelist:
original_site_in_use = set(wfi.request['SiteWhitelist'] & set(secondary_locations))
else:
original_site_in_use = set(secondary_locations)
mode = 'AddWhitelist'
if not prim and i_task==0:
print "because there isn't any input, one should be able to just replace the sitewhitelist instead of adding, with the restriction of not reaching every possible sites"
mode='ReplaceSiteWhitelist'
## remove the sites that have already running jobs
gmon = wfi.getGlideMon()
if gmon and task_name in gmon and 'Sites' in gmon[task_name] and mode=='AddWhitelist':
site_in_use = set(gmon[task_name]['Sites'])
site_in_use = set([]) ## at this time I cannot find a reason to apply such limitation
print "removing",sorted(site_in_use)
## that determines where you want to run in addition
augment_by = list((set(secondary_locations)- site_in_use) & original_site_in_use)
else:
print "no existing running site"
augment_by = list(original_site_in_use)
if not augment_by: print "Nowhere to extend to"
needs_overide = overide_from_agent( wfi, needs_overide)
if augment_by and (needs or needs_overide or force) and PU_overflow[campaign]['pending'] < PU_overflow[campaign]['max']:
PU_overflow[campaign]['pending'] += idled
print "raising overflow to",PU_overflow[campaign]['pending'],"for",PU_overflow[campaign]['max']
## the step with an input ought to be the digi part : make this one go anywhere
modifications[wfo.name][task.pathName] = { mode : augment_by , "Running" : running, "Pending" : idled, "Priority" : wfi.request['RequestPriority']}
altered_tasks.add( task.pathName )
wfi.sendLog('equalizor','%s of %s is running %d and pending %d, taking action : %s \n %s'%( task_name, wfo.name,
running, idled,
mode,
json.dumps( sorted(augment_by), indent=2 )))
else:
print task_name,"of",wfo.name,"running",running,"and pending",idled
### overflow the skims back to multi-core
if campaign in ['Run2015D','Run2015C_25ns'] and task.taskType =='Skim':
original_swl = wfi.request['SiteWhitelist']
needs, task_name, running, idled = needs_action(wfi, task)
if (needs or needs_overide):
modifications[wfo.name][task.pathName] = { 'AddWhitelist' : original_swl,
"Running" : running, "Pending" : idled, "Priority" : wfi.request['RequestPriority']}
altered_tasks.add( task.pathName )
wfi.sendLog('equalizor','%s of %s is running %d and pending %d, taking action : AddWhitelist \n %s'%( task_name, wfo.name,
running, idled,
json.dumps( sorted(original_swl), indent=2 )))
if options.augment:
#print "uhm ....",sorted(wfi.request['SiteWhitelist']),i_task,use_HLT
pass
### this is a hack when we need to kick gensim out of everything
if campaign in [
#'RunIIWinter15GS',
#'RunIISummer15GS',
#'RunIISummer15wmLHEGS',
#'Summer12',
] and task.taskType in ['Production'] and is_chain:
#what are the site you want to take out. What are the jobs in whitelist, make the diff and replace
t1s = set([site for site in SI.all_sites if site.startswith('T1')])
ust2s = set([site for site in SI.all_sites if site.startswith('T2_US')])
#ust2s = set([site for site in SI.sites_mcore_ready if site.startswith('T2_US')])
allmcores = set(SI.sites_mcore_ready)
#set_for = set(wfi.request['SiteWhitelist']) - t1s
#set_for = set(wfi.request['SiteWhitelist']) - t1s - ust2s
#set_for = set(wfi.request['SiteWhitelist']) - allmcores
set_for = set(wfi.request['SiteWhitelist']) & t1s
print wfo.name,"going for",set_for
print task.pathName
if set_for:
modifications[wfo.name][task.pathName] = { "ReplaceSiteWhitelist" : sorted(set_for) }
### add the HLT at partner of CERN
if 'T2_CH_CERN' in wfi.request['SiteWhitelist'] and i_task in [0,1] and use_HLT and not wfi.request['TrustSitelists']:
needs, task_name, running, idled = needs_action(wfi, task)
if options.augment: needs=True
needs = True
##needs = random.random()<0.40 remove the random, just add up to a limit
if (needs or needs_overide) and pending_HLT < max_HLT:
pending_HLT += idled
if task.pathName in modifications[wfo.name] and 'AddWhitelist' in modifications[wfo.name][task.pathName]:
modifications[wfo.name][task.pathName]["AddWhitelist"].append( "T2_CH_CERN_HLT" )
wfi.sendLog('equalizor','also adding the HLT in whitelist of %s to %d for %d'%( task.pathName, pending_HLT, max_HLT))
## this Replace does not work at all for HLT
elif task.pathName in modifications[wfo.name] and 'ReplaceSiteWhitelist' in modifications[wfo.name][task.pathName]:
#modifications[wfo.name][task.pathName]["ReplaceSiteWhitelist"].append( "T2_CH_CERN_HLT" )
#print "\t",wfo.name,"adding replace HLT up to",pending_HLT,"for",max_HLT
print "already having a site replacement, not adding the HLT for now"
pass
else:
modifications[wfo.name][task.pathName] = { "AddWhitelist" : ["T2_CH_CERN_HLT"],
"Priority" : wfi.request['RequestPriority'],
"Running" : running,
"Pending" : idled}
wfi.sendLog('equalizor','adding the HLT in whitelist of %s to %d for %d'%( task.pathName, pending_HLT, max_HLT))
if i_task==0 and not sec and use_T0:
needs, task_name, running, idled = needs_action(wfi, task)
if options.augment: needs=True
#needs = True
good_type = wfi.request['RequestType'] in ['MonteCarlo','MonteCarloFromGEN']
read_lhe = ((not 'LheInputFiles' in wfi.request) or bool(wfi.request['LheInputFiles']))
good_type &= not read_lhe
if not good_type and not options.augment: needs = False
##needs = random.random()<0.40 remove the random, just add up to a limit
if (needs or needs_overide):
pending_T0 += idled
if task.pathName in modifications[wfo.name] and 'AddWhitelist' in modifications[wfo.name][task.pathName]:
if not "T0_CH_CERN" in modifications[wfo.name][task.pathName]["AddWhitelist"]:
modifications[wfo.name][task.pathName]["AddWhitelist"].append( "T0_CH_CERN" )
wfi,sendLog('equalizor','adding the T0 for %s to %d for %d'%( task.pathName, pending_T0, max_T0))
elif task.pathName in modifications[wfo.name] and 'ReplaceSiteWhitelist' in modifications[wfo.name][task.pathName]:
if not "T0_CH_CERN" in modifications[wfo.name][task.pathName]["ReplaceSiteWhitelist"]:
modifications[wfo.name][task.pathName]["ReplaceSiteWhitelist"].append( "T0_CH_CERN" )
wfi,sendLog('equalizor','adding the T0 to replacement for %s to %d for %d'%( task.pathName, pending_T0, max_T0))
else:
modifications[wfo.name][task.pathName] = { "AddWhitelist" : ["T0_CH_CERN"],
"Priority" : wfi.request['RequestPriority'],
"Running" : running,
"Pending" : idled}
wfi,sendLog('equalizor','adding the T0 for %s to %d for %d'%( task.pathName, pending_T0, max_T0))
interface['modifications'].update( modifications )
### manage the number of core and job resizing
#interface['cores']={'T2_CH_CERN_HLT': {'min':4,'max':16}, 'default': {'min':1, 'max':4}}
#interface['resizes'] = ['RunIISpring16DR80']
interface['resizing'] = resizing
### manage the modification of the memory and target time
max_N_mem = 2
max_N_time = 4
## discretize the memory to 10 at most values
mems = set([o['memory'] for t,o in performance.items() if 'memory' in o])
times = set([o['time'] for t,o in performance.items() if 'time' in o])
if len(mems)>max_N_mem:
mem_step = int((max(mems) - min(mems))/ float(max_N_mem))
print "rebinning memory"
for t in performance:
if not 'memory' in performance[t]: continue
(m,r) = divmod(performance[t]['memory'], mem_step)
performance[t]['memory'] = (m+1)*mem_step
if len(times)>max_N_time:
print "rebinning memory"
time_step = int((max(times) - min(times))/float(max_N_time))
for t in performance:
if not 'time' in performance[t]: continue
(m,r) = divmod(performance[t]['time'], time_step)
performance[t]['time'] = (m+1)*time_step
new_times = defaultdict(list)
new_memories = defaultdict(list)
for t,o in performance.items():
if 'time' in o:
new_times[str(o['time'])].append( t )
if 'memory' in o:
new_memories[str(o['memory'])].append( t )
interface['time'].update( new_times )
interface['memory'].update( new_memories )
## close and save
close( interface )
def stagor(url,specific =None, options=None):
if not componentInfo().check(): return
SI = siteInfo()
CI = campaignInfo()
UC = unifiedConfiguration()
done_by_wf_id = {}
done_by_input = {}
completion_by_input = {}
good_enough = 100.0
lost_blocks = json.loads(open('%s/lost_blocks_datasets.json'%monitor_dir).read())
lost_files = json.loads(open('%s/lost_files_datasets.json'%monitor_dir).read())
known_lost_blocks = {}
known_lost_files = {}
for dataset in set(lost_blocks.keys()+lost_files.keys()):
b,f = findLostBlocksFiles(url, dataset)
if dataset in lost_blocks and not b:
print dataset,"has no really lost blocks"
else:
known_lost_blocks[dataset] = [i['name'] for i in b]
if dataset in lost_files and not f:
print dataset,"has no really lost files"
else:
known_lost_files[dataset] = [i['name'] for i in f]
try:
cached_transfer_statuses = json.loads(open('cached_transfer_statuses.json').read())
except:
print "inexisting transfer statuses. starting fresh"
cached_transfer_statuses = {}
return False
transfer_statuses = {}
def time_point(label="",sub_lap=False):
now = time.mktime(time.gmtime())
nows = time.asctime(time.gmtime())
print "Time check (%s) point at : %s"%(label, nows)
print "Since start: %s [s]"% ( now - time_point.start)
if sub_lap:
print "Sub Lap : %s [s]"% ( now - time_point.sub_lap )
time_point.sub_lap = now
else:
print "Lap : %s [s]"% ( now - time_point.lap )
time_point.lap = now
time_point.sub_lap = now
time_point.sub_lap = time_point.lap = time_point.start = time.mktime(time.gmtime())
time_point("Check cached transfer")
## collect all datasets that are needed for wf in staging, correcting the status of those that are not really in staging
wfois = []
needs = defaultdict(list)
needs_by_priority = defaultdict(list)
for wfo in session.query(Workflow).filter(Workflow.status == 'staging').all():
wfi = workflowInfo(url, wfo.name)
if wfi.request['RequestStatus'] in ['running-open','running-closed','completed','assigned','acquired']:
wfi.sendLog('stagor', "is in status %s"%wfi.request['RequestStatus'])
wfi.status='away'
session.commit()
continue
if not wfi.request['RequestStatus'] in ['assignment-approved']:
## should be setting 'away' too
print wfo.name,"is",wfi.request['RequestStatus']
sendEmail("wrong status in staging. debug","%s is in %s, should set away."%(wfo.name,wfi.request['RequestStatus']))
wfois.append( (wfo,wfi) )
_,primaries,_,secondaries = wfi.getIO()
for dataset in list(primaries)+list(secondaries):
needs[wfo.name].append( dataset)
done_by_input[dataset] = {}
completion_by_input[dataset] = {}
needs_by_priority[wfi.request['RequestPriority']].append( dataset )
wfi.sendLog('stagor', '%s needs %s'%( wfo.name, dataset))
time_point("Check staging workflows")
open('%s/dataset_requirements.json'%monitor_dir,'w').write( json.dumps( needs, indent=2))
for prio in needs_by_priority: needs_by_priority[prio] = list(set(needs_by_priority[prio]))
open('%s/dataset_priorities.json'%monitor_dir,'w').write( json.dumps( needs_by_priority , indent=2))
dataset_endpoints = defaultdict(set)
endpoint_in_downtime = defaultdict(set)
#endpoint_completed = defaultdict(set)
endpoint_incompleted = defaultdict(set)
#endpoint = defaultdict(set)
send_back_to_considered = set()
## first check if anything is inactive
all_actives = set([transfer.phedexid for transfer in session.query(TransferImp).filter(TransferImp.active).all()])
for active_phedexid in all_actives:
skip = True
transfers_phedexid = session.query(TransferImp).filter(TransferImp.phedexid == active_phedexid).all()
for imp in transfers_phedexid:
if imp.workflow.status == 'staging':
skip =False
sendLog('stagor',"\t%s is staging for %s"%(imp.phedexid, imp.workflow.name))
if skip:
sendLog('stagor',"setting %s inactive" % active_phedexid)
for imp in transfers_phedexid:
imp.active = False
session.commit()
all_actives = sorted(set([transfer.phedexid for transfer in session.query(TransferImp).filter(TransferImp.active).all()]))
for phedexid in all_actives:
if specific: continue
## check on transfer completion
not_cached = False
if str(phedexid) in cached_transfer_statuses:
### use a cache for transfer that already looked done
sendLog('stagor',"read %s from cache"%phedexid)
checks = cached_transfer_statuses[str(phedexid)]
else:
## I actually would like to avoid that all I can
sendLog('stagor','Performing spurious transfer check on %s'% phedexid, level='critical')
checks = checkTransferStatus(url, phedexid, nocollapse=True)
if not checks:
## this is going to bias quite heavily the rest of the code. we should abort here
#sendLog('stagor','Ending stagor because of skewed input from checkTransferStatus', level='critical')
#return False
sendLog('stagor','Stagor has got a skewed input from checkTransferStatus', level='critical')
checks = {}
pass
#checks = {}
#not_cached = True
time_point("Check transfer status %s"% phedexid, sub_lap=True)
## just write this out
transfer_statuses[str(phedexid)] = copy.deepcopy(checks)
if not specific:
for dsname in checks:
if not dsname in done_by_input: done_by_input[dsname]={}
if not dsname in completion_by_input: completion_by_input[dsname] = {}
done_by_input[dsname][phedexid]=all(map(lambda i:i>=good_enough, checks[dsname].values()))
completion_by_input[dsname][phedexid]=checks[dsname].values()
if checks:
sendLog('stagor',"Checks for %s are %s"%( phedexid, [node.values() for node in checks.values()]))
done = all(map(lambda i:i>=good_enough,list(itertools.chain.from_iterable([node.values() for node in checks.values()]))))
else:
## it is empty, is that a sign that all is done and away ?
if not_cached:
print "Transfer status was not cached"
else:
print "ERROR with the scubscriptions API of ",phedexid
print "Most likely something else is overiding the transfer request. Need to work on finding the replacement automatically, if the replacement exists"
done = False
transfers_phedexid = session.query(TransferImp).filter(TransferImp.phedexid == phedexid).all()
for imp in transfers_phedexid:
tr_wf = imp.workflow
if tr_wf:# and tr_wf.status == 'staging':
if not tr_wf.id in done_by_wf_id: done_by_wf_id[tr_wf.id]={}
done_by_wf_id[tr_wf.id][phedexid]=done
if done:
imp.active = False
session.commit()
for ds in checks:
for s,v in checks[ds].items():
dataset_endpoints[ds].add( s )
if done:
sendLog('stagor',"%s is done"%phedexid)
cached_transfer_statuses[str(phedexid)] = copy.deepcopy(checks)
else:
sendLog('stagor',"%s is not finished %s"%(phedexid, pprint.pformat( checks )))
##pprint.pprint( checks )
## check if the destination is in down-time
for ds in checks:
sites_incomplete = [SI.SE_to_CE(s) for s,v in checks[ds].items() if v<good_enough]
sites_incomplete_down = [s for s in sites_incomplete if not s in SI.sites_ready]
## no space means no transfer should go there : NO, it does not work in the long run
#sites_incomplete_down = [SI.SE_to_CE(s) for s,v in checks[ds].items() if (v<good_enough and (SI.disk[s]==0 or (not SI.SE_to_CE(s) in SI.sites_ready)))]
if sites_incomplete_down:
sendLog('stagor',"%s are in downtime, while waiting for %s to get there"%( ",".join(sites_incomplete_down), ds))
#sites_complete = [SI.SE_to_CE(s) for s,v in checks[ds].items() if v>=good_enough]
#endpoint[ds].update( sites_complete )
#endpoint[ds].update( sites_incomplete )
#endpoint_completed[ds].update( sites_complete )
endpoint_incompleted[ds].update( sites_incomplete )
endpoint_in_downtime[ds].update( sites_incomplete_down )
time_point("Check on-going transfers")
print "End points"
for k in dataset_endpoints: dataset_endpoints[k] = list(dataset_endpoints[k])
print json.dumps( dataset_endpoints , indent=2)
print "End point in down time"
for k in endpoint_in_downtime: endpoint_in_downtime[k] = list(endpoint_in_downtime[k])
print json.dumps( endpoint_in_downtime , indent=2)
print "End point incomplete in down time"
for k in endpoint_incompleted: endpoint_incompleted[k] = list(endpoint_incompleted[k])
print json.dumps( endpoint_incompleted , indent=2)
#open('cached_transfer_statuses.json','w').write( json.dumps( cached_transfer_statuses, indent=2))
open('%s/transfer_statuses.json'%monitor_dir,'w').write( json.dumps( transfer_statuses, indent=2))
open('%s/dataset_endpoints.json'%monitor_dir,'w').write( json.dumps(dataset_endpoints, indent=2))
already_stuck = json.loads( open('%s/stuck_transfers.json'%monitor_pub_dir).read() ).keys()
already_stuck.extend( getAllStuckDataset() )
missing_in_action = defaultdict(list)
print "-"*10,"Checking on workflows in staging","-"*10
#forget_about = ['/MinBias_TuneCUETP8M1_13TeV-pythia8/RunIISummer15GS-MCRUN2_71_V1-v2/GEN-SIM']
#for what in forget_about:
# if not done_by_input[what]:
# done_by_input[what] = {'fake':True}
## come back to workflows and check if they can go
available_cache = defaultdict(lambda : defaultdict(float))
presence_cache = defaultdict(dict)
time_point("Preparing for more")
for wfo,wfi in wfois:
print "#"*30
time_point("Forward checking %s"% wfo.name,sub_lap=True)
## the site white list takes site, campaign, memory and core information
(_,primaries,_,secondaries,sites_allowed) = wfi.getSiteWhiteList(verbose=False)
se_allowed = [SI.CE_to_SE(site) for site in sites_allowed]
se_allowed.sort()
se_allowed_key = ','.join(se_allowed)
readys={}
for need in list(primaries)+list(secondaries):
if not need in done_by_input:
wfi.sendLog('stagor',"missing transfer report for %s"%need)
readys[need] = False
## should warn someone about this !!!
## it cannot happen, by construction
sendEmail('missing transfer report','%s does not have a transfer report'%(need))
continue
if not done_by_input[need] and need in list(secondaries):
wfi.sendLog('stagor',"assuming it is OK for secondary %s to have no attached transfers"% need)
readys[need] = True
done_by_input[need] = { "fake" : True }
continue
if len(done_by_input[need]) and all(done_by_input[need].values()):
wfi.sendLog('stagor',"%s is ready"%need)
print json.dumps( done_by_input[need] , indent=2)
readys[need] = True
else:
wfi.sendLog('stagor',"%s is not ready \n%s"%(need,json.dumps( done_by_input[need] , indent=2)))
readys[need] = False
if readys and all(readys.values()):
if wfo.status == 'staging':
wfi.sendLog('stagor',"all needs are fullfilled, setting staged")
wfo.status = 'staged'
session.commit()
else:
wfi.sendLog('stagor',"all needs are fullfilled, already")
print json.dumps( readys, indent=2 )
else:
wfi.sendLog('stagor',"missing requirements")
copies_needed,_ = wfi.getNCopies()
jump_ahead = False
re_transfer = False
## there is missing input let's do something more elaborated
for need in list(primaries):#+list(secondaries):
if endpoint_in_downtime[need] == endpoint_incompleted[need]:
#print need,"is going to an end point in downtime"
wfi.sendLog('stagor',"%s has only incomplete endpoint in downtime"%need)
re_transfer=True
if not se_allowed_key in available_cache[need]:
available_cache[need][se_allowed_key] = getDatasetBlocksFraction( url , need, sites=se_allowed )
if available_cache[need][se_allowed_key] >= copies_needed:
wfi.sendLog('stagor',"assuming it is OK to move on like this already for %s"%need)
jump_ahead = True
else:
wfi.sendLog('stagor',"Available %s times"% available_cache[need][se_allowed_key])
missing_and_downtime = list(set(endpoint_in_downtime[need]) & set(endpoint_incompleted[need]))
if missing_and_downtime:
wfi.sendLog('stagor',"%s is incomplete at %s which is in downtime, trying to move along"%(need, ','.join(missing_and_downtime)))
jump_ahead = True
else:
wfi.sendLog('stagor',"continue waiting for transfers for optimum production performance.")
## compute a time since staging to filter jump starting ?
# check whether the inputs is already in the stuck list ...
for need in list(primaries)+list(secondaries):
if need in already_stuck:
wfi.sendLog('stagor',"%s is stuck, so try to jump ahead"%need)
jump_ahead = True
if jump_ahead or re_transfer:
details_text = "checking on availability for %s to jump ahead"%wfo.name
details_text += '\n%s wants %s copies'%(wfo.name,copies_needed)
copies_needed = max(1,copies_needed-1)
details_text += '\nlowering by one unit to %s'%copies_needed
wfi.sendLog('stagor', details_text)
all_check = True
prim_where = set()
for need in list(primaries):
if not se_allowed_key in presence_cache[need]:
presence_cache[need][se_allowed_key] = getDatasetPresence( url, need , within_sites=se_allowed)
presence = presence_cache[need][se_allowed_key]
prim_where.update( presence.keys() )
available = available_cache[need][se_allowed_key]
this_check = (available >= copies_needed)
wfi.sendLog('stagor', "%s is available %s times (%s), at %s"%( need, available, this_check, se_allowed_key))
all_check &= this_check
if not all_check: break
for need in list(secondaries):
## I do not want to check on the secon
## this below does not function because the primary could be all available, and the secondary not complete at a certain site that does not matter at that point
this_check = all(done_by_input[need].values())
wfi.sendLog('stagor',"%s is this much transfered %s"%(need, json.dumps(done_by_input[need], indent=2)))
all_check&= this_check
#if not se_allowed_key in presence_cache[need]:
# presence_cache[need][se_allowed_key] = getDatasetPresence( url, need , within_sites=se_allowed)
## restrict to where the primary is
#presence = dict([(k,v) for (k,v) in presence_cache[need][se_allowed_key].items() if k in prim_where])
#this_check = all([there for (there,frac) in presence.values()])
#print need,"is present at all sites:",this_check
#all_check&= this_check
if all_check and not re_transfer:
wfi.sendLog('stagor',"needs are sufficiently fullfilled, setting staged")
wfo.status = 'staged'
session.commit()
else:
print wfo.name,"has to wait a bit more"
wfi.sendLog('stagor',"needs to wait a bit more")
else:
wfi.sendLog('stagor',"not checking availability")
if re_transfer:
wfi.sendLog('stagor',"Sending back to considered because of endpoint in downtime")
if wfo.status == 'staging':
wfo.status = 'considered'
session.commit()
send_back_to_considered.add( wfo.name )
time_point("Checked affected workflows")
if send_back_to_considered:
#sendEmail("transfer to endpoint in downtime","sending back to considered the following workflows \n%s"%('\n'.join( send_back_to_considered)))
sendLog('stagor', "sending back to considered the following workflows \n%s"%('\n'.join( send_back_to_considered)), level='critical')
print "-"*10,"Checking on non-available datasets","-"*10
## now check on those that are not fully available
for dsname in available_cache.keys():
## squash the se_allowed_key key
available_cache[dsname] = min( available_cache[dsname].values() )
really_stuck_dataset = set()
for dsname,available in available_cache.items():
using_its = getWorkflowByInput(url, dsname)
#print using_its
using_wfos = []
for using_it in using_its:
wf = session.query(Workflow).filter(Workflow.name == using_it).first()
if wf:
using_wfos.append( wf )
if not len(done_by_input[dsname]):
print "For dataset",dsname,"there are no transfer report. That's an issue."
for wf in using_wfos:
if wf.status == 'staging':
if UC.get("stagor_sends_back"):
print "sending",wf.name,"back to considered"
wf.status = 'considered'
session.commit()
#sendEmail( "send back to considered","%s was send back and might be trouble"% wf.name)
sendLog('stagor', "%s was send back and might be trouble"% wf.name, level='critical')
else:
print "would send",wf.name,"back to considered"
#sendEmail( "subscription lagging behind","susbscriptions to get %s running are not appearing in phedex. I would have send it back to considered but that's not good."% wf.name)
sendLog('stagor', "susbscriptions to get %s running are not appearing in phedex. I would have send it back to considered but that's not good."% wf.name, level='critical')
continue
## not compatible with checking on secondary availability
#if all([wf.status != 'staging' for wf in using_wfos]):
# ## means despite all checks that input is not needed
# continue
if available < 1.:
print "incomplete",dsname
## there is a problem in the method below that it does not account for files stuck in T1*Buffer only
lost_blocks,lost_files = findLostBlocksFiles( url, dsname ) if (not dsname.endswith('/RAW')) else ([],[])
lost_block_names = [item['name'] for item in lost_blocks]
lost_file_names = [item['name'] for item in lost_files]
if lost_blocks:
#print json.dumps( lost , indent=2 )
## estimate for how much !
fraction_loss,_,n_missing = getDatasetBlockFraction(dsname, lost_block_names)
print "We have lost",len(lost_block_names),"blocks",lost_block_names,"for %f%%"%(100.*fraction_loss)
if fraction_loss > 0.05: ## 95% completion mark
#sendEmail('we have lost too many blocks','%s is missing %d blocks, for %d events, %f %% loss'%(dsname, len(lost_block_names), n_missing, fraction_loss))
sendLog('stagor', '%s is missing %d blocks, for %d events, %3.2f %% loss'%(dsname, len(lost_block_names), n_missing, 100*fraction_loss), level='critical')
## the workflow should be rejected !
for wf in using_wfos:
if wf.status == 'staging':
print wf.name,"is doomed. setting to trouble"
wf.status = 'trouble'
session.commit()
sendLog('stagor', '%s has too much loss on the input dataset %s. Missing %d blocks, for %d events, %3.2f %% loss'%(wf.name, dsname, len(lost_block_names), n_missing, 100*fraction_loss), level='critical')
else:
## probably enough to make a ggus and remove
if not dsname in known_lost_blocks:
#sendEmail('we have lost a few blocks', '%s is missing %d blocks, for %d events, %f %% loss\n\n%s'%(dsname, len(lost_block_names), n_missing, fraction_loss, '\n'.join( lost_block_names ) ))
sendLog('stagor', '%s is missing %d blocks, for %d events, %f %% loss\n\n%s'%(dsname, len(lost_block_names), n_missing, fraction_loss, '\n'.join( lost_block_names ) ), level='critical')
known_lost_blocks[dsname] = [i['name'] for i in lost_blocks]
really_stuck_dataset.add( dsname )
if lost_files:
fraction_loss,_,n_missing = getDatasetFileFraction(dsname, lost_file_names)
print "We have lost",len(lost_file_names),"files",lost_file_names,"for %f%%"%fraction_loss
if fraction_loss > 0.05:
#sendEmail('we have lost too many files','%s is missing %d files, for %d events, %f %% loss'%(dsname, len(lost_file_names),n_missing, fraction_loss))
sendLog('stagor', '%s is missing %d files, for %d events, %f %% loss'%(dsname, len(lost_file_names),n_missing, fraction_loss), level='critical')
for wf in using_wfos:
if wf.status == 'staging':
print wf.name,"is doomed. setting to trouble"
wf.status = 'trouble'
session.commit()
else:
## probably enough to make a ggus and remove
if not dsname in known_lost_files:
#sendEmail('we have lost a few files','%s is missing %d files, for %d events, %f %% loss\n\n%s'%(dsname, len(lost_file_names),n_missing, fraction_loss, '\n'.join(lost_file_names)))
sendLog('stagor', '%s is missing %d files, for %d events, %f %% loss\n\n%s'%(dsname, len(lost_file_names),n_missing, fraction_loss, '\n'.join(lost_file_names)), level='critical')
known_lost_files[dsname] = [i['name'] for i in lost_files]
## should the status be change to held-staging and pending on a ticket
missings = [pid for (pid,d) in done_by_input[dsname].items() if d==False]
print "\t",done_by_input[dsname]
print "\tneeds",len(done_by_input[dsname])
print "\tgot",done_by_input[dsname].values().count(True)
print "\tmissing",missings
missing_in_action[dsname].extend( missings )
rr= open('%s/lost_blocks_datasets.json'%monitor_dir,'w')
rr.write( json.dumps( known_lost_blocks, indent=2))
rr.close()
rr= open('%s/lost_files_datasets.json'%monitor_dir,'w')
rr.write( json.dumps( known_lost_files, indent=2))
rr.close()
open('%s/incomplete_transfers.json'%monitor_dir,'w').write( json.dumps(missing_in_action, indent=2) )
print "Stuck transfers and datasets"
print json.dumps( missing_in_action, indent=2 )
datasets_by_phid = defaultdict(set)
for dataset in missing_in_action:
for phid in missing_in_action[dataset]:
#print dataset,"stuck through",phid
datasets_by_phid[phid].add( dataset )
for k in datasets_by_phid:
datasets_by_phid[k] = list(datasets_by_phid[k])
open('datasets_by_phid.json','w').write( json.dumps(datasets_by_phid, indent=2 ))
open('really_stuck_dataset.json','w').write( json.dumps(list(really_stuck_dataset), indent=2 ))
print '\n'*2,"Datasets really stuck"
print '\n'.join( really_stuck_dataset )
#############
## not going further for what matters
#############
return
def cleanor(url, specific=None):
print "Deprecated"
return
if duplicateLock() : return
delete_per_site = {}
do_not_autoapprove = []#'T2_FR_CCIN2P3']
SI = siteInfo()
CI = campaignInfo()
LI = lockInfo()
counts=0
for wfo in session.query(Workflow).filter(Workflow.status == 'done').all():
keep_a_copy = False
if specific and not specific in wfo.name: continue
## what was in input
wl = getWorkLoad(url, wfo.name )
if 'Campaign' in wl and wl['Campaign'] in CI.campaigns and 'clean-in' in CI.campaigns[wl['Campaign']] and CI.campaigns[wl['Campaign']]['clean-in']==False:
print "Skipping cleaning on input for campaign",wl['Campaign'], "as per campaign configuration"
continue
dataset= 'N/A'
if 'InputDataset' in wl:
dataset = wl['InputDataset']
print dataset,"in input"
#print json.dumps(wl, indent=2)
announced_log = filter(lambda change : change["Status"] in ["closed-out","normal-archived","announced"],wl['RequestTransition'])
if not announced_log:
print "Cannot figure out when",wfo.name,"was finished"
continue
now = time.mktime(time.gmtime()) / (60*60*24.)
then = announced_log[-1]['UpdateTime'] / (60.*60.*24.)
if (now-then) <2:
print "workflow",wfo.name, "finished",now-then,"days ago. Too fresh to clean"
continue
else:
print "workflow",wfo.name,"has finished",now-then,"days ago."
if not 'InputDataset' in wl:
## should we set status = clean ? or something even further
print "passing along",wfo.name,"with no input"
wfo.status = 'clean'
session.commit()
continue
if 'MinBias' in dataset:
print "Should not clean anything using",dataset,"setting status further"
wfo.status = 'clean'
session.commit()
continue
total_size = getDatasetSize( dataset ) ## in Gb
#if counts> 20: break
counts+=1
## find any location it is at
our_presence = getDatasetPresence(url, dataset, complete=None, group="DataOps")
also_our_presence = getDatasetPresence(url, dataset, complete=None, group="")
## is there a custodial !!!
custodials = findCustodialLocation(url, dataset)
if not len(custodials):
print dataset,"has no custodial site yet, excluding from cleaning"
continue
## find out whether it is still in use
using_the_same = getWorkflowByInput(url, dataset, details=True)
conflict=False
for other in using_the_same:
if other['RequestName'] == wfo.name: continue
if other['RequestType'] == 'Resubmission': continue
if not other['RequestStatus'] in ['announced','normal-archived','aborted','rejected','aborted-archived','aborted-completed','rejected-archived','closed-out','None',None,'new']:
print other['RequestName'],'is in status',other['RequestStatus'],'preventing from cleaning',dataset
conflict=True
break
if 'Campaign' in other and other['Campaign'] in CI.campaigns and 'clean-in' in CI.campaigns[other['Campaign']] and CI.campaigns[other['Campaign']]['clean-in']==False:
print other['RequestName'],'is in campaign',other['Campaign']
conflict = True
break
if conflict: continue
print "other statuses:",[other['RequestStatus'] for other in using_the_same if other['RequestName'] != wfo.name]
## find all disks
to_be_cleaned = filter(lambda site : site.startswith('T2') or site.endswith('Disk') ,our_presence.keys())
to_be_cleaned.extend( filter(lambda site : site.startswith('T2') or site.endswith('Disk') ,also_our_presence.keys()))
print to_be_cleaned,"for",total_size,"GB"
anaops_presence = getDatasetPresence(url, dataset, complete=None, group="AnalysisOps")
own_by_anaops = anaops_presence.keys()
print "Own by analysis ops and vetoing"
print own_by_anaops
## need to black list the sites where there is a copy of analysis ops
to_be_cleaned = [site for site in to_be_cleaned if not site in own_by_anaops ]
## keep one copy out there
if 'Campaign' in wl and wl['Campaign'] in CI.campaigns and 'keep-one' in CI.campaigns[wl['Campaign']] and CI.campaigns[wl['Campaign']]['keep-one']==True:
print "Keeping a copy of input for",wl['Campaign']
keep_a_copy = True
if keep_a_copy:
keep_at = None
full_copies = [site for (site,(there,_)) in our_presence.items() if there and site.startswith('T1')]
full_copies.extend( [site for (site,(there,_)) in also_our_presence.items() if there and site.startswith('T1')] )
if not full_copies:
full_copies = [site for (site,(there,_)) in our_presence.items() if there and site.startswith('T2')]
full_copies.extend( [site for (site,(there,_)) in also_our_presence.items() if there and site.startswith('T2')] )
if full_copies:
keep_at = random.choice( full_copies )
if not keep_at:
print "We are enable to find a place to keep a full copy of",dataset,"skipping"
continue
else:
## keeping that copy !
print "Keeping a full copy of",dataset,"at",keep_at,"not setting the status further"
to_be_cleaned.remove( keep_at )
else:
wfo.status = 'clean'
## collect delete request per site
for site in to_be_cleaned :
if not site in delete_per_site: delete_per_site[site] = []
if not dataset in [existing[0] for existing in delete_per_site[site]]:
delete_per_site[site].append( (dataset, total_size) )
session.commit()
#open('deletes.json','w').write( json.dumps(delete_per_site,indent=2) )
print json.dumps(delete_per_site, indent=2)
print "\n\n ------- \n\n"
## unroll the deletion per site
## maybe find the optimum site/dataset dataset/site to limit the number of ph requests
for site in delete_per_site:
dataset_list = [info[0] for info in delete_per_site[site]]
size_removal = sum([info[1] for info in delete_per_site[site]]) / 1024.
if site in SI.disk:
free = SI.disk[site]
print site,"has",size_removal,"TB of potential cleanup.",free,"TB available."
else:
print site,"has",size_removal,"TB of potential cleanup. no info on available."
print "\t",','.join(dataset_list)
## make deletion requests
for site in delete_per_site:
site_datasets = [info[0] for info in delete_per_site[site]]
is_tape = any([v in site for v in ['MSS','Export','Buffer'] ])
#comments="Cleanup input after production. DataOps will take care of approving it."
#if is_tape:
# comments="Cleanup input after production."
for item in site_datasets:
LI.release( item, site, 'cleanup of input after production')
def assignor(url, specific=None, talk=True, options=None):
if userLock(): return
if duplicateLock(): return
if not componentInfo().check(): return
UC = unifiedConfiguration()
CI = campaignInfo()
#SI = siteInfo()
SI = global_SI()
#NLI = newLockInfo()
#if not NLI.free() and not options.go: return
LI = lockInfo()
if not LI.free() and not options.go: return
n_assigned = 0
n_stalled = 0
wfos = []
fetch_from = []
if specific or options.early:
fetch_from.extend(['considered', 'staging'])
if specific:
fetch_from.extend(['considered-tried'])
if options.early:
print "Option Early is on"
fetch_from.extend(['staged'])
if options.from_status:
fetch_from = options.from_status.split(',')
print "Overriding to read from", fetch_from
for status in fetch_from:
print "getting wf in", status
wfos.extend(
session.query(Workflow).filter(Workflow.status == status).all())
print len(wfos)
## in case of partial, go for fetching a list from json ?
#if options.partial and not specific:
# pass
dataset_endpoints = json.loads(
open('%s/dataset_endpoints.json' % monitor_dir).read())
aaa_mapping = json.loads(
open('%s/equalizor.json' % monitor_pub_dir).read())['mapping']
all_stuck = set()
all_stuck.update(
json.loads(open('%s/stuck_transfers.json' % monitor_pub_dir).read()))
all_stuck.update(getAllStuckDataset())
max_per_round = UC.get('max_per_round').get('assignor', None)
max_cpuh_block = UC.get('max_cpuh_block')
##order by priority instead of random
if options.early:
cache = sorted(getWorkflows(url, 'assignment-approved', details=True),
key=lambda r: r['RequestPriority'])
cache = [r['RequestName'] for r in cache]
def rank(wfn):
return cache.index(wfn) if wfn in cache else 0
wfos = sorted(wfos, key=lambda wfo: rank(wfo.name), reverse=True)
print "10 first", [wfo.name for wfo in wfos[:10]]
print "10 last", [wfo.name for wfo in wfos[-10:]]
else:
random.shuffle(wfos)
for wfo in wfos:
if options.limit and (n_stalled + n_assigned) > options.limit:
break
if max_per_round and (n_stalled + n_assigned) > max_per_round:
break
if specific:
if not any(map(lambda sp: sp in wfo.name, specific.split(','))):
continue
#if not specific in wfo.name: continue
print "\n\n"
wfh = workflowInfo(url, wfo.name)
if wfh.request['RequestStatus'] in [
'rejected', 'aborted', 'aborted-completed', 'aborted-archived',
'rejected-archived'
] and wfh.isRelval():
wfo.status = 'forget'
session.commit()
n_stalled += 1
continue
if options.priority and int(
wfh.request['RequestPriority']) < options.priority:
continue
options_text = ""
if options.early: options_text += ", early option is ON"
if options.partial:
options_text += ", partial option is ON"
options_text += ", good fraction is %.2f" % options.good_enough
wfh.sendLog('assignor',
"%s to be assigned%s" % (wfo.name, options_text))
## the site whitelist takes into account siteInfo, campaignInfo, memory and cores
(lheinput, primary, parent, secondary,
sites_allowed) = wfh.getSiteWhiteList()
output_tiers = list(
set([o.split('/')[-1] for o in wfh.request['OutputDatasets']]))
if not output_tiers:
n_stalled += 1
wfh.sendLog('assignor', 'There is no output at all')
sendLog('assignor',
'Workflow %s has no output at all' % (wfo.name),
level='critical')
continue
is_stuck = (all_stuck & primary)
if is_stuck:
wfh.sendLog('assignor',
"%s are stuck input" % (','.join(is_stuck)))
## check if by configuration we gave it a GO
no_go = False
if not wfh.go(log=True) and not options.go:
no_go = True
allowed_secondary = {}
assign_parameters = {}
check_secondary = (not wfh.isRelval())
for campaign in wfh.getCampaigns():
if campaign in CI.campaigns:
assign_parameters.update(CI.campaigns[campaign])
if campaign in CI.campaigns and 'secondaries' in CI.campaigns[
campaign]:
if CI.campaigns[campaign]['secondaries']:
allowed_secondary.update(
CI.campaigns[campaign]['secondaries'])
check_secondary = True
if campaign in CI.campaigns and 'banned_tier' in CI.campaigns[
campaign]:
banned_tier = list(
set(CI.campaigns[campaign]['banned_tier'])
& set(output_tiers))
if banned_tier:
no_go = True
wfh.sendLog(
'assignor', 'These data tiers %s are not allowed' %
(','.join(banned_tier)))
sendLog('assignor',
'These data tiers %s are not allowed' %
(','.join(banned_tier)),
level='critical')
if secondary and check_secondary:
if (set(secondary) & set(allowed_secondary.keys()) !=
set(secondary)):
wfh.sendLog(
'assignor', '%s is not an allowed secondary' %
(', '.join(set(secondary) -
set(allowed_secondary.keys()))))
sendLog(
'assignor',
'%s is not an allowed secondary' %
(', '.join(set(secondary) -
set(allowed_secondary.keys()))),
level='critical')
if not options.go:
no_go = True
## then get whether there is something more to be done by secondary
for sec in secondary:
if sec in allowed_secondary: # and 'parameters' in allowed_secondary[sec]:
assign_parameters.update(allowed_secondary[sec])
if no_go:
n_stalled += 1
## make a very loud noise if >100k priority stalled
continue
## check on current status for by-passed assignment
if wfh.request['RequestStatus'] != 'assignment-approved':
if not options.test:
wfh.sendLog('assignor',
"setting %s away and skipping" % wfo.name)
## the module picking up from away will do what is necessary of it
wfo.wm_status = wfh.request['RequestStatus']
wfo.status = 'away'
session.commit()
continue
else:
print wfo.name, wfh.request['RequestStatus']
## retrieve from the schema, dbs and reqMgr what should be the next version
version = wfh.getNextVersion()
if not version:
if options and options.ProcessingVersion:
version = options.ProcessingVersion
else:
wfh.sendLog('assignor', "cannot decide on version number")
n_stalled += 1
wfo.status = 'trouble'
session.commit()
continue
original_sites_allowed = copy.deepcopy(sites_allowed)
wfh.sendLog('assignor', "Site white list %s" % sorted(sites_allowed))
override_sec_location = CI.get(wfh.request['Campaign'],
'SecondaryLocation', [])
blocks = wfh.getBlockWhiteList()
rwl = wfh.getRunWhiteList()
if rwl:
## augment with run white list
for dataset in primary:
blocks = list(set(blocks +
getDatasetBlocks(dataset, runs=rwl)))
lwl = wfh.getLumiWhiteList()
if lwl:
## augment with lumi white list
for dataset in primary:
blocks = list(
set(blocks + getDatasetBlocks(dataset, lumis=lwl)))
wfh.sendLog('assignor', "Allowed %s" % sorted(sites_allowed))
secondary_locations = None
primary_aaa = options.primary_aaa
secondary_aaa = options.secondary_aaa
do_partial = False #options.good_enough if options.partial else 0
if 'Campaign' in wfh.request and wfh.request[
'Campaign'] in CI.campaigns:
assign_parameters.update(CI.campaigns[wfh.request['Campaign']])
if 'primary_AAA' in assign_parameters:
primary_aaa = primary_aaa or assign_parameters['primary_AAA']
if 'secondary_AAA' in assign_parameters:
secondary_aaa = secondary_aaa or assign_parameters['secondary_AAA']
if 'partial_copy' in assign_parameters:
## can this only work if there is a stuck input ? maybe not
## this is a number. 0 means no
print "Could do partial disk copy assignment"
if is_stuck or options.partial:
do_partial = assign_parameters['partial_copy']
wfh.sendLog(
'assignor',
"Overiding partial copy assignment to %.2f fraction" %
do_partial)
#sendEmail('stuck input to assignment','%s is stuck for assigning %s and going fractional'%(','.join( is_stuck), wfo.name))
do_partial = options.good_enough if options.partial else do_partial
for sec in list(secondary):
if override_sec_location:
print "We don't care where the secondary is"
print "Cannot pass for now"
#sendEmail("tempting to pass sec location check","but we cannot yet IMO")
#pass
presence = getDatasetPresence(url, sec)
print sec
print json.dumps(presence, indent=2)
one_secondary_locations = [
site for (site, (there, frac)) in presence.items()
if frac > 98.
]
if secondary_aaa:
if not one_secondary_locations:
sec_availability = getDatasetBlocksFraction(url, sec)
if sec_availability >= 1. and options.go:
## there is at least one copy of each block on disk. We should go ahead and let it go.
wfh.sendLog(
'assignor',
"The secondary %s is available %s times on disk, and usable"
% (sec, sec_availability))
else:
## not even a copy on disk anywhere !!!!
sites_allowed = [] ## will block the assignment
wfh.sendLog(
'assignor',
"The secondary %s is nowhere on disk" % sec)
#just continue without checking
continue
#one_secondary_locations = [site for (site,(there,frac)) in presence.items() if there]
if secondary_locations == None:
secondary_locations = one_secondary_locations
else:
secondary_locations = list(
set(secondary_locations) & set(one_secondary_locations))
## reduce the site white list to site with secondary only
#sites_allowed = [site for site in sites_allowed if any([osite.startswith(site) for osite in one_secondary_locations])]
sites_allowed = [
site for site in sites_allowed
if SI.CE_to_SE(site) in one_secondary_locations
]
wfh.sendLog(
'assignor', "From/after secondary requirement, now Allowed%s" %
sorted(sites_allowed))
initial_sites_allowed = copy.deepcopy(
sites_allowed
) ## keep track of this, after secondary input location restriction : that's how you want to operate it
sites_all_data = copy.deepcopy(sites_allowed)
sites_with_data = copy.deepcopy(sites_allowed)
sites_with_any_data = copy.deepcopy(sites_allowed)
primary_locations = None
available_fractions = {}
set_lfn = '/store/mc' ## by default
endpoints = set()
for prim in list(primary):
if prim in dataset_endpoints:
print "endpoints from stagor", dataset_endpoints[prim]
endpoints.update(dataset_endpoints[prim])
set_lfn = getLFNbase(prim)
presence = getDatasetPresence(url, prim, only_blocks=blocks)
if talk:
print prim
print json.dumps(presence, indent=2)
available_fractions[prim] = getDatasetBlocksFraction(
url,
prim,
sites=[SI.CE_to_SE(site) for site in sites_allowed],
only_blocks=blocks)
if primary_aaa:
available_fractions[prim] = getDatasetBlocksFraction(
url, prim, only_blocks=blocks)
sites_all_data = [
site for site in sites_with_data if SI.CE_to_SE(site) in [
psite for (psite, (there, frac)) in presence.items()
if there
]
]
if primary_aaa:
sites_all_data = list(
set([
SI.SE_to_CE(psite)
for (psite, (there, frac)) in presence.items() if there
]))
sites_with_data = [
site for site in sites_with_data if SI.CE_to_SE(site) in
[psite for (psite, frac) in presence.items() if frac[1] > 90.]
]
sites_with_any_data = [
site for site in sites_with_any_data
if SI.CE_to_SE(site) in presence.keys()
]
if primary_aaa:
sites_with_any_data = list(
set([SI.SE_to_CE(psite) for psite in presence.keys()]))
wfh.sendLog(
'assignor', "Holding the data but not allowed %s" % sorted(
list(
set([
se_site for se_site in presence.keys()
if not SI.SE_to_CE(se_site) in sites_allowed
]))))
if primary_locations == None:
primary_locations = presence.keys()
else:
primary_locations = list(
set(primary_locations) & set(presence.keys()))
sites_with_data = list(set(sites_with_data))
sites_with_any_data = list(set(sites_with_any_data))
opportunistic_sites = []
down_time = False
## opportunistic running where any piece of data is available
if secondary_locations or primary_locations:
## intersection of both any pieces of the primary and good IO
#opportunistic_sites = [SI.SE_to_CE(site) for site in list((set(secondary_locations) & set(primary_locations) & set(SI.sites_with_goodIO)) - set(sites_allowed))]
if secondary_locations and primary_locations:
opportunistic_sites = [
SI.SE_to_CE(site) for site in
list((set(secondary_locations) & set(primary_locations)) -
set([SI.CE_to_SE(site) for site in sites_allowed]))
]
elif primary_locations:
opportunistic_sites = [
SI.SE_to_CE(site) for site in list(
set(primary_locations) -
set([SI.CE_to_SE(site) for site in sites_allowed]))
]
else:
opportunistic_sites = []
wfh.sendLog(
'assignor', "We could be running in addition at %s" %
sorted(opportunistic_sites))
if any(
[osite in SI.sites_not_ready
for osite in opportunistic_sites]):
wfh.sendLog(
'assignor', "One of the usable site is in downtime %s" % ([
osite for osite in opportunistic_sites
if osite in SI.sites_not_ready
]))
down_time = True
## should this be send back to considered ?
## should be 2 but for the time-being let's lower it to get things going
copies_wanted, cpuh = wfh.getNCopies()
wfh.sendLog('assignor', "we need %s CPUh" % cpuh)
if cpuh > max_cpuh_block and not options.go:
#sendEmail('large workflow','that wf %s has a large number of CPUh %s, not assigning, please check the logs'%(wfo.name, cpuh))#,destination=['*****@*****.**'])
sendLog(
'assignor',
'%s requires a large numbr of CPUh %s , not assigning, please check with requester'
% (wfo.name, cpuh),
level='critical')
wfh.sendLog(
'assignor',
"Requiring a large number of CPUh %s, not assigning" % cpuh)
continue
if 'Campaign' in wfh.request and wfh.request[
'Campaign'] in CI.campaigns and 'maxcopies' in CI.campaigns[
wfh.request['Campaign']]:
copies_needed_from_campaign = CI.campaigns[
wfh.request['Campaign']]['maxcopies']
copies_wanted = min(copies_needed_from_campaign, copies_wanted)
if not options.early:
less_copies_than_requested = UC.get("less_copies_than_requested")
copies_wanted = max(
1, copies_wanted -
less_copies_than_requested) # take one out for the efficiency
else:
## find out whether there is a site in the whitelist, that is lacking jobs and reduce to 1 copy needed to get things going
pass
wfh.sendLog('assignor',
"needed availability fraction %s" % copies_wanted)
## should also check on number of sources, if large enough, we should be able to overflow most, efficiently
## default back to white list to original white list with any data
wfh.sendLog('assignor', "Allowed sites :%s" % sorted(sites_allowed))
if primary_aaa:
## remove the sites not reachable localy if not in having the data
if not sites_all_data:
wfh.sendLog('assignor',
"Overiding the primary on AAA setting to Off")
primary_aaa = False
else:
aaa_grid = set(sites_all_data)
for site in list(aaa_grid):
aaa_grid.update(aaa_mapping.get(site, []))
sites_allowed = list(set(initial_sites_allowed) & aaa_grid)
wfh.sendLog(
'assignor', "Selected to read primary through xrootd %s" %
sorted(sites_allowed))
isStoreResults = ('StoreResults' == wfh.request.setdefault(
'RequestType', None))
if isStoreResults:
if 'MergedLFNBase' in wfh.request:
set_lfn = wfh.request['MergedLFNBase']
else:
n_stalled += 1
wfh.sendLog(
'assignor',
"Cannot assign StoreResults request because MergedLFN is missing"
)
sendLog(
'assignor',
'Cannot assign StoreResults request because MergedLFN is missing',
level='critical')
continue
if not primary_aaa:
if not isStoreResults:
sites_allowed = sites_with_any_data
else:
## if we are dealing with a StoreResults request, we don't need to check dataset availability and
## should use the SiteWhiteList set in the original request
if 'SiteWhitelist' in wfh.request:
sites_allowed = wfh.request['SiteWhitelist']
else:
wfh.sendLog(
'assignor',
"Cannot assign StoreResults request because SiteWhitelist is missing"
)
sendLog(
'assignor',
'Cannot assign StoreResults request because SiteWhitelist is missing',
level='critical')
n_stalled += 1
continue
available_fractions = {}
wfh.sendLog('assignor',
"Selected for any data %s" % sorted(sites_allowed))
### check on endpoints for on-going transfers
if do_partial:
if endpoints:
end_sites = [SI.SE_to_CE(s) for s in endpoints]
sites_allowed = list(set(sites_allowed + end_sites))
if down_time and not any(osite in SI.sites_not_ready
for osite in end_sites):
print "Flip the status of downtime, since our destinations are good"
down_time = False
print "with added endpoints", sorted(end_sites)
else:
print "Cannot do partial assignment without knowin the endpoints"
n_stalled += 1
continue
#if not len(sites_allowed):
# if not options.early:
# wfh.sendLog('assignor',"cannot be assign with no matched sites")
# sendLog('assignor','%s has no whitelist'% wfo.name, level='critical')
# n_stalled+=1
# continue
low_pressure = SI.sites_low_pressure(0.4)
## if any of the site allowed is low pressure : reduce to 1 copy so that it gets started
allowed_and_low = sorted(set(low_pressure) & set(sites_allowed))
if allowed_and_low:
wfh.sendLog(
'assignor',
"The workflow can run at %s under low pressure currently" %
(','.join(allowed_and_low)))
copies_wanted = max(1., copies_wanted - 1.)
if available_fractions and not all([
available >= copies_wanted
for available in available_fractions.values()
]):
not_even_once = not all([
available >= 1. for available in available_fractions.values()
])
above_good = all([
available >= do_partial
for available in available_fractions.values()
])
wfh.sendLog(
'assignor',
"The input dataset is not available %s times, only %s" %
(copies_wanted, available_fractions.values()))
if down_time and not options.go and not options.early:
wfo.status = 'considered'
session.commit()
wfh.sendLog(
'assignor',
"sending back to considered because of site downtime, instead of waiting"
)
#sendEmail( "cannot be assigned due to downtime","%s is not sufficiently available, due to down time of a site in the whitelist. check the assignor logs. sending back to considered."% wfo.name)
sendLog(
'assignor',
'%s is not sufficiently available, due to down time of a site in the whitelist. sending back to considered.'
% (wfo.name),
level='delay')
n_stalled += 1
continue
#pass
print json.dumps(available_fractions)
if (options.go and not_even_once) or not options.go:
known = []
try:
known = json.loads(open('cannot_assign.json').read())
except:
pass
if not wfo.name in known and not options.limit and not options.go and not options.early and not (
do_partial and above_good):
wfh.sendLog(
'assignor',
"cannot be assigned, %s is not sufficiently available.\n %s"
% (wfo.name, json.dumps(available_fractions)))
#sendEmail( "cannot be assigned","%s is not sufficiently available.\n %s"%(wfo.name,json.dumps(available_fractions)))
known.append(wfo.name)
open('cannot_assign.json',
'w').write(json.dumps(known, indent=2))
if options.early:
if wfo.status == 'considered':
wfh.sendLog('assignor', "setting considered-tried")
wfo.status = 'considered-tried'
session.commit()
else:
print "tried but status is", wfo.status
if do_partial and above_good:
print "Will move on with partial locations"
else:
n_stalled += 1
continue
if not len(sites_allowed):
if not options.early:
wfh.sendLog('assignor',
"cannot be assign with no matched sites")
sendLog('assignor',
'%s has no whitelist' % wfo.name,
level='critical')
n_stalled += 1
continue
t1_only = [ce for ce in sites_allowed if ce.startswith('T1')]
if t1_only:
# try to pick from T1 only first
sites_out = [SI.pick_dSE([SI.CE_to_SE(ce) for ce in t1_only])]
else:
# then pick any otherwise
sites_out = [
SI.pick_dSE([SI.CE_to_SE(ce) for ce in sites_allowed])
]
wfh.sendLog('assignor', "Placing the output on %s" % sites_out)
parameters = {
'SiteWhitelist': sites_allowed,
'NonCustodialSites': sites_out,
'AutoApproveSubscriptionSites': list(set(sites_out)),
'AcquisitionEra': wfh.acquisitionEra(),
'ProcessingString': wfh.processingString(),
'MergedLFNBase': set_lfn,
'ProcessingVersion': version,
}
if primary_aaa:
parameters['TrustSitelists'] = True
wfh.sendLog(
'assignor',
"Reading primary through xrootd at %s" % sorted(sites_allowed))
if secondary_aaa:
parameters['TrustPUSitelists'] = True
wfh.sendLog(
'assignor', "Reading secondary through xrootd at %s" %
sorted(sites_allowed))
## plain assignment here
team = 'production'
if os.getenv('UNIFIED_TEAM'): team = os.getenv('UNIFIED_TEAM')
if options and options.team:
team = options.team
parameters['Team'] = team
if lheinput:
## throttle reading LHE article
wfh.sendLog('assignor',
'Setting the number of events per job to 500k max')
parameters['EventsPerJob'] = 500000
def pick_options(options, parameters):
##parse options entered in command line if any
if options:
for key in reqMgrClient.assignWorkflow.keys:
v = getattr(options, key)
if v != None:
if type(v) == str and ',' in v:
parameters[key] = filter(None, v.split(','))
else:
parameters[key] = v
def pick_campaign(assign_parameters, parameters):
## pick up campaign specific assignment parameters
parameters.update(assign_parameters.get('parameters', {}))
if options.force_options:
pick_campaign(assign_parameters, parameters)
pick_options(options, parameters)
else:
## campaign parameters update last
pick_options(options, parameters)
pick_campaign(assign_parameters, parameters)
if not options.test:
parameters['execute'] = True
hold_split, split_check = wfh.checkSplitting()
if hold_split and not options.go:
if split_check:
wfh.sendLog(
'assignor', 'Holding on to the change in splitting %s' %
('\n\n'.join([str(i) for i in split_check])))
else:
wfh.sendLog('assignor', 'Change of splitting is on hold')
n_stalled += 1
continue
if split_check == None or split_check == False:
n_stalled += 1
continue
elif split_check:
## operate all recommended changes
reqMgrClient.setWorkflowSplitting(url, wfo.name, split_check)
wfh.sendLog(
'assignor', 'Applying the change in splitting %s' %
('\n\n'.join([str(i) for i in split_check])))
split_check = True ## bypass completely and use the above
# Handle run-dependent MC
pstring = wfh.processingString()
if 'PU_RD' in pstring:
numEvents = wfh.getRequestNumEvents()
eventsPerLumi = [getDatasetEventsPerLumi(prim) for prim in primary]
eventsPerLumi = sum(eventsPerLumi) / float(len(eventsPerLumi))
reqJobs = 500
if 'PU_RD2' in pstring:
reqJobs = 2000
eventsPerJob = int(numEvents / (reqJobs * 1.4))
lumisPerJob = int(eventsPerJob / eventsPerLumi)
if lumisPerJob == 0:
#sendEmail("issue with event splitting for run-dependent MC","%s needs to be split by event with %s per job"%(wfo.name, eventsPerJob))
sendLog('assignor',
"%s needs to be split by event with %s per job" %
(wfo.name, eventsPerJob),
level='critical')
wfh.sendLog(
'assignor',
"%s needs to be split by event with %s per job" %
(wfo.name, eventsPerJob))
parameters['EventsPerJob'] = eventsPerJob
else:
spl = wfh.getSplittings()[0]
eventsPerJobEstimated = spl[
'events_per_job'] if 'events_per_job' in spl else None
eventsPerJobEstimated = spl[
'avg_events_per_job'] if 'avg_events_per_job' in spl else None
if eventsPerJobEstimated and eventsPerJobEstimated > eventsPerJob:
#sendEmail("setting lumi splitting for run-dependent MC","%s was assigned with %s lumis/job"%( wfo.name, lumisPerJob))
sendLog('assignor',
"%s was assigned with %s lumis/job" %
(wfo.name, lumisPerJob),
level='critical')
wfh.sendLog(
'assignor', "%s was assigned with %s lumis/job" %
(wfo.name, lumisPerJob))
parameters['LumisPerJob'] = lumisPerJob
else:
#sendEmail("leaving splitting untouched for PU_RD*","please check on "+wfo.name)
sendLog(
'assignor',
"leaving splitting untouched for %s, please check on %s"
% (pstring, wfo.name),
level='critical')
wfh.sendLog(
'assignor',
"leaving splitting untouched for PU_RD*, please check."
)
if isHEPCloudReady(url) and wfh.isGoodForNERSC():
parameters['Team'] = 'hepcloud'
parameters['SiteWhitelist'] = ['T3_US_NERSC']
if primary:
parameters['TrustSitelists'] = True
if secondary:
parameters['TrustPUSitelists'] = True
sendEmail("sending work to hepcloud",
"pleasse check on %s" % wfh.request['RequestName'],
destination=['*****@*****.**'])
## make sure to autoapprove all NonCustodialSites
parameters['AutoApproveSubscriptionSites'] = list(
set(parameters['NonCustodialSites'] +
parameters.get('AutoApproveSubscriptionSites', [])))
result = reqMgrClient.assignWorkflow(
url, wfo.name, None,
parameters) ## team is not relevant anymore here
# set status
if not options.test:
if result:
wfo.status = 'away'
session.commit()
n_assigned += 1
wfh.sendLog(
'assignor', "Properly assigned\n%s" %
(json.dumps(parameters, indent=2)))
try:
## refetch information and lock output
new_wfi = workflowInfo(url, wfo.name)
(_, prim, _, sec) = new_wfi.getIO()
for secure in list(prim) + list(
sec) + new_wfi.request['OutputDatasets']:
## lock all outputs
LI.lock(secure, reason='assigning')
except Exception as e:
print "fail in locking output"
print str(e)
sendEmail("failed locking of output", str(e))
else:
wfh.sendLog(
'assignor',
"Failed to assign %s.\n%s \n Please check the logs" %
(wfo.name, reqMgrClient.assignWorkflow.errorMessage))
sendLog('assignor',
"Failed to assign %s.\n%s \n Please check the logs" %
(wfo.name, reqMgrClient.assignWorkflow.errorMessage),
level='critical')
print "ERROR could not assign", wfo.name
else:
pass
print "Assignment summary:"
sendLog('assignor', "Assigned %d Stalled %s" % (n_assigned, n_stalled))
if n_stalled and not options.go and not options.early:
sendLog('assignor',
"%s workflows cannot be assigned. Please take a look" %
(n_stalled),
level='critical')
def assignor(url ,specific = None, talk=True, options=None):
if userLock('assignor'): return
CI = campaignInfo()
SI = siteInfo()
wfos=[]
if specific:
wfos = session.query(Workflow).filter(Workflow.name==specific).all()
if not wfos:
if specific:
wfos = session.query(Workflow).filter(Workflow.status=='considered').all()
wfos.extend( session.query(Workflow).filter(Workflow.status=='staging').all())
wfos.extend(session.query(Workflow).filter(Workflow.status=='staged').all())
for wfo in wfos:
if specific:
if not any(map(lambda sp: sp in wfo.name,specific.split(','))): continue
#if not specific in wfo.name: continue
print wfo.name,"to be assigned"
wfh = workflowInfo( url, wfo.name)
## check if by configuration we gave it a GO
if not CI.go( wfh.request['Campaign'] ) and not options.go:
print "No go for",wfh.request['Campaign']
continue
## check on current status for by-passed assignment
if wfh.request['RequestStatus'] !='assignment-approved':
print wfo.name,wfh.request['RequestStatus'],"skipping"
if not options.test:
continue
## retrieve from the schema, dbs and reqMgr what should be the next version
version=wfh.getNextVersion()
if not version:
if options and options.ProcessingVersion:
version = options.ProcessingVersion
else:
print "cannot decide on version number"
continue
(lheinput,primary,parent,secondary) = wfh.getIO()
sites_allowed = getSiteWhiteList( (lheinput,primary,parent,secondary) )
print "Allowed",sites_allowed
sites_out = [SI.pick_dSE([SI.CE_to_SE(ce) for ce in sites_allowed])]
sites_custodial = []
if len(sites_custodial)==0:
print "No custodial, it's fine, it's covered in close-out"
if len(sites_custodial)>1:
print "more than one custodial for",wfo.name
sys.exit(36)
secondary_locations=None
for sec in list(secondary):
presence = getDatasetPresence( url, sec )
print sec
print json.dumps(presence, indent=2)
#one_secondary_locations = [site for (site,(there,frac)) in presence.items() if frac>90.]
one_secondary_locations = [site for (site,(there,frac)) in presence.items() if there]
if secondary_locations==None:
secondary_locations = one_secondary_locations
else:
secondary_locations = list(set(secondary_locations) & set(one_secondary_locations))
## reduce the site white list to site with secondary only
sites_allowed = [site for site in sites_allowed if any([osite.startswith(site) for osite in one_secondary_locations])]
sites_all_data = copy.deepcopy( sites_allowed )
sites_with_data = copy.deepcopy( sites_allowed )
sites_with_any_data = copy.deepcopy( sites_allowed )
primary_locations = None
available_fractions = {}
for prim in list(primary):
presence = getDatasetPresence( url, prim )
if talk:
print prim
print json.dumps(presence, indent=2)
available_fractions[prim] = getDatasetBlocksFraction(url, prim, sites = [SI.CE_to_SE(site) for site in sites_allowed] )
sites_all_data = [site for site in sites_with_data if any([osite.startswith(site) for osite in [psite for (psite,(there,frac)) in presence.items() if there]])]
sites_with_data = [site for site in sites_with_data if any([osite.startswith(site) for osite in [psite for (psite,frac) in presence.items() if frac[1]>90.]])]
sites_with_any_data = [site for site in sites_with_any_data if any([osite.startswith(site) for osite in presence.keys()])]
if primary_locations==None:
primary_locations = presence.keys()
else:
primary_locations = list(set(primary_locations) & set(presence.keys() ))
sites_with_data = list(set(sites_with_data))
sites_with_any_data = list(set(sites_with_any_data))
opportunistic_sites=[]
## opportunistic running where any piece of data is available
if secondary_locations and primary_locations:
## intersection of both any pieces of the primary and good IO
#opportunistic_sites = [SI.SE_to_CE(site) for site in list((set(secondary_locations) & set(primary_locations) & set(SI.sites_with_goodIO)) - set(sites_allowed))]
opportunistic_sites = [SI.SE_to_CE(site) for site in list((set(secondary_locations) & set(primary_locations)) - set(sites_allowed))]
print "We could be running at",opportunistic_sites,"in addition"
if available_fractions and not all([available>=1. for available in available_fractions.values()]):
print "The input dataset is not located in full at any site"
print json.dumps(available_fractions)
if not options.test and not options.go: continue ## skip skip skip
copies_wanted = 2.
if available_fractions and not all([available>=copies_wanted for available in available_fractions.values()]):
print "The input dataset is not available",copies_wanted,"times, only",available_fractions.values()
if not options.go:
continue
## default back to white list to original white list with any data
print "Allowed",sites_allowed
sites_allowed = sites_with_any_data
print "Selected for any data",sites_allowed
if options.restrict:
print "Allowed",sites_allowed
sites_allowed = sites_with_any_data
print "Selected",sites_allowed
else:
if set(sites_with_data) != set(sites_allowed):
## the data is not everywhere we wanted to run at : enable aaa
print "Sites with 90% data not matching site white list (block choping!)"
print "Resorting to AAA reading for",list(set(sites_allowed) - set(sites_with_data)),"?"
print "Whitelist site with any data",list(set(sites_allowed) - set(sites_with_any_data))
#options.useSiteListAsLocation = True
#print "Not commissioned yet"
#continue
#print "We could be running at",opportunistic_sites,"in addition"
##sites_allowed = list(set(sites_allowed+ opportunistic_sites))
if not len(sites_allowed):
print wfo.name,"cannot be assign with no matched sites"
continue
parameters={
'SiteWhitelist' : sites_allowed,
'CustodialSites' : sites_custodial,
'NonCustodialSites' : sites_out,
'AutoApproveSubscriptionSites' : list(set(sites_out)),
'AcquisitionEra' : wfh.acquisitionEra(),
'ProcessingString' : wfh.processingString(),
'MergedLFNBase' : '/store/mc', ## to be figured out ! from Hi shit
'ProcessingVersion' : version,
}
##parse options entered in command line if any
if options:
for key in reqMgrClient.assignWorkflow.keys:
v=getattr(options,key)
if v!=None:
if ',' in v: parameters[key] = filter(None,v.split(','))
else: parameters[key] = v
## pick up campaign specific assignment parameters
parameters.update( CI.parameters(wfh.request['Campaign']) )
if not options.test:
parameters['execute'] = True
if not wfh.checkWorkflowSplitting():
## needs to go to event based ? fail for now
print "Falling back to event splitting ?"
#parameters['SplittingAlgorithm'] = 'EventBased'
continue
## plain assignment here
team='production'
if options and options.team:
team = options.team
result = reqMgrClient.assignWorkflow(url, wfo.name, team, parameters)
# set status
if not options.test:
if result:
wfo.status = 'away'
session.commit()
else:
print "ERROR could not assign",wfo.name
else:
pass
def outcleanor(url, options):
if options.approve:
for user in ['*Vlimant']: #,'*Cremonesi']:
deletes = listDelete(url, user=user)
for (site, who, tid) in deletes:
if 'MSS' in site: continue ### ever
print site, who, tid
print "approving deletion"
print approveSubscription(
url,
tid,
nodes=[site],
comments='Production cleaning by data ops')
return
sites_and_datasets = defaultdict(list)
our_copies = defaultdict(list)
wf_cleaned = {}
wfs = []
for fetch in options.fetch.split(','):
wfs.extend(
session.query(Workflow).filter(Workflow.status == fetch).all())
random.shuffle(wfs)
last_answer = None
for wfo in wfs:
if options.number and len(wf_cleaned) >= options.number:
print "Reached", options.number, "cleaned"
break
print '-' * 100
wfi = workflowInfo(url, wfo.name)
goes = {} # boolean per output
for dataset in wfi.request['OutputDatasets']:
goes[dataset] = False
keep_one_out = True
status = getDatasetStatus(dataset)
print "\n\tLooking at", dataset, status, "\n"
vetoes = None
if status == 'INVALID':
vetoes = ['Export', 'Buffer'] ## can take themselves out
keep_one_out = False # just wipe clean
elif status == None:
print dataset, "actually does not exist. skip"
goes[dataset] = True
continue
elif status in ['PRODUCTION', 'VALID'
] and wfo.status in ['forget', 'trouble']:
print dataset, "should probably be invalidated. (", wfo.status, ") skip"
keep_one_out = False # just wipe clean
continue ## you are not sure. just skip it for the time being
elif status == 'PRODUCTION' and wfo.status in ['clean']:
print dataset, "should probably be set valid .skip"
continue ## you are not sure. just skip it for the time being
if status == 'VALID' and dataset.startswith('/MinBias'):
print "This is a /MinBias. skip"
continue
if '/DQM' in dataset:
keep_one_out = False
total_size = getDatasetSize(dataset)
our_presence = getDatasetPresence(url,
dataset,
complete=None,
group="DataOps",
vetoes=vetoes)
also_our_presence = getDatasetPresence(url,
dataset,
complete=None,
group="",
vetoes=vetoes)
## merge in one unique dict
for site in also_our_presence:
if site in our_presence:
there, frac = our_presence[site]
other, ofrac = also_our_presence[site]
our_presence[site] = (max(there, other), max(frac, ofrac))
else:
our_presence[site] = also_our_presence[site]
if our_presence: print our_presence
## analysis ops copies need to be taken into account
anaops_presence = getDatasetPresence(url,
dataset,
complete=None,
group="AnalysisOps")
own_by_anaops = anaops_presence.keys()
## all our copies
to_be_cleaned = our_presence.keys()
if not len(to_be_cleaned):
print "nowhere to be found of ours,", len(
own_by_anaops), "in analysi ops pool"
goes[dataset] = True
continue
print "Where we own bits of dataset"
print to_be_cleaned
if len(own_by_anaops):
## remove site with the anaops copies
to_be_cleaned = list(set(to_be_cleaned) - set(own_by_anaops))
keep_one_out = False ## in that case, just remove our copies
print "Own by anaops (therefore not keep a copy of ours)"
print own_by_anaops
else:
## we should not be looking at anything that was not passed to DDM, otherwise we'll be cutting the grass under our feet
using_the_same = getWorkflowByInput(url, dataset, details=True)
conflict = False
for other in using_the_same:
if other['RequestName'] == wfo.name: continue
if other['RequestType'] == 'Resubmission': continue
if not other['RequestStatus'] in [
'announced', 'normal-archived', 'aborted',
'rejected', 'aborted-archived',
'rejected-archived', 'closed-out', 'None', None
]:
print other['RequestName'], 'is in status', other[
'RequestStatus'], 'preventing from cleaning', dataset
conflict = True
break
if conflict:
continue
## not being used. a bit less dangerous to clean-out
## keep one full copy out there
full_copies = [
site for (site, (there, fract)) in our_presence.items()
if there
]
if keep_one_out:
if not len(full_copies):
print "we do not own a full copy of", dataset, status, wfo.status, ".skip"
continue
stay_there = random.choice(
full_copies) #at a place own by ops
print "Where we keep a full copy", stay_there
to_be_cleaned.remove(stay_there)
our_copies[stay_there].append(dataset)
else:
print "We do not want to keep a copy of ", dataset, status, wfo.status
if len(to_be_cleaned):
print "Where we can clean"
print to_be_cleaned
for site in to_be_cleaned:
sites_and_datasets[site].append(
(dataset, total_size * our_presence[site][1] / 100.,
status))
goes[dataset] = True
else:
print "no cleaning to be done"
goes[dataset] = True
print wfo.name, "scrutinized"
if all(goes.values()):
print "\t", wfo.name, "can toggle -out"
def ask():
global last_answer
last_answer = raw_input('go on ?')
return last_answer
if options.auto or ask() in ['y', '']:
if all(goes.values()):
wfo.status = wfo.status + '-out'
wf_cleaned[wfo.name] = wfo.status
continue
elif last_answer in ['q', 'n']:
break
else:
return
if options.auto:
pass
elif last_answer in ['q']:
return
print "Potential cleanups"
for (site, items) in sites_and_datasets.items():
cleanup = sum([size for (_, size, _) in items])
print "\n\t potential cleanup of", "%8.4f" % cleanup, "GB at ", site
print "\n".join([ds + " " + st for ds, _, st in items])
datasets = [ds for ds, _, st in items]
print "Copies and bits we are going to delete"
print json.dumps(sites_and_datasets, indent=2)
print "Copies we are keeping"
print json.dumps(our_copies, indent=2)
print "Workflows cleaned for output"
print json.dumps(wf_cleaned, indent=2)
stamp = time.strftime("%Y%m%d%H%M%S", time.localtime())
open('outcleaning_%s.json' % stamp,
'w').write(json.dumps(sites_and_datasets, indent=2))
open('keepcopies_%s.json' % stamp,
'w').write(json.dumps(our_copies, indent=2))
open('wfcleanout_%s.json' % stamp,
'w').write(json.dumps(wf_cleaned, indent=2))
if (not options.test) and (options.auto or raw_input(
"Satisfied ? (y will trigger status change and deletion requests)")
in ['y']):
for (site, items) in sites_and_datasets.items():
datasets = [ds for ds, _, st in items]
print "making deletion to", site
result = makeDeleteRequest(
url, site, datasets,
"Cleanup output after production. DataOps will take care of approving it."
)
print result
## approve it right away ?
if 'MSS' in site: continue
if 'Export' in site: continue
if 'Buffer' in site: continue
for did in [
item['id'] for item in result['phedex']['request_created']
]:
print "auto-approve disabled, but ready"
#approveSubscription(url, did, nodes = [site], comments = 'Auto-approving production cleaning deletion')
pass
session.commit()
else:
print "Not making the deletion and changing statuses"
def transferor(url ,specific = None, talk=True, options=None):
if options and options.test:
execute = False
else:
execute = True
SI = siteInfo()
CI = campaignInfo()
all_transfers=defaultdict(list)
workflow_dependencies = defaultdict(set) ## list of wf.id per input dataset
data_to_wf = {}
for wfo in session.query(Workflow).filter(Workflow.status=='considered').all():
if specific and not specific in wfo.name: continue
print wfo.name,"to be transfered"
wfh = workflowInfo( url, wfo.name)
#injection_time = time.mktime(time.strptime('.'.join(map(str,wfh.request['RequestDate'])),"%Y.%m.%d.%H.%M.%S")) / (60.*60.)
#now = time.mktime(time.gmtime()) / (60.*60.)
#if float(now - injection_time) < 4.:
# print "It is too soon to transfer", now, injection_time
# continue
(lheinput,primary,parent,secondary) = wfh.getIO()
if options and options.tosites:
sites_allowed = options.tosites.split(',')
else:
sites_allowed = getSiteWhiteList( (lheinput,primary,parent,secondary) )
if 'SiteWhitelist' in CI.parameters(wfh.request['Campaign']):
sites_allowed = CI.parameters(wfh.request['Campaign'])['SiteWhitelist']
blocks = []
if 'BlockWhitelist' in wfh.request and wfh.request['BlockWhitelist']:
blocks = wfh.request['BlockWhitelist']
can_go = True
if primary:
if talk:
print wfo.name,'reads',', '.join(primary),'in primary'
## chope the primary dataset
for prim in primary:
workflow_dependencies[prim].add( wfo.id )
presence = getDatasetPresence( url, prim )
prim_location = [site for site,pres in presence.items() if pres[0]==True]
subscriptions = listSubscriptions( url , prim )
prim_destination = [site for site in subscriptions]
prim_to_distribute = [site for site in sites_allowed if not any([osite.startswith(site) for osite in prim_location])]
prim_to_distribute = [site for site in prim_to_distribute if not any([osite.startswith(site) for osite in prim_destination])]
if len(prim_to_distribute)>0: ## maybe that a parameter we can play with to limit the
if not options or options.chop:
spreading = distributeToSites( [[prim]]+getDatasetChops(prim), prim_to_distribute, n_copies = 3, weights=SI.cpu_pledges)
else:
spreading = {}
for site in prim_to_distribute: spreading[site]=[prim]
can_go = False
for (site,items) in spreading.items():
all_transfers[site].extend( items )
if secondary:
if talk:
print wfo.name,'reads',', '.join(secondary),'in secondary'
for sec in secondary:
workflow_dependencies[sec].add( wfo.id )
presence = getDatasetPresence( url, sec )
sec_location = [site for site,pres in presence.items() if pres[1]>90.] ## more than 90% of the minbias at sites
subscriptions = listSubscriptions( url ,sec )
sec_destination = [site for site in subscriptions]
sec_to_distribute = [site for site in sites_allowed if not any([osite.startswith(site) for osite in sec_location])]
sec_to_distribute = [site for site in sec_to_distribute if not any([osite.startswith(site) for osite in sec_destination])]
if len( sec_to_distribute )>0:
for site in sec_to_distribute:
all_transfers[site].append( sec )
can_go = False
## is that possible to do something more
if can_go:
print wfo.name,"should just be assigned NOW to",sites_allowed
wfo.status = 'staged'
session.commit()
continue
else:
print wfo.name,"needs a transfer"
#print json.dumps(all_transfers)
fake_id=-1
wf_id_in_prestaging=set()
for (site,items_to_transfer) in all_transfers.iteritems():
items_to_transfer = list(set(items_to_transfer))
## convert to storage element
site_se = SI.CE_to_SE(site)
## massage a bit the items
blocks = [it for it in items_to_transfer if '#' in it]
datasets = [it for it in items_to_transfer if not '#' in it]
print "Making a replica to",site,"(CE)",site_se,"(SE) for"
print "\t",len(blocks),"blocks"
## remove blocks if full dataset is send out
blocks = [block for block in blocks if not block.split('#')[0] in datasets]
print "\t",len(blocks),"needed blocks"
print "\t",len(datasets),"datasets"
print "\t",datasets
items_to_transfer = blocks + datasets
## operate the transfer
if options and options.stop:
## ask to move-on
answer = raw_input('Continue with that ?')
if not answer.lower() in ['y','yes','go']:
continue
if execute:
result = makeReplicaRequest(url, site_se, items_to_transfer, 'prestaging')
else:
#result= {'phedex':{'request_created' : [{'id' : fake_id}]}}
result= {'phedex':{'request_created' : []}}
fake_id-=1
if not result:
print "ERROR Could not make a replica request for",site,items_to_transfer,"pre-staging"
continue
for phedexid in [o['id'] for o in result['phedex']['request_created']]:
new_transfer = session.query(Transfer).filter(Transfer.phedexid == phedexid).first()
if not new_transfer:
new_transfer = Transfer( phedexid = phedexid)
session.add( new_transfer )
new_transfer.workflows_id = set()
for transfering in list(set(map(lambda it : it.split('#')[0], items_to_transfer))):
new_transfer.workflows_id.update( workflow_dependencies[transfering] )
new_transfer.workflows_id = list(new_transfer.workflows_id)
wf_id_in_prestaging.update(new_transfer.workflows_id)
session.commit()
## auto approve it
if execute:
approved = approveSubscription(url, phedexid, [site_se])
for wfid in wf_id_in_prestaging:
tr_wf = session.query(Workflow).get(wfid)
if tr_wf and tr_wf.status!='staging':
if execute:
tr_wf.status = 'staging'
if talk:
print "setting",tr_wf.name,"to staging"
session.commit()
def transferor(url, specific=None, talk=True, options=None):
if userLock(): return
if duplicateLock(): return
use_mcm = True
up = componentInfo(mcm=use_mcm, soft=['mcm'])
if not up.check(): return
use_mcm = up.status['mcm']
if options and options.test:
execute = False
else:
execute = True
SI = siteInfo()
CI = campaignInfo()
NLI = newLockInfo()
mcm = McMClient(dev=False)
dss = DSS()
#allowed_secondary = UC.get('')
print "counting all being handled..."
being_handled = len(
session.query(Workflow).filter(Workflow.status == 'away').all())
being_handled += len(
session.query(Workflow).filter(
Workflow.status.startswith('stag')).all())
being_transfered = len(
session.query(Workflow).filter(Workflow.status == 'staging').all())
being_handled += len(
session.query(Workflow).filter(
Workflow.status.startswith('assistance-')).all())
max_to_handle = options.maxworkflows
max_to_transfer = options.maxstaging
allowed_to_handle = max(0, max_to_handle - being_handled)
allowed_to_transfer = max(0, max_to_transfer - being_transfered)
wf_buffer = 5
if allowed_to_handle <= wf_buffer: ## buffer for having several wf per transfer
print "Not allowed to run more than", max_to_handle, "at a time. Currently", being_handled, "and", wf_buffer, "buffer"
else:
print being_handled, "already being handled", max_to_handle, "max allowed,", allowed_to_handle, "remaining", "and", wf_buffer, "buffer"
if allowed_to_transfer <= wf_buffer:
print "Not allowed to transfer more than", max_to_transfer, "at a time. Currently", being_transfered, "and", wf_buffer, "buffer"
else:
print being_transfered, "already being transfered", max_to_transfer, "max allowed,", allowed_to_transfer, "remaining", "and", wf_buffer, "buffer"
print "... done"
all_transfers = defaultdict(list)
workflow_dependencies = defaultdict(
set) ## list of wf.id per input dataset
wfs_and_wfh = []
print "getting all wf to consider ..."
cache = getWorkflows(url, 'assignment-approved', details=True)
for wfo in session.query(Workflow).filter(
Workflow.status.startswith('considered')).all():
print "\t", wfo.name
if specific and not specific in wfo.name: continue
cache_r = filter(lambda d: d['RequestName'] == wfo.name, cache)
if len(cache_r):
wfs_and_wfh.append((wfo,
workflowInfo(url,
wfo.name,
spec=False,
request=cache_r[0])))
else:
wfs_and_wfh.append((wfo, workflowInfo(url, wfo.name, spec=False)))
print "... done"
transfers_per_sites = defaultdict(int)
input_sizes = {}
ignored_input_sizes = {}
input_cput = {}
input_st = {}
## list the size of those in transfer already
in_transfer_priority = None
min_transfer_priority = None
print "getting all wf in staging ..."
stucks = json.loads(open('%s/stuck_transfers.json' % monitor_dir).read())
for wfo in session.query(Workflow).filter(
Workflow.status == 'staging').all():
wfh = workflowInfo(url, wfo.name, spec=False)
#(lheinput,primary,parent,secondary) = wfh.getIO()
#sites_allowed = getSiteWhiteList( (lheinput,primary,parent,secondary) )
(lheinput, primary, parent, secondary,
sites_allowed) = wfh.getSiteWhiteList()
for site in sites_allowed: ## we should get the actual transfer destination instead of the full white list
transfers_per_sites[site] += 1
#input_cput[wfo.name] = wfh.getComputingTime()
#input_st[wfo.name] = wfh.getSystemTime()
for prim in primary:
ds_s = dss.get(prim)
if prim in stucks:
sendLog('transferor',
"%s appears stuck, so not counting it %s [GB]" %
(prim, ds_s),
wfi=wfh)
ignored_input_sizes[prim] = ds_s
else:
input_sizes[prim] = ds_s
sendLog('transferor',
"%s needs %s [GB]" % (wfo.name, ds_s),
wfi=wfh)
if in_transfer_priority == None:
in_transfer_priority = int(wfh.request['RequestPriority'])
else:
in_transfer_priority = max(in_transfer_priority,
int(wfh.request['RequestPriority']))
if min_transfer_priority == None:
min_transfer_priority = int(wfh.request['RequestPriority'])
else:
min_transfer_priority = min(min_transfer_priority,
int(wfh.request['RequestPriority']))
if min_transfer_priority == None or in_transfer_priority == None:
print "nothing is lining up for transfer"
sendEmail("no request in staging", "no request in staging")
return
pass
try:
print "Ignored input sizes"
ignored_values = list(ignored_input_sizes.items())
ignored_values.sort(key=lambda i: i[1])
print "\n".join(map(str, ignored_values))
print "Considered input sizes"
considered_values = list(input_sizes.items())
considered_values.sort(key=lambda i: i[1])
print "\n".join(map(str, considered_values))
except Exception as e:
print "trying to print the summary of input size"
print str(e)
print "... done"
print "Max priority in transfer already", in_transfer_priority
print "Min priority in transfer already", min_transfer_priority
print "transfers per sites"
print json.dumps(transfers_per_sites, indent=2)
in_transfer_already = sum(input_sizes.values())
cput_in_transfer_already = sum(input_cput.values())
st_in_transfer_already = sum(input_st.values())
## list the size of all inputs
primary_input_per_workflow_gb = defaultdict(float)
print "getting all input sizes ..."
for (wfo, wfh) in wfs_and_wfh:
(_, primary, _, _) = wfh.getIO()
#input_cput[wfo.name] = wfh.getComputingTime()
#input_st[wfo.name] = wfh.getSystemTime()
for prim in primary:
## do not count it if it appears stalled !
prim_size = dss.get(prim)
input_sizes[prim] = prim_size
primary_input_per_workflow_gb[wfo.name] += prim_size
print "... done"
# shuffle first by name
random.shuffle(wfs_and_wfh)
# Sort smallest transfers first; allows us to transfer as many as possible workflows.
def prio_and_size(i, j):
if int(i[1].request['RequestPriority']) == int(
j[1].request['RequestPriority']):
return cmp(int(primary_input_per_workflow_gb.get(j[0].name, 0)),
int(primary_input_per_workflow_gb.get(i[0].name, 0)))
else:
return cmp(int(i[1].request['RequestPriority']),
int(j[1].request['RequestPriority']))
#wfs_and_wfh.sort(cmp = prio_and_size, reverse=True)
#wfs_and_wfh.sort(cmp = lambda i,j : cmp(int(primary_input_per_workflow_gb.get(i[0].name, 0)), int(primary_input_per_workflow_gb.get(j[0].name, 0)) ))
#sort by priority higher first
wfs_and_wfh.sort(cmp=lambda i, j: cmp(int(i[1].request[
'RequestPriority']), int(j[1].request['RequestPriority'])),
reverse=True)
cput_grand_total = sum(input_cput.values())
cput_to_transfer = cput_grand_total - cput_in_transfer_already
st_grand_total = sum(input_st.values())
st_to_transfer = st_grand_total - st_in_transfer_already
print "%15.4f [CPU h] worth already in transfer" % cput_in_transfer_already
print "%15.4f [CPU h] worth is the current requested transfer load" % cput_to_transfer
print "%15.4f [h] worth of absolute system time in transfer" % (
cput_in_transfer_already / SI.availableSlots())
print "%15.4f [h] worth of absolute system time is the current requested transfer load" % (
cput_to_transfer / SI.availableSlots())
print "%15.4f [h] worth of theoritical system time in transfer" % (
st_in_transfer_already)
print "%15.4f [h] worth of theoritical system time is the current requested transfer load" % (
st_to_transfer)
grand_total = sum(input_sizes.values())
to_transfer = grand_total - in_transfer_already
grand_transfer_limit = options.maxtransfer
#grand_transfer_limit = SI.total_disk()*0.25*1024## half of the free sapce in TB->GB
transfer_limit = grand_transfer_limit - in_transfer_already
print "%15.4f GB already being transfered" % in_transfer_already
print "%15.4f GB is the current requested transfer load" % to_transfer
print "%15.4f GB is the global transfer limit" % grand_transfer_limit
print "%15.4f GB is the available limit" % transfer_limit
max_staging_per_site = options.maxstagingpersite
# the max priority value per dataset.
max_priority = defaultdict(int)
needs_transfer = 0 ## so that we can count'em
passing_along = 0
transfer_sizes = {}
went_over_budget = False
destination_cache = {}
no_goes = set()
max_per_round = UC.get('max_per_round').get('transferor', None)
if max_per_round and not spec:
wfs_and_wfh = wfs_and_wfh[:max_per_round]
for (wfo, wfh) in wfs_and_wfh:
print wfo.name, "to be transfered with priority", wfh.request[
'RequestPriority']
if wfh.request['RequestStatus'] != 'assignment-approved':
if wfh.request['RequestStatus'] in [
'aborted', 'rejected', 'rejected-archived',
'aborted-archived'
]:
wfo.status = 'trouble' ## so that we look or a replacement
else:
wfo.status = 'away'
wfh.sendLog(
'transferor', '%s in status %s, setting %s' %
(wfo.name, wfh.request['RequestStatus'], wfo.status))
continue
(_, primary, _, _) = wfh.getIO()
this_load = sum([input_sizes[prim] for prim in primary])
no_budget = False
if (this_load
and (sum(transfer_sizes.values()) + this_load > transfer_limit
or went_over_budget)):
if went_over_budget:
wfh.sendLog('transferor', "Transfer has gone over bubget.")
else:
wfh.sendLog('transferor', "Transfer will go over bubget.")
wfh.sendLog(
'transferor',
"%15.4f GB this load, %15.4f GB already this round, %15.4f GB is the available limit"
% (this_load, sum(transfer_sizes.values()), transfer_limit))
#if sum(transfer_sizes.values()) > transfer_limit:
went_over_budget = True
if in_transfer_priority != None and min_transfer_priority != None:
if int(
wfh.request['RequestPriority']
) >= in_transfer_priority and min_transfer_priority != in_transfer_priority:
wfh.sendLog(
'transferor',
"Higher priority sample %s >= %s go-on over budget" %
(wfh.request['RequestPriority'], in_transfer_priority))
else:
if not options.go:
wfh.sendLog(
'transferor',
"%s minimum priority %s < %s : stop" %
(min_transfer_priority,
wfh.request['RequestPriority'],
in_transfer_priority))
no_budget = True
## throtlle by campaign go
no_go = False
if not wfh.go(log=True) and not options.go:
no_go = True
no_goes.add(wfo.name)
allowed_secondary = set()
for campaign in wfh.getCampaigns():
if campaign in CI.campaigns and 'secondaries' in CI.campaigns[
campaign]:
allowed_secondary.update(CI.campaigns[campaign]['secondaries'])
if secondary:
if (secondary and allowed_secondary) and (
set(secondary) & allowed_secondary != set(secondary)):
wfh.sendLog(
'assignor', '%s is not an allowed secondary' %
(', '.join(set(secondary) - allowed_secondary)))
no_go = True
if no_go:
continue
## check if the batch is announced
def check_mcm(wfn):
announced = False
is_real = False
if not wfn.startswith('pdmvserv'):
is_real = True
try:
for b in mcm.getA('batches', query='contains=%s' % wfo.name):
is_real = True
if b['status'] == 'announced':
announced = True
break
except:
try:
for b in mcm.getA('batches',
query='contains=%s' % wfo.name):
is_real = True
if b['status'] == 'announced':
announced = True
break
except:
print "could not get mcm batch announcement, assuming not real"
return announced, is_real
if not use_mcm:
announced, is_real = False, True
else:
if wfh.request['RequestType'] in ['ReReco']:
announced, is_real = True, True
else:
announced, is_real = check_mcm(wfo.name)
if not announced:
wfh.sendLog('transferor', "does not look announced.")
if not is_real:
wfh.sendLog('transferor', "does not appear to be genuine.")
## prevent any duplication. if the wf is not mentioned in any batch, regardless of status
continue
## check on a grace period
injection_time = time.mktime(
time.strptime('.'.join(map(str, wfh.request['RequestDate'])),
"%Y.%m.%d.%H.%M.%S")) / (60. * 60.)
now = time.mktime(time.gmtime()) / (60. * 60.)
if float(now - injection_time) < 4.:
if not options.go and not announced:
wfh.sendLog(
'transferor',
"It is too soon to start transfer: %3.2fH remaining" %
(now - injection_time))
continue
if passing_along >= allowed_to_handle:
#if int(wfh.request['RequestPriority']) >= in_transfer_priority and min_transfer_priority!=in_transfer_priority:
if in_transfer_priority != None and min_transfer_priority != None:
if int(wfh.request['RequestPriority']
) >= in_transfer_priority and int(
wfh.request['RequestPriority']
) != min_transfer_priority:
## higher priority, and not only this priority being transfered
wfh.sendLog(
'transferor',
"Higher priority sample %s >= %s go-on over %s" %
(wfh.request['RequestPriority'], in_transfer_priority,
max_to_handle))
else:
wfh.sendLog(
'transferor',
" Not allowed to pass more than %s at a time. Currently %s handled, and adding %s"
% (max_to_handle, being_handled, passing_along))
if not options.go:
## should not allow to jump that fence
break
if this_load and needs_transfer >= allowed_to_transfer:
if in_transfer_priority != None and min_transfer_priority != None:
if int(wfh.request['RequestPriority']
) >= in_transfer_priority and int(
wfh.request['RequestPriority']
) != min_transfer_priority:
## higher priority, and not only this priority being transfered
wfh.sendLog(
'transferor',
"Higher priority sample %s >= %s go-on over %s" %
(wfh.request['RequestPriority'], in_transfer_priority,
max_to_transfer))
else:
wfh.sendLog(
'transferor',
"Not allowed to transfer more than %s at a time. Currently %s transfering, and adding %s"
% (max_to_transfer, being_transfered, needs_transfer))
if not options.go:
no_budget = True
if no_budget:
continue
## the site white list considers site, campaign, memory and core information
(lheinput, primary, parent, secondary,
sites_allowed) = wfh.getSiteWhiteList()
if options and options.tosites:
sites_allowed = options.tosites.split(',')
for dataset in list(primary) + list(parent) + list(secondary):
## lock everything flat
NLI.lock(dataset)
if not sites_allowed:
wfh.sendLog('transferor', "not possible site to run at")
#sendEmail("no possible sites","%s has no possible sites to run at"%( wfo.name ))
sendLog('transferor',
"%s has no possible sites to run at" % (wfo.name),
level='critical')
continue
blocks = []
if 'BlockWhitelist' in wfh.request and wfh.request['BlockWhitelist']:
blocks = wfh.request['BlockWhitelist']
if 'RunWhitelist' in wfh.request and wfh.request['RunWhitelist']:
## augment with run white list
for dataset in primary:
blocks = list(
set(blocks + getDatasetBlocks(
dataset, runs=wfh.request['RunWhitelist'])))
if 'LumiList' in wfh.request and wfh.request['LumiList']:
## augment with the lumi white list
blocks = list(
set(blocks +
getDatasetBlocks(dataset, lumis=wfh.request['LumiList'])))
if blocks:
print "Reading", len(blocks), "in block whitelist"
can_go = True
staging = False
allowed = True
primary_destinations = set()
if primary:
copies_needed_from_CPUh, CPUh = wfh.getNCopies()
if talk:
print wfo.name, 'reads', ', '.join(primary), 'in primary'
## chope the primary dataset
for prim in primary:
## keep track of what needs what
workflow_dependencies[prim].add(wfo.id)
max_priority[prim] = max(max_priority[prim],
int(wfh.request['RequestPriority']))
wfh.sendLog(
'transferor', "Would make %s from cpu requirement %s" %
(copies_needed_from_CPUh, CPUh))
copies_needed = copies_needed_from_CPUh
if 'Campaign' in wfh.request and wfh.request[
'Campaign'] in CI.campaigns and 'maxcopies' in CI.campaigns[
wfh.request['Campaign']]:
copies_needed_from_campaign = CI.campaigns[
wfh.request['Campaign']]['maxcopies']
copies_needed = min(copies_needed_from_campaign,
copies_needed)
wfh.sendLog(
'transferor',
"Maxed to %s by campaign configuration %s" %
(copies_needed, wfh.request['Campaign']))
### new ways of making the whole thing
destinations, all_block_names = getDatasetDestinations(
url,
prim,
within_sites=[SI.CE_to_SE(site) for site in sites_allowed],
only_blocks=blocks)
print json.dumps(destinations, indent=2)
## get where the dataset is in full and completed
prim_location = [
site for (site, info) in destinations.items()
if info['completion'] == 100 and info['data_fraction'] == 1
]
## the rest is places it is going to be
prim_destination = [
site for site in destinations.keys()
if not site in prim_location
]
if len(prim_location) >= copies_needed:
wfh.sendLog(
'transferor',
"The input is all fully in place at %s sites %s" %
(len(prim_location), sorted(prim_location)))
continue
copies_needed = max(0, copies_needed - len(prim_location))
wfh.sendLog(
'transferor',
"not counting existing copies ; now need %s" %
copies_needed)
copies_being_made = [
sum([
info['blocks'].keys().count(block)
for site, info in destinations.items()
if site in prim_destination
]) for block in all_block_names
]
latching_on_transfers = set()
[
latching_on_transfers.update(info['blocks'].values())
for site, info in destinations.items()
if site in prim_destination
]
latching_on_transfers = list(latching_on_transfers)
#print latching_on_transfers
## figure out where all this is going to go
prim_to_distribute = [
site for site in sites_allowed
if not SI.CE_to_SE(site) in prim_location
]
prim_to_distribute = [
site for site in prim_to_distribute
if not SI.CE_to_SE(site) in prim_destination
]
## take out the ones that cannot receive transfers
prim_to_distribute = [
site for site in prim_to_distribute if not any([
osite.startswith(site)
for osite in SI.sites_veto_transfer
])
]
wfh.sendLog(
'transferor',
"Could be going to: %s" % sorted(prim_to_distribute))
if not prim_to_distribute or any([
transfers_per_sites[site] < max_staging_per_site
for site in prim_to_distribute
]):
## means there is openings let me go
print "There are transfer slots available:", [
(site, transfers_per_sites[site])
for site in prim_to_distribute
]
#for site in sites_allowed:
# #increment accross the board, regardless of real destination: could be changed
# transfers_per_sites[site] += 1
else:
if int(
wfh.request['RequestPriority']
) >= in_transfer_priority and min_transfer_priority != in_transfer_priority:
wfh.sendLog(
'transferor',
"Higher priority sample %s >= %s go-on over transfer slots available"
% (wfh.request['RequestPriority'],
in_transfer_priority))
else:
wfh.sendLog(
'transferor',
"Not allowed to transfer more than %s per site at a time. Going overboard for %s"
% (max_staging_per_site,
sorted([
site for site in prim_to_distribute
if transfers_per_sites[site] >=
max_staging_per_site
])))
if not options.go:
allowed = False
break
for latching in latching_on_transfers:
tfo = session.query(Transfer).filter(
Transfer.phedexid == int(latching)).first()
if not tfo:
tfo = session.query(Transfer).filter(
Transfer.phedexid == -int(latching)).first()
if not tfo:
tfo = Transfer(phedexid=latching)
tfo.workflows_id = []
session.add(tfo)
else:
tfo.phedexid = latching ## make it positive ever
if not wfo.id in tfo.workflows_id:
print "adding", wfo.id, "to", tfo.id, "with phedexid", latching
l = copy.deepcopy(tfo.workflows_id)
l.append(wfo.id)
tfo.workflows_id = l
if not options.test:
session.commit()
else:
session.flush(
) ## regardless of commit later on, we need to let the next wf feeding on this transfer to see it in query
can_go = False
transfer_sizes[prim] = input_sizes[prim]
staging = True
# reduce the number of copies required by the on-going full transfer : how do we bootstrap on waiting for them ??
#copies_needed = max(0,copies_needed - len(prim_destination))
copies_needed = max(0, copies_needed - min(copies_being_made))
wfh.sendLog(
'transferor',
"Not counting the copies being made ; then need %s" %
copies_needed)
if copies_needed == 0:
wfh.sendLog(
'transferor',
"The output is either fully in place or getting in full somewhere with %s"
% latching_on_transfers)
can_go = True
continue
elif len(prim_to_distribute) == 0:
wfh.sendLog(
'transferor',
"We are going to need extra copies, but no destinations seems available"
)
prim_to_distribute = [
site for site in sites_allowed
if not SI.CE_to_SE(site) in prim_location
]
prim_to_distribute = [
site for site in prim_to_distribute if not any([
osite.startswith(site)
for osite in SI.sites_veto_transfer
])
]
if len(
prim_to_distribute
) > 0: ## maybe that a parameter we can play with to limit the
if not options or options.chop:
### hard include the tape disk andpoint ?
#tapes = [site for site in getDatasetPresence( url, prim, vetos=['T0','T2','T3','Disk']) if site.endswith('MSS')]
chops, sizes = getDatasetChops(
prim,
chop_threshold=options.chopsize,
only_blocks=blocks)
spreading = distributeToSites(chops,
prim_to_distribute,
n_copies=copies_needed,
weights=SI.cpu_pledges,
sizes=sizes)
transfer_sizes[prim] = sum(sizes)
if not spreading:
sendLog(
'transferor',
'cannot send %s to any site, it cannot fit anywhere'
% prim,
level='critical')
wfh.sendLog(
'transferor',
"cannot send to any site. %s cannot seem to fit anywhere"
% (prim))
staging = False
can_go = False
else:
spreading = {}
for site in prim_to_distribute:
if blocks:
spreading[site] = blocks
else:
spreading[site] = [prim]
transfer_sizes[prim] = input_sizes[
prim] ## this is approximate if blocks are specified
can_go = False
wfh.sendLog(
'transferor', "selected CE destinations %s" %
(sorted(spreading.keys())))
for (site, items) in spreading.items():
all_transfers[site].extend(items)
transfers_per_sites[site] += 1
primary_destinations.add(site)
if not allowed:
wfh.sendLog('transferor', "Not allowed to move on with")
continue
if secondary:
override_sec_destination = []
if 'SecondaryLocation' in CI.campaigns[wfh.request['Campaign']]:
override_sec_destination = CI.campaigns[
wfh.request['Campaign']]['SecondaryLocation']
print wfo.name, 'reads', ', '.join(secondary), 'in secondary'
for sec in secondary:
workflow_dependencies[sec].add(wfo.id)
if True:
## new style, failing on minbias
if not sec in destination_cache:
## this is barbbaric, and does not show the correct picture on workflow by workflow with different whitelist
destination_cache[sec], _ = getDatasetDestinations(
url, sec) ## NO SITE WHITE LIST ADDED
#destination_cache[sec],_ = getDatasetDestinations(url, sec, within_sites = [SI.CE_to_SE(site) for site in sites_allowed])
## limit to the site whitelist NOW
se_allowed = [SI.CE_to_SE(site) for site in sites_allowed]
destinations = dict([
(k, v) for (k, v) in destination_cache[sec].items()
if site in se_allowed
])
## truncate location/destination to those making up for >90% of the dataset
bad_destinations = [
destinations.pop(site)
for (site, info) in destinations.items()
if info['data_fraction'] < 0.9
]
sec_location = [
site for (site, info) in destinations.items()
if info['completion'] >= 95
]
sec_destination = [
site for site in destinations.keys()
if not site in sec_location
]
else:
## old style
presence = getDatasetPresence(url, sec)
sec_location = [
site for site, pres in presence.items()
if pres[1] > 90.
] ## more than 90% of the minbias at sites
subscriptions = listSubscriptions(url, sec)
sec_destination = [site for site in subscriptions]
sec_to_distribute = [
site for site in sites_allowed if
not any([osite.startswith(site) for osite in sec_location])
]
sec_to_distribute = [
site for site in sec_to_distribute if not any(
[osite.startswith(site) for osite in sec_destination])
]
sec_to_distribute = [
site for site in sec_to_distribute if not any([
osite.startswith(site)
for osite in SI.sites_veto_transfer
])
]
if override_sec_destination:
## intersect with where we want the PU to be
not_needed_anymore = list(
set(sec_to_distribute) - set(override_sec_destination))
#sendEmail("secondary superfluous","the dataset %s could be removed from %s"%( sec, not_needed_anymore ))
sendLog(
'transferor',
"the dataset %s could be removed from %s" %
(sec, not_needed_anymore))
sec_to_distribute = list(
set(sec_to_distribute) & set(override_sec_destination))
if len(sec_to_distribute) > 0:
print "secondary could go to", sorted(sec_to_distribute)
sec_size = dss.get(sec)
for site in sec_to_distribute:
site_se = SI.CE_to_SE(site)
if (SI.disk[site_se] * 1024.) > sec_size:
all_transfers[site].append(sec)
can_go = False
else:
print "could not send the secondary input to", site_se, "because it is too big for the available disk", SI.disk[
site_se] * 1024, "GB need", sec_size
if primary_destinations and site in primary_destinations:
#sendEmail('secondary input too big','%s is too big (%s) for %s (%s)'%( sec, sec_size, site_se, SI.disk[site_se]*1024))
sendLog('transferor',
'%s is too big (%s) for %s (%s)' %
(sec, sec_size, site_se,
SI.disk[site_se] * 1024),
level='critical')
else:
print "the secondary input does not have to be send to site"
## is that possible to do something more
if can_go:
## no explicit transfer required this time
if staging:
## but using existing ones
wfh.sendLog(
'transferor',
"latches on existing transfers, and nothing else, settin staging"
)
wfo.status = 'staging'
needs_transfer += 1
else:
wfh.sendLog(
'transferor', "should just be assigned now to %s" %
sorted(sites_allowed))
wfo.status = 'staged'
passing_along += 1
wfh.sendLog('transferor', "setting status to %s" % wfo.status)
session.commit()
continue
else:
## there is an explicit transfer required
if staging:
## and also using an existing one
wfh.sendLog('transferor', "latches on existing transfers")
if not options.test:
wfo.status = 'staging'
wfh.sendLog('transferor',
"setting status to %s" % wfo.status)
session.commit()
wfh.sendLog('transferor', "needs a transfer")
needs_transfer += 1
passing_along += 1
if no_goes:
#sendEmail("no go for managing","No go for \n"+"\n".join( no_goes ))
sendLog('transferor',
"No go for \n" + "\n".join(no_goes),
level='critical')
print "accumulated transfers"
print json.dumps(all_transfers, indent=2)
fake_id = -1
wf_id_in_prestaging = set()
for (site, items_to_transfer) in all_transfers.iteritems():
items_to_transfer = list(set(items_to_transfer))
## convert to storage element
site_se = SI.CE_to_SE(site)
## site that do not want input datasets
if site in SI.sites_veto_transfer:
print site, "does not want transfers"
continue
## throttle the transfer size to T2s ? we'd be screwed by a noPU sample properly configured.
## massage a bit the items
blocks = [it for it in items_to_transfer if '#' in it]
block_datasets = list(set([it.split('#')[0] for it in blocks]))
datasets = [it for it in items_to_transfer if not '#' in it]
details_text = "Making a replica to %s (CE) %s (SE) for" % (site,
site_se)
#print "\t",len(blocks),"blocks"
## remove blocks if full dataset is send out
blocks = [
block for block in blocks if not block.split('#')[0] in datasets
]
#print "\t",len(blocks),"needed blocks for",list(set([block.split('#')[0] for block in blocks]))
#print "\t",len(datasets),"datasets"
#print "\t",datasets
details_text += '\n\t%d blocks' % len(blocks)
details_text += '\n\t%d needed blocks for %s' % (
len(blocks),
sorted(list(set([block.split('#')[0] for block in blocks]))))
details_text += '\n\t%d datasets' % len(datasets)
details_text += '\n\t%s' % sorted(datasets)
items_to_transfer = blocks + datasets
if execute:
sendLog('transferor', details_text)
else:
print "Would make a replica to", site, "(CE)", site_se, "(SE) for"
print details_text
## operate the transfer
if options and options.stop:
## ask to move-on
answer = raw_input('Continue with that ?')
if not answer.lower() in ['y', 'yes', 'go']:
continue
if execute:
priority = 'normal'
cds = [
ds for ds in datasets + block_datasets if ds in max_priority
]
if cds and False: ## I don't think this is working. subscription should be updated on the fly and regularly for raising the priority if needed
## decide on an overall priority : that's a bit too large though
if any([max_priority[ds] >= 90000 for ds in cds]):
priority = 'high'
elif all([max_priority[ds] < 80000 for ds in cds]):
priority = 'low'
result = makeReplicaRequest(url,
site_se,
items_to_transfer,
'prestaging',
priority=priority)
else:
result = {'phedex': {'request_created': []}}
fake_id -= 1
if not result:
print "ERROR Could not make a replica request for", site, items_to_transfer, "pre-staging"
continue
for phedexid in [o['id'] for o in result['phedex']['request_created']]:
new_transfer = session.query(Transfer).filter(
Transfer.phedexid == int(phedexid)).first()
if not new_transfer:
new_transfer = session.query(Transfer).filter(
Transfer.phedexid == -int(phedexid)).first()
print phedexid, "transfer created"
if not new_transfer:
new_transfer = Transfer(phedexid=phedexid)
session.add(new_transfer)
else:
new_transfer.phedexid = phedexid ## make it positive again
new_transfer.workflows_id = set()
for transfering in list(
set(map(lambda it: it.split('#')[0], items_to_transfer))):
new_transfer.workflows_id.update(
workflow_dependencies[transfering])
new_transfer.workflows_id = list(new_transfer.workflows_id)
wf_id_in_prestaging.update(new_transfer.workflows_id)
session.commit()
## auto approve it
if execute:
approved = approveSubscription(url, phedexid, [site_se])
for wfid in wf_id_in_prestaging:
tr_wf = session.query(Workflow).get(wfid)
if tr_wf and tr_wf.status != 'staging':
if execute:
tr_wf.status = 'staging'
if talk:
print "setting", tr_wf.name, "to staging"
session.commit()
def close(self):
if os.path.isfile('.closor_stop'):
print "The closing of workflows is shortened"
return
url = self.url
batch_go = self.batch_go
CI = self.CI
UC = self.UC
wfo = self.wfo
jump_the_line = self.jump_the_line
batch_goodness = self.batch_goodness
check_fullcopy_to_announce = UC.get('check_fullcopy_to_announce')
## what is the expected #lumis
self.wfi = workflowInfo(url, wfo.name )
wfi = self.wfi
wfo.wm_status = wfi.request['RequestStatus']
if wfi.isRelval():
has_batch_go = False
batch_name = wfi.getCampaign()
if not batch_name in batch_go:
## do the esimatation whethere this can be announced : only once per batch
in_batches = getWorkflowByCampaign(url , batch_name, details=True)
batch_go[ batch_name ] = all(map(lambda s : not s in ['completed','running-open','running-closed','acquired','assigned','assignment-approved'], [r['RequestStatus'] for r in in_batches]))
## already verified
has_batch_go = batch_go[batch_name]
if not has_batch_go:
wfi.sendLog('closor', 'Cannot close for now because the batch <a href=https://dmytro.web.cern.ch/dmytro/cmsprodmon/workflows.php?campaign=%s>%s</a> is not all close'%( batch_name, batch_name))
return
if wfi.request['RequestStatus'] in ['announced','normal-archived'] and not options.force:
## manually announced ??
self.to_status = 'done'
self.to_wm_status = wfi.request['RequestStatus']
wfi.sendLog('closor','%s is announced already : %s'%( wfo.name,self.to_wm_status))
return
if jump_the_line:
wfi.sendLog('closor','Announcing while completing')
expected_lumis = 1
if not 'TotalInputLumis' in wfi.request:
print wfo.name,"has not been assigned yet, or the database is corrupted"
elif wfi.request['TotalInputLumis']==0:
print wfo.name,"is corrupted with 0 expected lumis"
else:
expected_lumis = wfi.request['TotalInputLumis']
## what are the outputs
outputs = wfi.request['OutputDatasets']
## check whether the number of lumis is as expected for each
all_OK = defaultdict(lambda : False)
stats = defaultdict(int)
#print outputs
if len(outputs):
print wfo.name,wfi.request['RequestStatus']
for out in outputs:
event_count,lumi_count = getDatasetEventsAndLumis(dataset=out)
self.outs.append( Output( datasetname = out ))
odb = self.outs[-1]
odb.workflow = wfo
odb.nlumis = lumi_count
odb.nevents = event_count
odb.workfow_id = wfo.id
if odb.expectedlumis < expected_lumis:
odb.expectedlumis = expected_lumis
else:
expected_lumis = odb.expectedlumis
odb.date = time.mktime(time.gmtime())
fraction = lumi_count/float(expected_lumis)*100.
completion_line = "%60s %d/%d = %3.2f%%"%(out,lumi_count,expected_lumis,fraction)
wfi.sendLog('closor',"\t%s"% completion_line)
if wfi.isRelval() and fraction < batch_goodness:
self.batch_warnings[ wfi.getCampaign()].add( completion_line )
stats[out] = lumi_count
all_OK[out] = True
## check for at least one full copy prior to moving on
in_full = {}
for out in outputs:
in_full[out] = []
presence = getDatasetPresence( url, out )
where = [site for site,info in presence.items() if info[0]]
if where:
all_OK[out] = True
print out,"is in full at",",".join(where)
in_full[out] = copy.deepcopy(where)
else:
going_to = wfi.request['NonCustodialSites']+wfi.request['CustodialSites']
wfi.sendLog('closor',"%s is not in full anywhere. send to %s"%(out, ",".join(sorted(going_to))))
at_destination = dict([(k,v) for (k,v) in presence.items() if k in going_to])
else_where = dict([(k,v) for (k,v) in presence.items() if not k in going_to])
print json.dumps( at_destination )
print json.dumps( else_where, indent=2 )
## do the full stuck transfer study, missing files and shit !
for there in going_to:
late_info = findLateFiles(url, out, going_to = there )
for l in late_info:
l.update({"workflow":wfo.name,"dataset":out})
self.all_late_files.extend( late_info )
if check_fullcopy_to_announce:
## only set this false if the check is relevant
all_OK[out] = False
## verify if we have to do harvesting
if not options.no_harvest and not jump_the_line:
(OK, requests) = spawn_harvesting(url, wfi, in_full)
all_OK.update( OK )
## only that status can let me go into announced
if all(all_OK.values()) and ((wfi.request['RequestStatus'] in ['closed-out']) or options.force or jump_the_line):
print wfo.name,"to be announced"
results=[]
if not results:
for out in outputs:
print "dealing with",out
if out in stats and not stats[out]:
continue
_,dsn,process_string,tier = out.split('/')
if all_OK[out]:
print "setting valid"
results.append(setDatasetStatus(out, 'VALID', withFiles=False))
if all_OK[out] and wfi.isRelval():
## make the specific relval rules and the replicas
## figure the destination(s) out
destinations = set()
if tier != "RECO" and tier != "ALCARECO":
destinations.add('T2_CH_CERN')
if tier == "GEN-SIM":
destinations.add('T1_US_FNAL_Disk')
if tier == "GEN-SIM-DIGI-RAW":
destinations.add('T1_US_FNAL_Disk')
if tier == "GEN-SIM-RECO":
destinations.add('T1_US_FNAL_Disk')
if "RelValTTBar" in dsn and "TkAlMinBias" in process_string and tier != "ALCARECO":
destinations.add('T2_CH_CERN')
if "MinimumBias" in dsn and "SiStripCalMinBias" in process_string and tier != "ALCARECO":
destinations.add('T2_CH_CERN')
if destinations:
wfi.sendLog('closor', '%s to go to %s'%(out, ', '.join( sorted( destinations ))))
## call to makereplicarequest under relval => done
for site in destinations:
result = makeReplicaRequest(url, site, [out], 'Copy for release validation consumption', priority='normal', approve=True, mail=False, group='RelVal')
try:
request_id = result['phedex']['request_created'][0]['id']
results.append( True )
except:
results.append( 'Failed relval transfer' )
elif all_OK[out]:
campaign = None
try:
campaign = out.split('/')[2].split('-')[0]
except:
if 'Campaign' in wfi.request and wfi.request['Campaign']:
campaign = wfi.request['Campaign']
to_DDM = False
## campaign override
if campaign and campaign in CI.campaigns and 'toDDM' in CI.campaigns[campaign] and tier in CI.campaigns[campaign]['toDDM']:
to_DDM = True
## by typical enabling
if tier in UC.get("tiers_to_DDM"):
to_DDM = True
## check for unitarity
if not tier in UC.get("tiers_no_DDM")+UC.get("tiers_to_DDM"):
print "tier",tier,"neither TO or NO DDM for",out
results.append('Not recognitized tier %s'%tier)
#sendEmail("failed DDM injection","could not recognize %s for injecting in DDM"% out)
sendLog('closor', "could not recognize %s for injecting in DDM"% out, level='critical')
continue
n_copies = 1
destinations=[]
if to_DDM and campaign and campaign in CI.campaigns and 'DDMcopies' in CI.campaigns[campaign]:
ddm_instructions = CI.campaigns[campaign]['DDMcopies']
if type(ddm_instructions) == int:
n_copies = CI.campaigns[campaign]['DDMcopies']
elif type(ddm_instructions) == dict:
## a more fancy configuration
for ddmtier,indication in ddm_instructions.items():
if ddmtier==tier or ddmtier in ['*','all']:
## this is for us
if 'N' in indication:
n_copies = indication['N']
if 'host' in indication:
destinations = indication['host']
destination_spec = ""
if destinations:
destination_spec = "--destination="+",".join( destinations )
group_spec = "" ## not used yet
### should make this a campaign configuration
## inject to DDM when necessary
if to_DDM:
print "Sending",out," to DDM"
status = pass_to_dynamo( [out], N = n_copies, sites=destinations if destinations else None, group = group_spec if group_spec else None)
results.append( status )
if status == True:
wfi.sendLog('closor','%s is send to dynamo in %s copies %s %s'%( out, n_copies, sorted(destinations), group_spec))
else:
sendLog('closor',"could not add "+out+" to dynamo pool. check closor logs.", level='critical')
wfi.sendLog('closor',"could not add "+out+" to dynamo pool. check closor logs.")
else:
print wfo.name,"no stats for announcing",out
results.append('No Stats')
if all(map(lambda result : result in ['None',None,True],results)):
if not jump_the_line:
## only announce if all previous are fine
res = reqMgrClient.announceWorkflowCascade(url, wfo.name)
if not res in ['None',None]:
## check the status again, it might well have toggled
wl_bis = workflowInfo(url, wfo.name)
self.to_wm_status = wl_bis.request['RequestStatus']
if wl_bis.request['RequestStatus'] in ['announced','normal-archived']:
res = None
else:
res = reqMgrClient.announceWorkflowCascade(url, wfo.name)
results.append( res )
print results
if all(map(lambda result : result in ['None',None,True],results)):
if jump_the_line:
if not 'announced' in wfo.status:
self.to_status = wfo.status.replace('announce','announced')
else:
self.to_status = 'done'
self.closing = True
wfi.sendLog('closor',"workflow outputs are announced")
else:
wfi.sendLog('closor',"Error with %s to be announced \n%s"%( wfo.name, json.dumps( results )))
elif wfi.request['RequestStatus'] in ['failed','aborted','aborted-archived','rejected','rejected-archived','aborted-completed']:
if wfi.isRelval():
self.to_status = 'forget'
self.to_wm_status = wfi.request['RequestStatus']
wfi.sendLog('closor',"%s is %s, but will not be set in trouble to find a replacement."%( wfo.name, self.to_wm_status))
else:
self.to_status = 'trouble'
self.to_wm_status = wfi.request['RequestStatus']
else:
print wfo.name,"not good for announcing:",wfi.request['RequestStatus']
wfi.sendLog('closor',"cannot be announced")
self.held.add( wfo.name )
def stagor(url, specific=None, options=None):
if not componentInfo().check(): return
SI = siteInfo()
CI = campaignInfo()
UC = unifiedConfiguration()
TS = transferStatuses()
cached_transfer_statuses = TS.content()
transfer_statuses = {}
done_by_wf_id = {}
done_by_input = {}
completion_by_input = {}
good_enough = 100.0
lost_blocks = json.loads(
eosRead('%s/lost_blocks_datasets.json' % monitor_dir))
lost_files = json.loads(
eosRead('%s/lost_files_datasets.json' % monitor_dir))
known_lost_blocks = {}
known_lost_files = {}
for dataset in set(lost_blocks.keys() + lost_files.keys()):
b, f = findLostBlocksFiles(url, dataset)
if dataset in lost_blocks and not b:
print dataset, "has no really lost blocks"
else:
known_lost_blocks[dataset] = [i['name'] for i in b]
if dataset in lost_files and not f:
print dataset, "has no really lost files"
else:
known_lost_files[dataset] = [i['name'] for i in f]
def time_point(label="", sub_lap=False):
now = time.mktime(time.gmtime())
nows = time.asctime(time.gmtime())
print "Time check (%s) point at : %s" % (label, nows)
print "Since start: %s [s]" % (now - time_point.start)
if sub_lap:
print "Sub Lap : %s [s]" % (now - time_point.sub_lap)
time_point.sub_lap = now
else:
print "Lap : %s [s]" % (now - time_point.lap)
time_point.lap = now
time_point.sub_lap = now
time_point.sub_lap = time_point.lap = time_point.start = time.mktime(
time.gmtime())
time_point("Check cached transfer")
## collect all datasets that are needed for wf in staging, correcting the status of those that are not really in staging
wfois = []
needs = defaultdict(list)
needs_by_priority = defaultdict(list)
for wfo in session.query(Workflow).filter(
Workflow.status == 'staging').all():
wfi = workflowInfo(url, wfo.name)
if wfi.request['RequestStatus'] in [
'running-open', 'running-closed', 'completed', 'assigned',
'acquired'
]:
wfi.sendLog('stagor',
"is in status %s" % wfi.request['RequestStatus'])
wfo.status = 'away'
session.commit()
continue
if not wfi.request['RequestStatus'] in ['assignment-approved']:
## should be setting 'away' too
## that usually happens for relvals
if wfi.request['RequestStatus'] in [
'rejected', 'aborted', 'aborted-completed',
'aborted-archived', 'rejected-archived'
] and wfi.isRelval():
wfo.status = 'forget'
session.commit()
continue
else:
print wfo.name, "is", wfi.request['RequestStatus']
#sendEmail("wrong status in staging. debug","%s is in %s, should set away."%(wfo.name,wfi.request['RequestStatus']))
sendLog("stagor",
"%s is in %s, set away" %
(wfo.name, wfi.request['RequestStatus']),
level='critical')
wfo.status = 'away'
session.commit()
continue
wfois.append((wfo, wfi))
_, primaries, _, secondaries = wfi.getIO()
for dataset in list(primaries) + list(secondaries):
needs[wfo.name].append(dataset)
done_by_input[dataset] = {}
completion_by_input[dataset] = {}
needs_by_priority[wfi.request['RequestPriority']].append(dataset)
wfi.sendLog('stagor', '%s needs %s' % (wfo.name, dataset))
time_point("Check staging workflows")
open('%s/dataset_requirements.json' % monitor_dir,
'w').write(json.dumps(needs, indent=2))
for prio in needs_by_priority:
needs_by_priority[prio] = list(set(needs_by_priority[prio]))
open('%s/dataset_priorities.json' % monitor_dir,
'w').write(json.dumps(needs_by_priority, indent=2))
dataset_endpoints = defaultdict(set)
endpoint_in_downtime = defaultdict(set)
#endpoint_completed = defaultdict(set)
endpoint_incompleted = defaultdict(set)
#endpoint = defaultdict(set)
send_back_to_considered = set()
## first check if anything is inactive
all_actives = set([
transfer.phedexid for transfer in session.query(TransferImp).filter(
TransferImp.active).all()
])
for active_phedexid in all_actives:
skip = True
transfers_phedexid = session.query(TransferImp).filter(
TransferImp.phedexid == active_phedexid).all()
for imp in transfers_phedexid:
if imp.workflow.status == 'staging':
skip = False
sendLog(
'stagor', "\t%s is staging for %s" %
(imp.phedexid, imp.workflow.name))
if skip:
sendLog('stagor', "setting %s inactive" % active_phedexid)
for imp in transfers_phedexid:
imp.active = False
session.commit()
all_actives = sorted(
set([
transfer.phedexid for transfer in session.query(
TransferImp).filter(TransferImp.active).all()
]))
for phedexid in all_actives:
if specific: continue
## check on transfer completion
not_cached = False
if phedexid in cached_transfer_statuses:
### use a cache for transfer that already looked done
sendLog('stagor', "read %s from cache" % phedexid)
checks = cached_transfer_statuses[phedexid]
else:
## I actually would like to avoid that all I can
sendLog('stagor',
'Performing spurious transfer check on %s' % phedexid,
level='critical')
checks = checkTransferStatus(url, phedexid, nocollapse=True)
try:
print json.dumps(checks, indent=2)
except:
print checks
if not checks:
## this is going to bias quite heavily the rest of the code. we should abort here
#sendLog('stagor','Ending stagor because of skewed input from checkTransferStatus', level='critical')
#return False
sendLog(
'stagor',
'Stagor has got a skewed input from checkTransferStatus',
level='critical')
checks = {}
pass
else:
TS.add(phedexid, checks)
time_point("Check transfer status %s" % phedexid, sub_lap=True)
if not specific:
for dsname in checks:
if not dsname in done_by_input: done_by_input[dsname] = {}
if not dsname in completion_by_input:
completion_by_input[dsname] = {}
done_by_input[dsname][phedexid] = all(
map(lambda i: i >= good_enough, checks[dsname].values()))
completion_by_input[dsname][phedexid] = checks[dsname].values()
if checks:
sendLog(
'stagor', "Checks for %s are %s" %
(phedexid, [node.values() for node in checks.values()]))
done = all(
map(
lambda i: i >= good_enough,
list(
itertools.chain.from_iterable(
[node.values() for node in checks.values()]))))
else:
## it is empty, is that a sign that all is done and away ?
if not_cached:
print "Transfer status was not cached"
else:
print "ERROR with the scubscriptions API of ", phedexid
print "Most likely something else is overiding the transfer request. Need to work on finding the replacement automatically, if the replacement exists"
done = False
transfers_phedexid = session.query(TransferImp).filter(
TransferImp.phedexid == phedexid).all()
for imp in transfers_phedexid:
tr_wf = imp.workflow
if tr_wf: # and tr_wf.status == 'staging':
if not tr_wf.id in done_by_wf_id: done_by_wf_id[tr_wf.id] = {}
done_by_wf_id[tr_wf.id][phedexid] = done
if done:
imp.active = False
session.commit()
for ds in checks:
for s, v in checks[ds].items():
dataset_endpoints[ds].add(s)
if done:
sendLog('stagor', "%s is done" % phedexid)
TS.add(phedexid, checks)
else:
sendLog(
'stagor',
"%s is not finished %s" % (phedexid, pprint.pformat(checks)))
##pprint.pprint( checks )
## check if the destination is in down-time
for ds in checks:
sites_incomplete = [
SI.SE_to_CE(s) for s, v in checks[ds].items()
if v < good_enough
]
sites_incomplete_down = [
s for s in sites_incomplete if not s in SI.sites_ready
]
## no space means no transfer should go there : NO, it does not work in the long run
#sites_incomplete_down = [SI.SE_to_CE(s) for s,v in checks[ds].items() if (v<good_enough and (SI.disk[s]==0 or (not SI.SE_to_CE(s) in SI.sites_ready)))]
if sites_incomplete_down:
sendLog(
'stagor',
"%s are in downtime, while waiting for %s to get there"
% (",".join(sites_incomplete_down), ds))
endpoint_in_downtime[ds].update(sites_incomplete_down)
if sites_incomplete:
endpoint_incompleted[ds].update(sites_incomplete)
time_point("Check on-going transfers")
print "End points"
for k in dataset_endpoints:
dataset_endpoints[k] = list(dataset_endpoints[k])
print json.dumps(dataset_endpoints, indent=2)
print "End point in down time"
for k in endpoint_in_downtime:
endpoint_in_downtime[k] = list(endpoint_in_downtime[k])
print json.dumps(endpoint_in_downtime, indent=2)
print "End point incomplete in down time"
for k in endpoint_incompleted:
endpoint_incompleted[k] = list(endpoint_incompleted[k])
print json.dumps(endpoint_incompleted, indent=2)
#open('%s/transfer_statuses.json'%monitor_dir,'w').write( json.dumps( transfer_statuses, indent=2))
eosFile('%s/transfer_statuses.json' % monitor_dir,
'w').write(json.dumps(TS.content(), indent=2)).close()
eosFile('%s/dataset_endpoints.json' % monitor_dir,
'w').write(json.dumps(dataset_endpoints, indent=2)).close()
already_stuck = json.loads(
eosRead('%s/stuck_transfers.json' % monitor_pub_dir)).keys()
already_stuck.extend(getAllStuckDataset())
missing_in_action = defaultdict(list)
print "-" * 10, "Checking on workflows in staging", "-" * 10
#forget_about = ['/MinBias_TuneCUETP8M1_13TeV-pythia8/RunIISummer15GS-MCRUN2_71_V1-v2/GEN-SIM']
#for what in forget_about:
# if not done_by_input[what]:
# done_by_input[what] = {'fake':True}
## come back to workflows and check if they can go
available_cache = defaultdict(lambda: defaultdict(float))
presence_cache = defaultdict(dict)
time_point("Preparing for more")
for wfo, wfi in wfois:
print "#" * 30
time_point("Forward checking %s" % wfo.name, sub_lap=True)
## the site white list takes site, campaign, memory and core information
(_, primaries, _, secondaries,
sites_allowed) = wfi.getSiteWhiteList(verbose=False)
se_allowed = [SI.CE_to_SE(site) for site in sites_allowed]
se_allowed.sort()
se_allowed_key = ','.join(se_allowed)
readys = {}
for need in list(primaries) + list(secondaries):
if not need in done_by_input:
wfi.sendLog('stagor', "missing transfer report for %s" % need)
readys[need] = False
## should warn someone about this !!!
## it cannot happen, by construction
sendEmail('missing transfer report',
'%s does not have a transfer report' % (need))
continue
if not done_by_input[need] and need in list(secondaries):
wfi.sendLog(
'stagor',
"assuming it is OK for secondary %s to have no attached transfers"
% need)
readys[need] = True
done_by_input[need] = {"fake": True}
continue
if len(done_by_input[need]) and all(done_by_input[need].values()):
wfi.sendLog('stagor', "%s is ready" % need)
print json.dumps(done_by_input[need], indent=2)
readys[need] = True
else:
wfi.sendLog(
'stagor', "%s is not ready \n%s" %
(need, json.dumps(done_by_input[need], indent=2)))
readys[need] = False
if readys and all(readys.values()):
if wfo.status == 'staging':
wfi.sendLog('stagor',
"all needs are fullfilled, setting staged")
wfo.status = 'staged'
session.commit()
else:
wfi.sendLog('stagor', "all needs are fullfilled, already")
print json.dumps(readys, indent=2)
else:
wfi.sendLog('stagor', "missing requirements")
copies_needed, _ = wfi.getNCopies()
jump_ahead = False
re_transfer = False
## there is missing input let's do something more elaborated
for need in list(primaries): #+list(secondaries):
if endpoint_in_downtime[need] and endpoint_in_downtime[
need] == endpoint_incompleted[need]:
#print need,"is going to an end point in downtime"
wfi.sendLog(
'stagor',
"%s has only incomplete endpoint in downtime\n%s" %
(need, endpoint_in_downtime[need]))
re_transfer = True
if not se_allowed_key in available_cache[need]:
available_cache[need][
se_allowed_key] = getDatasetBlocksFraction(
url, need, sites=se_allowed)
if available_cache[need][se_allowed_key] >= copies_needed:
wfi.sendLog(
'stagor',
"assuming it is OK to move on like this already for %s"
% need)
jump_ahead = True
else:
wfi.sendLog(
'stagor', "Available %s times" %
available_cache[need][se_allowed_key])
missing_and_downtime = list(
set(endpoint_in_downtime[need])
& set(endpoint_incompleted[need]))
if missing_and_downtime:
wfi.sendLog(
'stagor',
"%s is incomplete at %s which is in downtime, trying to move along"
% (need, ','.join(missing_and_downtime)))
jump_ahead = True
else:
wfi.sendLog(
'stagor',
"continue waiting for transfers for optimum production performance."
)
## compute a time since staging to filter jump starting ?
# check whether the inputs is already in the stuck list ...
for need in list(primaries) + list(secondaries):
if need in already_stuck:
wfi.sendLog('stagor',
"%s is stuck, so try to jump ahead" % need)
jump_ahead = True
if jump_ahead or re_transfer:
details_text = "checking on availability for %s to jump ahead" % wfo.name
details_text += '\n%s wants %s copies' % (wfo.name,
copies_needed)
copies_needed = max(1, copies_needed - 1)
details_text += '\nlowering by one unit to %s' % copies_needed
wfi.sendLog('stagor', details_text)
all_check = True
prim_where = set()
for need in list(primaries):
if not se_allowed_key in presence_cache[need]:
presence_cache[need][
se_allowed_key] = getDatasetPresence(
url, need, within_sites=se_allowed)
presence = presence_cache[need][se_allowed_key]
prim_where.update(presence.keys())
available = available_cache[need][se_allowed_key]
this_check = (available >= copies_needed)
wfi.sendLog(
'stagor', "%s is available %s times (%s), at %s" %
(need, available, this_check, se_allowed_key))
all_check &= this_check
if not all_check: break
for need in list(secondaries):
## I do not want to check on the secon
## this below does not function because the primary could be all available, and the secondary not complete at a certain site that does not matter at that point
this_check = all(done_by_input[need].values())
wfi.sendLog(
'stagor', "%s is this much transfered %s" %
(need, json.dumps(done_by_input[need], indent=2)))
all_check &= this_check
#if not se_allowed_key in presence_cache[need]:
# presence_cache[need][se_allowed_key] = getDatasetPresence( url, need , within_sites=se_allowed)
## restrict to where the primary is
#presence = dict([(k,v) for (k,v) in presence_cache[need][se_allowed_key].items() if k in prim_where])
#this_check = all([there for (there,frac) in presence.values()])
#print need,"is present at all sites:",this_check
#all_check&= this_check
if all_check and not re_transfer:
wfi.sendLog(
'stagor',
"needs are sufficiently fullfilled, setting staged")
wfo.status = 'staged'
session.commit()
else:
print wfo.name, "has to wait a bit more"
wfi.sendLog('stagor', "needs to wait a bit more")
else:
wfi.sendLog('stagor', "not checking availability")
if re_transfer:
wfi.sendLog(
'stagor',
"Sending back to considered because of endpoint in downtime"
)
if wfo.status == 'staging':
wfo.status = 'considered'
session.commit()
send_back_to_considered.add(wfo.name)
time_point("Checked affected workflows")
if send_back_to_considered:
#sendEmail("transfer to endpoint in downtime","sending back to considered the following workflows \n%s"%('\n'.join( send_back_to_considered)))
sendLog('stagor',
"sending back to considered the following workflows \n%s" %
('\n'.join(send_back_to_considered)),
level='critical')
print "-" * 10, "Checking on non-available datasets", "-" * 10
## now check on those that are not fully available
for dsname in available_cache.keys():
## squash the se_allowed_key key
available_cache[dsname] = min(available_cache[dsname].values())
really_stuck_dataset = set()
for dsname, available in available_cache.items():
using_its = getWorkflowByInput(url, dsname)
#print using_its
using_wfos = []
for using_it in using_its:
wf = session.query(Workflow).filter(
Workflow.name == using_it).first()
if wf:
using_wfos.append(wf)
if not len(done_by_input[dsname]):
print "For dataset", dsname, "there are no transfer report. That's an issue."
for wf in using_wfos:
if wf.status == 'staging':
if UC.get("stagor_sends_back"):
print "sending", wf.name, "back to considered"
wf.status = 'considered'
session.commit()
#sendEmail( "send back to considered","%s was send back and might be trouble"% wf.name)
sendLog('stagor',
"%s was send back and might be trouble" %
wf.name,
level='critical')
else:
print "would send", wf.name, "back to considered"
#sendEmail( "subscription lagging behind","susbscriptions to get %s running are not appearing in phedex. I would have send it back to considered but that's not good."% wf.name)
sendLog(
'stagor',
"susbscriptions to get %s running are not appearing in phedex. I would have send it back to considered but that's not good."
% wf.name,
level='critical')
continue
## not compatible with checking on secondary availability
#if all([wf.status != 'staging' for wf in using_wfos]):
# ## means despite all checks that input is not needed
# continue
if available < 1.:
print "incomplete", dsname
## there is a problem in the method below that it does not account for files stuck in T1*Buffer only
lost_blocks, lost_files = findLostBlocksFiles(
url, dsname) if (not dsname.endswith('/RAW')) else ([], [])
lost_block_names = [item['name'] for item in lost_blocks]
lost_file_names = [item['name'] for item in lost_files]
if lost_blocks:
#print json.dumps( lost , indent=2 )
## estimate for how much !
fraction_loss, _, n_missing = getDatasetBlockFraction(
dsname, lost_block_names)
print "We have lost", len(
lost_block_names
), "blocks", lost_block_names, "for %f%%" % (100. *
fraction_loss)
if fraction_loss > 0.05: ## 95% completion mark
#sendEmail('we have lost too many blocks','%s is missing %d blocks, for %d events, %f %% loss'%(dsname, len(lost_block_names), n_missing, fraction_loss))
sendLog(
'stagor',
'%s is missing %d blocks, for %d events, %3.2f %% loss'
% (dsname, len(lost_block_names), n_missing,
100 * fraction_loss),
level='critical')
## the workflow should be rejected !
for wf in using_wfos:
if wf.status == 'staging':
print wf.name, "is doomed. setting to trouble"
wf.status = 'trouble'
session.commit()
sendLog(
'stagor',
'%s has too much loss on the input dataset %s. Missing %d blocks, for %d events, %3.2f %% loss'
% (wf.name, dsname, len(lost_block_names),
n_missing, 100 * fraction_loss),
level='critical')
else:
## probably enough to make a ggus and remove
if not dsname in known_lost_blocks:
#sendEmail('we have lost a few blocks', '%s is missing %d blocks, for %d events, %f %% loss\n\n%s'%(dsname, len(lost_block_names), n_missing, fraction_loss, '\n'.join( lost_block_names ) ))
sendLog(
'stagor',
'%s is missing %d blocks, for %d events, %f %% loss\n\n%s'
% (dsname, len(lost_block_names), n_missing,
fraction_loss, '\n'.join(lost_block_names)),
level='critical')
known_lost_blocks[dsname] = [
i['name'] for i in lost_blocks
]
really_stuck_dataset.add(dsname)
if lost_files:
fraction_loss, _, n_missing = getDatasetFileFraction(
dsname, lost_file_names)
print "We have lost", len(
lost_file_names
), "files", lost_file_names, "for %f%%" % fraction_loss
if fraction_loss > 0.05:
#sendEmail('we have lost too many files','%s is missing %d files, for %d events, %f %% loss'%(dsname, len(lost_file_names),n_missing, fraction_loss))
sendLog(
'stagor',
'%s is missing %d files, for %d events, %f %% loss' %
(dsname, len(lost_file_names), n_missing,
fraction_loss),
level='critical')
for wf in using_wfos:
if wf.status == 'staging':
print wf.name, "is doomed. setting to trouble"
wf.status = 'trouble'
session.commit()
else:
## probably enough to make a ggus and remove
if not dsname in known_lost_files:
#sendEmail('we have lost a few files','%s is missing %d files, for %d events, %f %% loss\n\n%s'%(dsname, len(lost_file_names),n_missing, fraction_loss, '\n'.join(lost_file_names)))
sendLog(
'stagor',
'%s is missing %d files, for %d events, %f %% loss\n\n%s'
% (dsname, len(lost_file_names), n_missing,
fraction_loss, '\n'.join(lost_file_names)),
level='critical')
known_lost_files[dsname] = [
i['name'] for i in lost_files
]
## should the status be change to held-staging and pending on a ticket
missings = [
pid for (pid, d) in done_by_input[dsname].items() if d == False
]
print "\t", done_by_input[dsname]
print "\tneeds", len(done_by_input[dsname])
print "\tgot", done_by_input[dsname].values().count(True)
print "\tmissing", missings
missing_in_action[dsname].extend(missings)
rr = eosFile('%s/lost_blocks_datasets.json' % monitor_dir, 'w')
rr.write(json.dumps(known_lost_blocks, indent=2))
rr.close()
rr = eosFile('%s/lost_files_datasets.json' % monitor_dir, 'w')
rr.write(json.dumps(known_lost_files, indent=2))
rr.close()
eosFile('%s/incomplete_transfers.json' % monitor_dir,
'w').write(json.dumps(missing_in_action, indent=2)).close()
print "Stuck transfers and datasets"
print json.dumps(missing_in_action, indent=2)
TD = transferDataset()
datasets_by_phid = defaultdict(set)
for dataset in missing_in_action:
for phid in missing_in_action[dataset]:
#print dataset,"stuck through",phid
datasets_by_phid[phid].add(dataset)
for k in datasets_by_phid:
#datasets_by_phid[k] = list(datasets_by_phid[k])
TD.add(k, list(datasets_by_phid[k]))
#eosFile('%s/datasets_by_phid.json'%base_eos_dir,'w').write( json.dumps(datasets_by_phid, indent=2 )).close()
eosFile('%s/really_stuck_dataset.json' % base_eos_dir,
'w').write(json.dumps(list(really_stuck_dataset),
indent=2)).close()
print '\n' * 2, "Datasets really stuck"
print '\n'.join(really_stuck_dataset)
#############
## not going further for what matters
#############
return
def closor(url, specific=None, options=None):
if userLock(): return
if duplicateLock(): return
if not componentInfo().check(): return
UC = unifiedConfiguration()
CI = campaignInfo()
all_late_files = []
check_fullcopy_to_announce = UC.get('check_fullcopy_to_announce')
jump_the_line = options.announce if options else False
if jump_the_line:
wfs = session.query(Workflow).filter(Workflow.status.contains('announce')).filter(sqlalchemy.not_(Workflow.status.contains('announced'))).all()
else:
wfs = session.query(Workflow).filter(Workflow.status=='close').all()
held = set()
print len(wfs),"closing"
max_per_round = UC.get('max_per_round').get('closor',None)
if options.limit: max_per_round = options.limit
random.shuffle( wfs )
if max_per_round: wfs = wfs[:max_per_round]
batch_go = {}
batch_warnings = defaultdict(set)
batch_goodness = UC.get("batch_goodness")
for wfo in wfs:
if specific and not specific in wfo.name: continue
## what is the expected #lumis
wfi = workflowInfo(url, wfo.name )
wfo.wm_status = wfi.request['RequestStatus']
if wfi.isRelval():
has_batch_go = False
batch_name = wfi.getCampaign()
if not batch_name in batch_go:
## do the esimatation whethere this can be announced : only once per batch
in_batches = getWorkflowByCampaign(url , batch_name, details=True)
batch_go[ batch_name ] = all(map(lambda s : not s in ['completed','running-open','running-closed','acquired','assigned','assignment-approved'], [r['RequestStatus'] for r in in_batches]))
## already verified
has_batch_go = batch_go[batch_name]
if not has_batch_go:
wfi.sendLog('closor', 'Cannot close for now because the batch %s is not all close'% batch_name)
continue
if wfi.request['RequestStatus'] in ['announced','normal-archived'] and not options.force:
## manually announced ??
wfo.status = 'done'
wfo.wm_status = wfi.request['RequestStatus']
wfi.sendLog('closor','%s is announced already : %s'%( wfo.name,wfo.wm_status))
session.commit()
if jump_the_line:
wfi.sendLog('closor','Announcing while completing')
expected_lumis = 1
if not 'TotalInputLumis' in wfi.request:
print wfo.name,"has not been assigned yet, or the database is corrupted"
elif wfi.request['TotalInputLumis']==0:
print wfo.name,"is corrupted with 0 expected lumis"
else:
expected_lumis = wfi.request['TotalInputLumis']
## what are the outputs
outputs = wfi.request['OutputDatasets']
## check whether the number of lumis is as expected for each
all_OK = defaultdict(lambda : False)
stats = defaultdict(int)
#print outputs
if len(outputs):
print wfo.name,wfi.request['RequestStatus']
for out in outputs:
event_count,lumi_count = getDatasetEventsAndLumis(dataset=out)
odb = session.query(Output).filter(Output.datasetname==out).first()
if not odb:
print "adding an output object",out
odb = Output( datasetname = out )
odb.workflow = wfo
session.add( odb )
odb.nlumis = lumi_count
odb.nevents = event_count
odb.workfow_id = wfo.id
if odb.expectedlumis < expected_lumis:
odb.expectedlumis = expected_lumis
else:
expected_lumis = odb.expectedlumis
odb.date = time.mktime(time.gmtime())
session.commit()
fraction = lumi_count/float(expected_lumis)*100.
completion_line = "%60s %d/%d = %3.2f%%"%(out,lumi_count,expected_lumis,fraction)
wfi.sendLog('closor',"\t%s"% completion_line)
if wfi.isRelval() and fraction < batch_goodness:
batch_warnings[ wfi.getCampaign()].add( completion_line )
stats[out] = lumi_count
all_OK[out] = True
## check for at least one full copy prior to moving on
in_full = {}
for out in outputs:
in_full[out] = []
presence = getDatasetPresence( url, out )
where = [site for site,info in presence.items() if info[0]]
if where:
all_OK[out] = True
print out,"is in full at",",".join(where)
in_full[out] = copy.deepcopy(where)
else:
going_to = wfi.request['NonCustodialSites']+wfi.request['CustodialSites']
wfi.sendLog('closor',"%s is not in full anywhere. send to %s"%(out, ",".join(sorted(going_to))))
at_destination = dict([(k,v) for (k,v) in presence.items() if k in going_to])
else_where = dict([(k,v) for (k,v) in presence.items() if not k in going_to])
print json.dumps( at_destination )
print json.dumps( else_where, indent=2 )
## do the full stuck transfer study, missing files and shit !
for there in going_to:
late_info = findLateFiles(url, out, going_to = there )
for l in late_info:
l.update({"workflow":wfo.name,"dataset":out})
all_late_files.extend( late_info )
if check_fullcopy_to_announce:
## only set this false if the check is relevant
all_OK[out] = False
## verify if we have to do harvesting
if not options.no_harvest and not jump_the_line:
(OK, requests) = spawn_harvesting(url, wfi, in_full)
all_OK.update( OK )
## only that status can let me go into announced
if all(all_OK.values()) and ((wfi.request['RequestStatus'] in ['closed-out']) or options.force or jump_the_line):
print wfo.name,"to be announced"
results=[]
if not results:
for out in outputs:
if out in stats and not stats[out]:
continue
_,dsn,process_string,tier = out.split('/')
if all_OK[out]:
results.append(setDatasetStatus(out, 'VALID'))
if all_OK[out] and wfi.isRelval():
## make the specific relval rules and the replicas
## figure the destination(s) out
destinations = set()
if tier != "RECO" and tier != "ALCARECO":
destinations.add('T2_CH_CERN')
if tier == "GEN-SIM":
destinations.add('T1_US_FNAL_Disk')
if tier == "GEN-SIM-DIGI-RAW":
destinations.add('T1_US_FNAL_Disk')
if tier == "GEN-SIM-RECO":
destinations.add('T1_US_FNAL_Disk')
if "RelValTTBar" in dsn and "TkAlMinBias" in process_string and tier != "ALCARECO":
destinations.add('T2_CH_CERN')
if "MinimumBias" in dsn and "SiStripCalMinBias" in process_string and tier != "ALCARECO":
destinations.add('T2_CH_CERN')
if destinations:
wfi.sendLog('closor', '%s to go to %s'%(out, ', '.join( sorted( destinations ))))
## call to makereplicarequest under relval => done
for site in destinations:
result = makeReplicaRequest(url, site, [out], 'Copy for release validation consumption', priority='normal', approve=True, mail=False, group='RelVal')
try:
request_id = result['phedex']['request_created'][0]['id']
results.append( True )
except:
results.append( 'Failed relval transfer' )
elif all_OK[out]:
campaign = None
try:
campaign = out.split('/')[2].split('-')[0]
except:
if 'Campaign' in wfi.request and wfi.request['Campaign']:
campaign = wfi.request['Campaign']
to_DDM = False
## campaign override
if campaign and campaign in CI.campaigns and 'toDDM' in CI.campaigns[campaign] and tier in CI.campaigns[campaign]['toDDM']:
to_DDM = True
## by typical enabling
if tier in UC.get("tiers_to_DDM"):
to_DDM = True
## check for unitarity
if not tier in UC.get("tiers_no_DDM")+UC.get("tiers_to_DDM"):
print "tier",tier,"neither TO or NO DDM for",out
results.append('Not recognitized tier %s'%tier)
#sendEmail("failed DDM injection","could not recognize %s for injecting in DDM"% out)
sendLog('closor', "could not recognize %s for injecting in DDM"% out, level='critical')
continue
n_copies = 2
destinations=[]
if to_DDM and campaign and campaign in CI.campaigns and 'DDMcopies' in CI.campaigns[campaign]:
ddm_instructions = CI.campaigns[campaign]['DDMcopies']
if type(ddm_instructions) == int:
n_copies = CI.campaigns[campaign]['DDMcopies']
elif type(ddm_instructions) == dict:
## a more fancy configuration
for ddmtier,indication in ddm_instructions.items():
if ddmtier==tier or ddmtier in ['*','all']:
## this is for us
if 'N' in indication:
n_copies = indication['N']
if 'host' in indication:
destinations = indication['host']
destination_spec = ""
if destinations:
destination_spec = "--destination="+",".join( destinations )
group_spec = "" ## not used yet
### should make this a campaign configuration
## inject to DDM when necessary
if to_DDM:
print "Sending",out," to DDM"
p = os.popen('python assignDatasetToSite.py --nCopies=%d --dataset=%s %s %s --debug 0 --exec'%(n_copies, out,destination_spec, group_spec))
ddm_text = p.read()
print ddm_text
status = p.close()
if status!=None:
print "Failed DDM, retrying to send",out,"a second time"
p = os.popen('python assignDatasetToSite.py --nCopies=%d --dataset=%s %s %s --debug 1 --exec'%(n_copies, out,destination_spec, group_spec))
ddm_text = p.read()
print ddm_text
status = p.close()
if status!=None:
#sendEmail("failed DDM injection","could not add "+out+" to DDM pool. check closor logs.")
sendLog('closor',"could not add "+out+" to DDM pool. check closor logs.", level='critical')
if options.force: status = True
results.append( status )
if status == None:
wfi.sendLog('closor',ddm_text)
wfi.sendLog('closor','%s is send to AnalysisOps DDM pool in %s copies %s'%( out, n_copies, destination_spec))
else:
print wfo.name,"no stats for announcing",out
results.append('No Stats')
if all(map(lambda result : result in ['None',None,True],results)):
if not jump_the_line:
## only announce if all previous are fine
res = reqMgrClient.announceWorkflowCascade(url, wfo.name)
if not res in ['None',None]:
## check the status again, it might well have toggled
wl_bis = workflowInfo(url, wfo.name)
wfo.wm_status = wl_bis.request['RequestStatus']
session.commit()
if wl_bis.request['RequestStatus'] in ['announced','normal-archived']:
res = None
else:
## retry ?
res = reqMgrClient.announceWorkflowCascade(url, wfo.name)
results.append( res )
#print results
if all(map(lambda result : result in ['None',None,True],results)):
if jump_the_line:
if not 'announced' in wfo.status:
wfo.status = wfo.status.replace('announce','announced')
else:
wfo.status = 'done'
session.commit()
wfi.sendLog('closor',"workflow outputs are announced")
else:
wfi.sendLog('closor',"Error with %s to be announced \n%s"%( wfo.name, json.dumps( results )))
elif wfi.request['RequestStatus'] in ['failed','aborted','aborted-archived','rejected','rejected-archived','aborted-completed']:
if wfi.isRelval():
wfo.status = 'forget'
wfo.wm_status = wfi.request['RequestStatus']
wfi.sendLog('closor',"%s is %s, but will not be set in trouble to find a replacement."%( wfo.name, wfo.wm_status))
else:
wfo.status = 'trouble'
wfo.wm_status = wfi.request['RequestStatus']
session.commit()
else:
print wfo.name,"not good for announcing:",wfi.request['RequestStatus']
wfi.sendLog('closor',"cannot be announced")
held.add( wfo.name )
days_late = 0.
retries_late = 10
really_late_files = [info for info in all_late_files if info['retries']>=retries_late]
really_late_files = [info for info in really_late_files if info['delay']/(60*60*24.)>=days_late]
if really_late_files:
subject = 'These %d files are lagging for %d days and %d retries announcing dataset \n%s'%(len(really_late_files), days_late, retries_late, json.dumps( really_late_files , indent=2) )
#sendEmail('waiting for files to announce', subject)
sendLog('closor', subject, level='warning')
sendLog('closor',subject)
print subject
open('%s/stuck_files.json'%monitor_dir,'w').write( json.dumps( really_late_files , indent=2))
if held:
sendLog('closor',"the workflows below are held up \n%s"%("\n".join( sorted(held) )), level='critical')
#batches = json.loads(open('batches.json').read())
for bname,go in batch_go.items():
if go:
subject = "Release Validation Samples Batch %s"% bname
issues=""
if batch_warnings[ bname ]:
issues="The following datasets have outstanding completion (<%d%%) issues:\n\n"% batch_goodness
issues+="\n".join( sorted( batch_warnings[ bname ] ))
issues+="\n\n"
text = """
Dear all,
a batch of release validation workflows has finished.
Batch ID:
%s
Detail of the workflows
https://dmytro.web.cern.ch/dmytro/cmsprodmon/requests.php?campaign=%s
%s
This is an automated message.
"""%( bname,
bname,
issues)
to = ['*****@*****.**']
sendEmail(subject, text, destination=to )
def transferor(url, specific=None, talk=True, options=None):
if userLock('transferor'): return
if options and options.test:
execute = False
else:
execute = True
SI = siteInfo()
CI = campaignInfo()
mcm = McMClient(dev=False)
dss = DSS()
print "counting all being handled..."
being_handled = len(
session.query(Workflow).filter(Workflow.status == 'away').all())
being_handled += len(
session.query(Workflow).filter(
Workflow.status.startswith('stag')).all())
being_handled += len(
session.query(Workflow).filter(
Workflow.status.startswith('assistance')).all())
max_to_handle = options.maxworkflows
allowed_to_handle = max(0, max_to_handle - being_handled)
wf_buffer = 5
if allowed_to_handle <= wf_buffer: ## buffer for having several wf per transfer
print "Not allowed to run more than", max_to_handle, "at a time. Currently", being_handled, "and", wf_buffer, "buffer"
else:
print being_handled, "already being handled", max_to_handle, "max allowed,", allowed_to_handle, "remaining", "and", wf_buffer, "buffer"
print "... done"
all_transfers = defaultdict(list)
workflow_dependencies = defaultdict(
set) ## list of wf.id per input dataset
wfs_and_wfh = []
print "getting all wf to consider ..."
cache = getWorkflows(url, 'assignment-approved', details=True)
for wfo in session.query(Workflow).filter(
Workflow.status == 'considered').all():
if specific and not specific in wfo.name: continue
cache_r = filter(lambda d: d['RequestName'] == wfo.name, cache)
if len(cache_r):
wfs_and_wfh.append((wfo,
workflowInfo(url,
wfo.name,
spec=False,
request=cache_r[0])))
else:
wfs_and_wfh.append((wfo, workflowInfo(url, wfo.name, spec=False)))
print "... done"
input_sizes = {}
## list the size of those in transfer already
in_transfer_priority = 0
min_transfer_priority = 100000000
print "getting all wf in staging ..."
for wfo in session.query(Workflow).filter(
Workflow.status == 'staging').all():
wfh = workflowInfo(url, wfo.name, spec=False)
(_, primary, _, _) = wfh.getIO()
for prim in primary:
input_sizes[prim] = dss.get(prim)
in_transfer_priority = max(in_transfer_priority,
int(wfh.request['RequestPriority']))
min_transfer_priority = min(min_transfer_priority,
int(wfh.request['RequestPriority']))
print "... done"
print "Max priority in transfer already", in_transfer_priority
print "Min priority in transfer already", min_transfer_priority
in_transfer_already = sum(input_sizes.values())
#sort by priority higher first
wfs_and_wfh.sort(cmp=lambda i, j: cmp(int(i[1].request[
'RequestPriority']), int(j[1].request['RequestPriority'])),
reverse=True)
## list the size of all inputs
print "getting all input sizes ..."
for (wfo, wfh) in wfs_and_wfh:
(_, primary, _, _) = wfh.getIO()
for prim in primary:
input_sizes[prim] = dss.get(prim)
print "... done"
grand_total = sum(input_sizes.values())
to_transfer = grand_total - in_transfer_already
grand_transfer_limit = options.maxtransfer
transfer_limit = grand_transfer_limit - in_transfer_already
print "%15.4f GB already being transfered" % in_transfer_already
print "%15.4f GB is the current requested transfer load" % to_transfer
print "%15.4f GB is the global transfer limit" % grand_transfer_limit
print "%15.4f GB is the available limit" % transfer_limit
# the max priority value per dataset.
max_priority = defaultdict(int)
needs_transfer = 0 ## so that we can count'em
passing_along = 0
transfer_sizes = {}
went_over_budget = False
for (wfo, wfh) in wfs_and_wfh:
print wfh.request['RequestPriority']
print wfo.name, "to be transfered"
#wfh = workflowInfo( url, wfo.name)
(_, primary, _, _) = wfh.getIO()
this_load = sum([input_sizes[prim] for prim in primary])
if (this_load
and (sum(transfer_sizes.values()) + this_load > transfer_limit
or went_over_budget)):
if went_over_budget:
print "Transfer has gone over bubget."
else:
print "Transfer will go over bubget."
print "%15.4f GB this load" % this_load
print "%15.4f GB already this round" % sum(transfer_sizes.values())
print "%15.4f GB is the available limit" % transfer_limit
went_over_budget = True
if int(
wfh.request['RequestPriority']
) >= in_transfer_priority and min_transfer_priority != in_transfer_priority:
print "Higher priority sample", wfh.request[
'RequestPriority'], ">=", in_transfer_priority, "go-on over budget"
else:
if not options.go:
print min_transfer_priority, "minimum priority", wfh.request[
'RequestPriority'], "<", in_transfer_priority, "stop"
continue
## throtlle by campaign go
if not CI.go(wfh.request['Campaign']):
print "No go for", wfh.request['Campaign']
if not options.go: continue
## check if the batch is announced
announced = False
is_real = False
for b in mcm.getA('batches', query='contains=%s' % wfo.name):
is_real = True
if b['status'] == 'announced':
announced = True
break
if not announced:
print wfo.name, "does not look announced." # skipping?, rejecting?, reporting?"
if not is_real:
print wfo.name, "does not appear to be genuine."
## prevent any duplication. if the wf is not mentioned in any batch, regardless of status
continue
## check on a grace period
injection_time = time.mktime(
time.strptime('.'.join(map(str, wfh.request['RequestDate'])),
"%Y.%m.%d.%H.%M.%S")) / (60. * 60.)
now = time.mktime(time.gmtime()) / (60. * 60.)
if float(now - injection_time) < 4.:
if not options.go and not announced:
print "It is too soon to start transfer: %3.2fH remaining" % (
now - injection_time)
continue
passing_along += 1
if passing_along >= allowed_to_handle:
if int(
wfh.request['RequestPriority']
) >= in_transfer_priority and min_transfer_priority != in_transfer_priority:
print "Higher priority sample", wfh.request[
'RequestPriority'], ">=", in_transfer_priority, "go-on over", max_to_handle
else:
print "Not allowed to pass more than", max_to_handle, "at a time. Currently", being_handled, "handled, and adding", passing_along
break
(lheinput, primary, parent, secondary) = wfh.getIO()
if options and options.tosites:
sites_allowed = options.tosites.split(',')
else:
sites_allowed = getSiteWhiteList(
(lheinput, primary, parent, secondary))
if 'SiteWhitelist' in CI.parameters(wfh.request['Campaign']):
sites_allowed = CI.parameters(
wfh.request['Campaign'])['SiteWhitelist']
blocks = []
if 'BlockWhitelist' in wfh.request and wfh.request['BlockWhitelist']:
blocks = wfh.request['BlockWhitelist']
can_go = True
staging = False
if primary:
if talk:
print wfo.name, 'reads', ', '.join(primary), 'in primary'
## chope the primary dataset
for prim in primary:
max_priority[prim] = max(max_priority[prim],
int(wfh.request['RequestPriority']))
sites_really_allowed = [
site for site in sites_allowed if not any([
osite.startswith(site)
for osite in SI.sites_veto_transfer
])
]
print "Sites allowed minus the vetoed transfer"
print sites_really_allowed
copies_needed = int(
0.35 * len(sites_really_allowed)
) + 1 ## should just go for a fixed number based if the white list grows that big
print "Would make", copies_needed, "copies"
if options.maxcopy > 0:
copies_needed = min(options.maxcopy, copies_needed)
## remove the sites that do not want transfers
print "need", copies_needed
workflow_dependencies[prim].add(wfo.id)
presence = getDatasetPresence(url, prim)
prim_location = [
site for site, pres in presence.items() if pres[0] == True
]
if len(prim_location) >= copies_needed:
print "The output is all fully in place at", len(
prim_location), "sites"
continue
# reduce the number of copies required by existing full copies
copies_needed = max(0, copies_needed - len(prim_location))
print "now need", copies_needed
subscriptions = listSubscriptions(url, prim)
prim_destination = list(
set([
site
for (site, (tid, decision)) in subscriptions.items()
if decision and not any([
site.endswith(veto)
for veto in ['MSS', 'Export', 'Buffer']
])
]))
## need to reject from that list the ones with a full copy already: i.e the transfer corresponds to the copy in place
prim_destination = [
site for site in prim_destination
if not site in prim_location
]
## add transfer dependencies
latching_on_transfers = list(
set([
tid
for (site, (tid, decision)) in subscriptions.items()
if decision and site in prim_destination and not any([
site.endswith(veto)
for veto in ['MSS', 'Export', 'Buffer']
])
]))
print latching_on_transfers
for latching in latching_on_transfers:
tfo = session.query(Transfer).filter(
Transfer.phedexid == latching).first()
if not tfo:
tfo = Transfer(phedexid=latching)
tfo.workflows_id = []
session.add(tfo)
if not wfo.id in tfo.workflows_id:
print "adding", wfo.id, "to", tfo.id, "with phedexid", latching
l = copy.deepcopy(tfo.workflows_id)
l.append(wfo.id)
tfo.workflows_id = l
if not options.test:
session.commit()
else:
session.flush(
) ## regardless of commit later on, we need to let the next wf feeding on this transfer to see it in query
can_go = False
transfer_sizes[prim] = input_sizes[prim]
staging = True
# reduce the number of copies required by the on-going full transfer : how do we bootstrap on waiting for them ??
copies_needed = max(0, copies_needed - len(prim_destination))
print "then need", copies_needed
if copies_needed == 0:
print "The output is either fully in place or getting in full somewhere with", latching_on_transfers
can_go = True
continue
prim_to_distribute = [
site for site in sites_allowed if not any(
[osite.startswith(site) for osite in prim_location])
]
prim_to_distribute = [
site for site in prim_to_distribute if not any(
[osite.startswith(site) for osite in prim_destination])
]
## take out the ones that cannot receive transfers
prim_to_distribute = [
site for site in prim_to_distribute if not any([
osite.startswith(site)
for osite in SI.sites_veto_transfer
])
]
if len(
prim_to_distribute
) > 0: ## maybe that a parameter we can play with to limit the
if not options or options.chop:
spreading = distributeToSites(getDatasetChops(prim),
prim_to_distribute,
n_copies=copies_needed,
weights=SI.cpu_pledges)
else:
spreading = {}
for site in prim_to_distribute:
spreading[site] = [prim]
can_go = False
transfer_sizes[prim] = input_sizes[prim]
for (site, items) in spreading.items():
all_transfers[site].extend(items)
if secondary:
if talk:
print wfo.name, 'reads', ', '.join(secondary), 'in secondary'
for sec in secondary:
workflow_dependencies[sec].add(wfo.id)
presence = getDatasetPresence(url, sec)
sec_location = [
site for site, pres in presence.items() if pres[1] > 90.
] ## more than 90% of the minbias at sites
subscriptions = listSubscriptions(url, sec)
sec_destination = [site for site in subscriptions]
sec_to_distribute = [
site for site in sites_allowed if
not any([osite.startswith(site) for osite in sec_location])
]
sec_to_distribute = [
site for site in sec_to_distribute if not any(
[osite.startswith(site) for osite in sec_destination])
]
sec_to_distribute = [
site for site in sec_to_distribute if not any([
osite.startswith(site)
for osite in SI.sites_veto_transfer
])
]
if len(sec_to_distribute) > 0:
for site in sec_to_distribute:
all_transfers[site].append(sec)
can_go = False
## is that possible to do something more
if can_go:
## no explicit transfer required this time
if staging:
## but using existing ones
print wfo.name, "latches on existing transfers, and nothing else"
wfo.status = 'staging'
else:
print wfo.name, "should just be assigned NOW to", sites_allowed
wfo.status = 'staged'
print "setting status to", wfo.status
session.commit()
continue
else:
## there is an explicit transfer required
if staging:
## and also using an existing one
print wfo.name, "latches on existing transfers"
if not options.test:
wfo.status = 'staging'
print "setting status to", wfo.status
session.commit()
print wfo.name, "needs a transfer"
needs_transfer += 1
#print json.dumps(all_transfers)
fake_id = -1
wf_id_in_prestaging = set()
for (site, items_to_transfer) in all_transfers.iteritems():
items_to_transfer = list(set(items_to_transfer))
## convert to storage element
site_se = SI.CE_to_SE(site)
## site that do not want input datasets
if site in SI.sites_veto_transfer:
print site, "does not want transfers"
continue
## throttle the transfer size to T2s ? we'd be screwed by a noPU sample properly configured.
## massage a bit the items
blocks = [it for it in items_to_transfer if '#' in it]
datasets = [it for it in items_to_transfer if not '#' in it]
if execute:
print "Making a replica to", site, "(CE)", site_se, "(SE) for"
else:
print "Would make a replica to", site, "(CE)", site_se, "(SE) for"
print "\t", len(blocks), "blocks"
## remove blocks if full dataset is send out
blocks = [
block for block in blocks if not block.split('#')[0] in datasets
]
print "\t", len(blocks), "needed blocks for", list(
set([block.split('#')[0] for block in blocks]))
print "\t", len(datasets), "datasets"
print "\t", datasets
items_to_transfer = blocks + datasets
## operate the transfer
if options and options.stop:
## ask to move-on
answer = raw_input('Continue with that ?')
if not answer.lower() in ['y', 'yes', 'go']:
continue
if execute:
result = makeReplicaRequest(url,
site_se,
items_to_transfer,
'prestaging',
priority='normal')
## make use of max_priority dataset:priority to set the subscriptions priority
"""
## does not function
once = True
for item in items_to_transfer:
bds = item.split('#')[0]
if max_priority[bds] >= 90000:
if once:
w=10
print "waiting",w,"s before raising priority"
time.sleep(w)
once=False
## raise it to high priority
print item,"subscription priority raised to high at",site_se
#print "This does not work yet properly it seems"
print updateSubscription(url, site_se, item, priority='high')
"""
else:
#result= {'phedex':{'request_created' : [{'id' : fake_id}]}}
result = {'phedex': {'request_created': []}}
fake_id -= 1
if not result:
print "ERROR Could not make a replica request for", site, items_to_transfer, "pre-staging"
continue
for phedexid in [o['id'] for o in result['phedex']['request_created']]:
new_transfer = session.query(Transfer).filter(
Transfer.phedexid == phedexid).first()
print phedexid, "transfer created"
if not new_transfer:
new_transfer = Transfer(phedexid=phedexid)
session.add(new_transfer)
new_transfer.workflows_id = set()
for transfering in list(
set(map(lambda it: it.split('#')[0], items_to_transfer))):
new_transfer.workflows_id.update(
workflow_dependencies[transfering])
new_transfer.workflows_id = list(new_transfer.workflows_id)
wf_id_in_prestaging.update(new_transfer.workflows_id)
session.commit()
## auto approve it
if execute:
approved = approveSubscription(url, phedexid, [site_se])
for wfid in wf_id_in_prestaging:
tr_wf = session.query(Workflow).get(wfid)
if tr_wf and tr_wf.status != 'staging':
if execute:
tr_wf.status = 'staging'
if talk:
print "setting", tr_wf.name, "to staging"
session.commit()
def parse_one(url, wfn, options=None):
def time_point(label="", sub_lap=False):
now = time.mktime(time.gmtime())
nows = time.asctime(time.gmtime())
print "[showError] Time check (%s) point at : %s" % (label, nows)
print "[showError] Since start: %s [s]" % (now - time_point.start)
if sub_lap:
print "[showError] Sub Lap : %s [s]" % (now - time_point.sub_lap)
time_point.sub_lap = now
else:
print "[showError] Lap : %s [s]" % (now - time_point.lap)
time_point.lap = now
time_point.sub_lap = now
time_point.sub_lap = time_point.lap = time_point.start = time.mktime(
time.gmtime())
task_error_site_count = {}
one_explanation = defaultdict(set)
per_task_explanation = defaultdict(set)
if wfn in [
'vlimant_task_EXO-RunIISummer15wmLHEGS-04800__v1_T_170906_141738_1357'
]:
return task_error_site_count, one_explanation
time_point("Starting with %s" % wfn)
threads = []
SI = global_SI()
UC = unifiedConfiguration()
wfi = workflowInfo(url, wfn)
time_point("wfi", sub_lap=True)
where_to_run, missing_to_run, missing_to_run_at = wfi.getRecoveryInfo()
time_point("acdcinfo", sub_lap=True)
all_blocks, needed_blocks_loc, files_in_blocks, files_and_loc_notin_dbs = wfi.getRecoveryBlocks(
)
time_point("inputs", sub_lap=True)
ancestor = workflowInfo(url, wfn)
lhe, prim, _, sec = ancestor.getIO()
high_order_acdc = 0
while ancestor.request['RequestType'] == 'Resubmission':
ancestor = workflowInfo(url, ancestor.request['OriginalRequestName'])
lhe, prim, _, sec = ancestor.getIO()
high_order_acdc += 1
no_input = (not lhe) and len(prim) == 0 and len(sec) == 0
cache = options.cache
print "cache timeout", cache
err = wfi.getWMErrors(cache=cache)
time_point("wmerrors", sub_lap=True)
stat = wfi.getWMStats(cache=cache)
time_point("wmstats", sub_lap=True)
#adcd = wfi.getRecoveryDoc()
total_by_code_dash = defaultdict(int)
total_by_site_dash = defaultdict(int)
r_dashb = defaultdict(lambda: defaultdict(int))
dash_board_h = 1
if False:
## NB get the since from when the wf has started, not a fixed value
## no dashboard until we get a better api
#dashb = wfi.getFullPicture(since=dash_board_h,cache=cache)
dashb = {}
#print json.dumps( dashb , indent=2)
for site, sinfo in dashb.items():
for s_code, counts in sinfo.items():
d_statuses = ['submitted', 'pending', 'app-unknown', 'done']
total_by_code_dash[str(s_code)] += counts.get('submitted', 0)
total_by_site_dash[site] += counts.get('submitted', 0)
r_dashb[str(s_code)][site] += counts.get('submitted', 0)
print json.dumps(total_by_code_dash, indent=2)
print json.dumps(total_by_site_dash, indent=2)
time_point("Got most input")
status_per_task = defaultdict(lambda: defaultdict(int))
if not 'AgentJobInfo' in stat:
stat['AgentJobInfo'] = {}
#print "bad countent ?"
#print json.dumps( stat, indent=2)
for agent in stat['AgentJobInfo']:
for task in stat['AgentJobInfo'][agent]['tasks']:
if not 'status' in stat['AgentJobInfo'][agent]['tasks'][task]:
continue
for status in stat['AgentJobInfo'][agent]['tasks'][task]['status']:
info = stat['AgentJobInfo'][agent]['tasks'][task]['status'][
status]
#print status,stat['AgentJobInfo'][agent]['tasks'][task]['status'][status]
if type(info) == dict:
status_per_task[task][status] += sum(
stat['AgentJobInfo'][agent]['tasks'][task]['status']
[status].values())
else:
status_per_task[task][status] += stat['AgentJobInfo'][
agent]['tasks'][task]['status'][status]
#print json.dumps( status_per_task, indent=2)
db_total_per_site = defaultdict(int)
db_total_per_code = defaultdict(int)
## cannot do that since there is no task count in dashboard and we have to take away the submitted
#for site in dashb:
# for error in dashb[site]:
# db_total_per_site[site] += dashb[site][error]
# db_total_per_code[code] += dashb[site][error]
print "ACDC Information"
print "\t where to re-run"
print json.dumps(where_to_run, indent=2)
print "\t Missing events"
print json.dumps(missing_to_run, indent=2)
print "\t Missing events per site"
print json.dumps(missing_to_run_at, indent=2)
if not where_to_run and not missing_to_run and not missing_to_run_at:
print "showError is unable to run"
#return task_error_site_count, one_explanation
pass
do_JL = not options.no_JL
do_CL = not options.no_CL
do_all_error_code = options.all_errors
if high_order_acdc >= 1:
print high_order_acdc, "order request, pulling down all logs"
do_all_error_code = True
if wfi.isRelval():
print "getting all codes for relval"
do_all_error_code = True
tasks = sorted(set(err.keys() + missing_to_run.keys()))
if not tasks:
print "no task to look at"
#return task_error_site_count
html = "<html> <center><h1>%s, Updated on %s (GMT)" % (
wfn, time.asctime(time.gmtime()))
html += '</center>'
html += '<a href=https://cmsweb.cern.ch/reqmgr2/fetch?rid=%s>dts</a>, ' % (
wfn)
html += '<a href=https://dmytro.web.cern.ch/dmytro/cmsprodmon/workflows.php?prep_id=%s>ac</a>, ' % (
wfi.request['PrepID'])
html += '<a href=https://cms-gwmsmon.cern.ch/prodview/%s>Job Progress</a>, ' % (
wfn)
r_type = wfi.request.get('OriginalRequestType',
wfi.request.get('RequestType', 'NaT'))
if r_type in ['ReReco']:
html += '<a href=../datalumi/lumi.%s.html>Lumisection Summary</a>, ' % wfi.request[
'PrepID']
html += '<a href="https://its.cern.ch/jira/issues/?jql=text~%s AND project = CMSCOMPPR" target="_blank">jira</a>' % (
wfi.request['PrepID'])
html += '<hr>'
html += '<a href=#IO>I/O</a>, <a href=#ERROR>Errors</a>, <a href=#BLOCK>blocks</a>, <a href=#FILE>files</a>, <a href=#CODES>Error codes</a><br>'
html += '<hr>'
time_point("Header writen")
html += '<a name=IO></a>'
if prim:
html += 'Reads in primary<br>'
rwl = wfi.getRunWhiteList()
lwl = wfi.getLumiWhiteList()
for dataset in prim:
html += '<b>%s </b>(events/lumi ~%d)' % (
dataset, getDatasetEventsPerLumi(dataset))
blocks = getDatasetBlocks(dataset, runs=rwl) if rwl else None
blocks = getDatasetBlocks(dataset, lumis=lwl) if lwl else None
available = getDatasetBlocksFraction(url,
dataset,
only_blocks=blocks)
html += '<br><br>Available %.2f (>1 more than one copy, <1 not in full on disk)<br>' % available
html += '<ul>'
presence = getDatasetPresence(url, dataset, only_blocks=blocks)
for site in sorted(presence.keys()):
html += '<li>%s : %.2f %%' % (site, presence[site][1])
html += '</ul><br>'
if sec:
html += 'Reads in secondary<br>'
for dataset in sec:
presence = getDatasetPresence(url, dataset)
html += '<b>%s</b><ul>' % dataset
for site in sorted(presence.keys()):
html += '<li>%s : %.2f %%' % (site, presence[site][1])
html += '</ul>'
outs = sorted(wfi.request['OutputDatasets'])
if outs:
html += 'Produces<br>'
for dataset in outs:
presence = getDatasetPresence(url, dataset)
html += '<b>%s </b>(events/lumi ~ %d)<ul>' % (
dataset, getDatasetEventsPerLumi(dataset))
for site in sorted(presence.keys()):
html += '<li>%s : %.2f %%' % (site, presence[site][1])
html += '</ul>'
time_point("Input checked")
html += """
<hr><br>
<a name=ERROR></a>
<ul>
<li> <b><i>dashboard numbers over %d days</b></i>
<li> ↑ %% with respect to total number of error in the code
<li> → %% with respect to total number of error at the site
</ul>
""" % (dash_board_h)
html += '<br>'
n_expose_base = options.expose # if options else UC.get('n_error_exposed')
print "getting", n_expose_base, "logs by default"
if tasks:
min_rank = min([task.count('/') for task in tasks])
for task in tasks:
n_expose = n_expose_base
expose_archive_code = dict([(str(code), defaultdict(lambda: n_expose))
for code in UC.get('expose_archive_code')])
expose_condor_code = dict([(str(code), defaultdict(lambda: n_expose))
for code in UC.get('expose_condor_code')])
#print task
task_rank = task.count('/')
task_short = task.split('/')[-1]
total_per_site = defaultdict(int)
time_point("Starting with task %s" % task_short, sub_lap=True)
notreported = 'NotReported'
total_count = defaultdict(int)
error_site_count = defaultdict(lambda: defaultdict(int))
all_not_reported = set()
for agent in stat['AgentJobInfo']:
for site in stat['AgentJobInfo'][agent]['tasks'].get(task, {}).get(
'skipped', {}):
info = stat['AgentJobInfo'][agent]['tasks'][task]['skipped'][
site]
#print info
all_not_reported.add(site)
ce = SI.SE_to_CE(site)
error_site_count[notreported][ce] += info.get(
'skippedFiles', 0)
total_count[notreported] += info.get('skippedFiles', 0)
for site in stat['AgentJobInfo'][agent]['tasks'].get(task, {}).get(
'sites', {}):
info = stat['AgentJobInfo'][agent]['tasks'][task]['sites'][
site]
for s in ['success', 'failure', 'cooloff', 'submitted']:
if not s in info: continue
data = info[s]
if type(data) == dict:
total_per_site[site] += sum(data.values())
else:
total_per_site[site] += data
#is the task relevant to recover (discard log, cleanup)
if any([v in task.lower() for v in ['logcol', 'cleanup']]): continue
#total_count= defaultdict(int)
#error_site_count = defaultdict( lambda : defaultdict(int))
if not task in err:
print task, "has not reported error"
err[task] = {}
#print err[task].keys()
for exittype in err[task]:
#print "\t",err[task][exittype].keys()
for errorcode_s in err[task][exittype]:
if errorcode_s == '0': continue
#print "\t\t",err[task][exittype][errorcode_s].keys()
for site in err[task][exittype][errorcode_s]:
ce = SI.SE_to_CE(site)
count = err[task][exittype][errorcode_s][site][
'errorCount']
total_count[errorcode_s] += count
#error_site_count[errorcode_s][site] += count
error_site_count[errorcode_s][ce] += count
## show the total
all_sites = set()
all_codes = set()
for code in error_site_count:
for site in error_site_count[code]:
all_sites.add(site)
if code != '0':
all_codes.add(code)
s_per_code = defaultdict(int)
for site in all_sites:
for code in sorted(all_codes):
s_per_code[code] += error_site_count[code][site]
expose_top_N = UC.get('expose_top_N')
count_top_N = min(
sorted(s_per_code.values(),
reverse=True)[:expose_top_N]) if s_per_code else -1
for exittype in err[task]:
#print "\t",err[task][exittype].keys()
for errorcode_s in err[task][exittype]:
if errorcode_s == '0': continue
#print "\t\t",err[task][exittype][errorcode_s].keys()
force_code = (count_top_N > 0
and s_per_code[errorcode_s] >= count_top_N)
if force_code: print "will expose", errorcode_s, "anyways"
for site in err[task][exittype][errorcode_s]:
ce = SI.SE_to_CE(site)
count = err[task][exittype][errorcode_s][site][
'errorCount']
###total_count[errorcode_s] += count
#error_site_count[errorcode_s][site] += count
###error_site_count[errorcode_s][ce] += count
for sample in err[task][exittype][errorcode_s][site][
'samples']:
#print sample.keys()
for step in sample['errors']:
for report in sample['errors'][step]:
if report['type'] == 'CMSExeption': continue
#if int(report['exitCode']) == int(errorcode_s):
one_explanation[errorcode_s].add(
"%s (Exit code: %s) \n%s" %
(report['type'], report['exitCode'],
report['details']))
per_task_explanation[
"%s:%s" % (task_short, errorcode_s)].add(
"%s (Exit code: %s) \n%s" %
(report['type'], report['exitCode'],
report['details']))
#one_explanation[errorcode_s].add( report['details'] )
#else:
#one_explanation[
agent = sample['agent_name']
wmbs = sample['wmbsid']
workflow = sample['workflow']
if force_code:
if not errorcode_s in expose_condor_code:
expose_condor_code[errorcode_s] = defaultdict(
lambda: n_expose)
if not errorcode_s in expose_archive_code:
expose_archive_code[errorcode_s] = defaultdict(
lambda: n_expose)
if do_CL and ((errorcode_s in expose_condor_code and
expose_condor_code[errorcode_s][agent])
) and 'cern' in agent:
if errorcode_s in expose_condor_code:
expose_condor_code[errorcode_s][agent] -= 1
print errorcode_s, agent, "error count", expose_condor_code.get(
errorcode_s, {}).get(agent, 0)
threads.append(
AgentBuster(agent=agent,
workflow=workflow,
wmbs=wmbs,
errorcode_s=errorcode_s,
base_eos_dir=base_eos_dir,
monitor_eos_dir=monitor_eos_dir,
task_short=task_short))
for out in sample['output']:
#print out
if out['type'] == 'logArchive':
if do_JL and (
(errorcode_s in expose_archive_code and
expose_archive_code[errorcode_s][agent] >
0)):
if errorcode_s in expose_archive_code:
expose_archive_code[errorcode_s][
agent] -= 1
print errorcode_s, agent, "error count", expose_archive_code.get(
errorcode_s, {}).get(agent, 0)
threads.append(
XRDBuster(
out_lfn=out['lfn'],
monitor_eos_dir=monitor_eos_dir,
wfn=wfn,
errorcode_s=errorcode_s,
task_short=task_short,
from_eos=(
not options.not_from_eos
), # if options else True),
))
#print task
#print json.dumps( total_count, indent=2)
#print json.dumps( explanations , indent=2)
all_sites = set()
all_codes = set()
for code in error_site_count:
for site in error_site_count[code]:
all_sites.add(site)
if code != '0':
all_codes.add(code)
## parse the dashboard data
for site in total_by_site_dash:
## no. cannot discriminate by task in dashboard...
#all_sites.add( site )
pass
## parse the acdc data
#notreported='NotReported'
#all_missing_stats = set()
#for site in missing_to_run_at[task] if task in missing_to_run_at else []:
# if not missing_to_run_at[task][site]: continue
# ce = SI.SE_to_CE( site )
# #all_sites.add( ce )
# all_missing_stats.add( ce )
#all_missing_stats = all_missing_stats &set(SI.all_sites)
#all_not_reported = all_missing_stats - all_sites
#print task
#print "site with no report",sorted(all_not_reported)
#print sorted(all_sites)
#print sorted(all_missing_stats)
#all_sites = all_missing_stats | all_sites
#all_sites = all_sites & set(SI.all_sites)
no_error = len(all_not_reported) != 0
if not no_error and notreported in all_codes:
all_codes.remove(notreported)
missing_events = missing_to_run[task] if task in missing_to_run else 0
feff = wfi.getFilterEfficiency(task.split('/')[-1])
html += "<a name=%s>" % task.split('/')[-1]
html += "<b>%s</b>" % task.split('/')[-1]
if missing_events:
if feff != 1.:
html += ' is missing %s events in input and <b>about %s events in output</b>' % (
"{:,}".format(missing_events), "{:,}".format(
int(missing_events * feff)))
else:
html += ' is missing <b>%s events in I/O</b>' % (
"{:,}".format(missing_events))
html += ' <a href="https://cmsweb.cern.ch/couchdb/acdcserver/_design/ACDC/_view/byCollectionName?key=%%22%s%%22&include_docs=true&reduce=false" target=_blank>AC/DC</a>' % (
wfn)
if no_error:
html += "<br><b><font color=red> and has UNreported error</font></b>"
html += "<br><table border=1><thead><tr><th>Sites/Errors</th>"
#for site in all_sites:
# html+='<th>%s</th>'%site
for code in sorted(all_codes):
#html+='<th><a href="#%s">%s</a>'%(code,code)
html += '<th><a href="#%s:%s">%s</a>' % (task_short, code, code)
if (str(code) in expose_archive_code
or do_all_error_code): # and n_expose_base:
html += ' <a href=%s/joblogs/%s/%s/%s>, JobLog</a>' % (
unified_url_eos, wfn, code, task_short)
if (str(code) in expose_condor_code
or do_all_error_code): # and n_expose_base:
html += ' <a href=%s/condorlogs/%s/%s/%s>, CondorLog</a>' % (
unified_url_eos, wfn, code, task_short)
html += '</th>'
html += '<th>Total jobs</th><th>Site Ready</th>'
html += '</tr></thead>\n'
html += '<tr><td>Total</td>'
for code in sorted(all_codes):
html += '<td bgcolor=orange width=100>%d' % (s_per_code[code])
if code in total_by_code_dash:
html += ' (<b><i>%d</i></b>)' % total_by_code_dash[code]
html += '</td>'
ulist = '<ul>'
grand = 0
for status in sorted(status_per_task[task].keys()):
ulist += '<li> %s %d' % (status, status_per_task[task][status])
grand += status_per_task[task][status]
ulist += '<li><b> Total %d </b>' % grand
ulist += '</ul>'
#html += '<td bgcolor=orange> %.2f%% </td>'% (100.*(float(sum(s_per_code.values()))/sum(total_per_site.values())) if sum(total_per_site.values()) else 0.)
html += '<td bgcolor=orange> → %.2f%% ← </td>' % (
100. * (float(sum(s_per_code.values())) / grand) if grand else 0.)
html += '<td bgcolor=orange> %s </td>' % ulist
html += '</tr>'
def palette(frac):
_range = {
0.0: 'green',
0.5: 'green',
0.6: 'darkgreen',
0.7: 'orange',
0.8: 'salmon',
0.9: 'red'
}
which = [k for k in _range.keys() if k <= frac]
if which:
there = max(which)
else:
there = max(_range.keys())
return _range[there]
for site in sorted(all_sites):
site_in = 'Yes'
color = 'bgcolor=lightblue'
if not site in SI.sites_ready:
color = 'bgcolor=indianred'
site_in = '<b>No</b>'
if task in missing_to_run_at and missing_to_run_at[task][
SI.CE_to_SE(site)] == 0 or min_rank == task_rank:
color = 'bgcolor=aquamarine'
site_in = '<b>No</b> but fine'
if not no_error:
site_in += " (%s events)" % ("{:,}".format(
missing_to_run_at[task][SI.CE_to_SE(site)]) if task
in missing_to_run_at else '--')
html += '<tr><td %s>%s</td>' % (color, site)
for code in sorted(all_codes):
if code == notreported:
html += '<td %s width=200>%s events </td>' % (
color, "{:,}".format(
missing_to_run_at[task][SI.CE_to_SE(site)]))
else:
if error_site_count[code][site]:
er_frac = float(
error_site_count[code][site]
) / s_per_code[code] if s_per_code[code] else 0.
si_frac = float(
error_site_count[code][site]) / total_per_site[
site] if total_per_site[site] else 0.
html += '<td %s width=200>%d' % (
color, error_site_count[code][site])
if code in r_dashb and site in r_dashb[code]:
html += ' (<b><i>%d</i></b>)' % (
r_dashb[code][site])
html += ', <font color=%s>↑ %.1f%%</font>, <font color=%s>→ %.1f%%</font></td>' % (
palette(er_frac), 100. * er_frac, palette(si_frac),
100. * si_frac)
else:
html += '<td %s>0</td>' % color
html += '<td bgcolor=orange>%d</td>' % total_per_site[site]
html += '<td %s>%s</td>' % (color, site_in)
html += '</tr>\n'
html += '</table><br>'
task_error_site_count[task] = error_site_count
## run all retrieval
run_threads = ThreadHandler(
threads=threads,
n_threads=options.log_threads, # if options else 5,
sleepy=10,
timeout=UC.get('retrieve_errors_timeout'),
verbose=True)
run_threads.start()
html += '<hr><br>'
html += '<a name=BLOCK></a>'
html += "<b>Blocks (%d/%d) needed for recovery</b><br>" % (
len(needed_blocks_loc), len(all_blocks))
for block in sorted(needed_blocks_loc.keys()):
html += '%s <b>@ %s</b><br>' % (block, ','.join(
sorted(needed_blocks_loc[block])))
html += '<a name=FILE></a>'
html += "<br><b>Files in no block</b><br>"
rthreads = []
check_files = [f for f in files_and_loc_notin_dbs.keys() if '/store' in f]
random.shuffle(check_files)
check_files = check_files[:100]
check_files = [] ## disable it completely
by_f = {}
if check_files:
for f in check_files:
rthreads.append(ReadBuster(file=f))
print "checking on existence of", len(rthreads), "files"
run_rthreads = ThreadHandler(threads=rthreads,
n_threads=20,
timeout=10)
run_rthreads.start()
while run_rthreads.is_alive():
time.sleep(10)
for t in run_rthreads.threads:
by_f[t.file] = t.readable
#print "checked",t.file,t.readable
for f in sorted(files_and_loc_notin_dbs.keys()):
readable = by_f.get(f, -1)
if readable == -1:
fs = '%s' % f
elif readable == 0:
fs = '<font color=light green>%s</font>' % f
#print f,"is readable"
else:
fs = '<font color=red>%s</font>' % f
#print f,"is not readable"
html += '%s <b>@</b> %s<br>' % (fs, ','.join(
sorted(files_and_loc_notin_dbs[f])))
#html +='%s <b>@</b> %s<br>'%(f , ','.join(sorted(files_and_loc_notin_dbs[f])) )
html += '<hr><br>'
html += '<a name=CODES></a>'
html += '<table border=1>'
for code in per_task_explanation:
html += '<tr><td><a name="%s">%s</a><br><a href=https://twiki.cern.ch/twiki/bin/view/CMSPublic/JobExitCodes>code twiki</a></td><td>%s</td></tr>' % (
code, code, '<br><br>'.join(per_task_explanation[code]).replace(
'\n', '<br>'))
#for code in one_explanation:
# html +='<tr><td><a name="%s">%s</a></td><td>%s</td></tr>'% ( code, code, '<br><br>'.join(one_explanation[code]).replace('\n','<br>' ))
html += '</table>'
html += ('<br>' * 30)
html += '</html>'
time_point("Report finished")
wfi.sendLog('error', html, show=False)
fn = '%s' % wfn
time_point("error send to ES")
open('%s/report/%s' % (monitor_dir, fn), 'w').write(html)
open('%s/report/%s' % (monitor_eos_dir, fn), 'w').write(html)
time_point("Finished with showError")
## then wait for the retrivals to complete
ping = 0
while run_threads.is_alive():
ping += 1
if ping % 100:
time_point("waiting for sub-threads to finish")
time.sleep(6)
time_point("Finished with retrieval threads")
return task_error_site_count, one_explanation
def equalizor(url, specific=None, options=None):
up = componentInfo(mcm=False, soft=['mcm'])
if not up.check(): return
if not specific:
workflows = getWorkflows(url, status='running-closed', details=True)
workflows.extend(getWorkflows(url, status='running-open',
details=True))
## start from scratch
modifications = defaultdict(dict)
## define regionality site => fallback allowed. feed on an ssb metric ??
mapping = defaultdict(list)
reversed_mapping = defaultdict(list)
regions = defaultdict(list)
SI = siteInfo()
CI = campaignInfo()
UC = unifiedConfiguration()
for site in SI.sites_ready:
region = site.split('_')[1]
if not region in ['US', 'DE', 'IT']: continue
regions[region] = [region]
def site_in_depletion(s):
return True
if s in SI.sites_pressure:
(m, r, pressure) = SI.sites_pressure[s]
if float(m) < float(r):
print s, m, r, "lacking pressure"
return True
else:
print s, m, r, "pressure"
pass
return False
for site in SI.sites_ready:
region = site.split('_')[1]
## fallback to the region, to site with on-going low pressure
mapping[site] = [
fb for fb in SI.sites_ready
if any([('_%s_' %
(reg) in fb and fb != site and site_in_depletion(fb))
for reg in regions[region]])
]
use_T0 = ('T0_CH_CERN' in UC.get("site_for_overflow"))
if options.t0: use_T0 = True
#if options.augment : use_T0 = True
use_HLT = ('T2_CH_CERN_HLT' in UC.get("site_for_overflow"))
if options.hlt: use_HLT = True
#if options.augment : use_HLT=True
if use_HLT:
mapping['T2_CH_CERN'].append('T2_CH_CERN_HLT')
if use_T0:
mapping['T2_CH_CERN'].append('T0_CH_CERN')
#mapping['T1_FR_CCIN2P3'].append('T0_CH_CERN')
#mapping['T2_IT_Legnaro'].append('T1_IT_CNAF')
for reg in ['IT', 'DE', 'UK']:
mapping['T2_CH_CERN'].extend(
[fb for fb in SI.sites_ready if '_%s_' % reg in fb])
## make them appear as OK to use
force_sites = []
## overflow CERN to underutilized T1s
upcoming = json.loads(open('%s/GQ.json' % monitor_dir).read())
for possible in SI.sites_T1s:
if not possible in upcoming:
mapping['T2_CH_CERN'].append(possible)
## remove add-hoc sites from overflow mapping
prevent_sites = ['T2_US_Purdue']
for prevent in prevent_sites:
if prevent in mapping: mapping.pop(prevent)
for src in mapping:
for prevent in prevent_sites:
if prevent in mapping[src]:
mapping[src].remove(prevent)
## create the reverse mapping for the condor module
for site, fallbacks in mapping.items():
for fb in fallbacks:
reversed_mapping[fb].append(site)
## this is the fallback mapping
print "Direct mapping : site => overflow"
print json.dumps(mapping, indent=2)
print "Reverse mapping : dest <= from origin"
print json.dumps(reversed_mapping, indent=2)
altered_tasks = set()
def running_idle(wfi, task_name):
gmon = wfi.getGlideMon()
#print gmon
if not gmon: return (0, 0)
if not task_name in gmon: return (0, 0)
return (gmon[task_name]['Running'], gmon[task_name]['Idle'])
def needs_action(wfi, task, min_idled=100, pressure=0.2):
task_name = task.pathName.split('/')[-1]
running, idled = running_idle(wfi, task_name)
go = True
if not idled and not running:
go = False
if idled < 100:
go = False
if (not running and idled) or (running and
(idled / float(running) > pressure)):
go = True
else:
go = False
return go, task_name, running, idled
def getPerf(task):
task = task.split('/')[1] + '/' + task.split('/')[-1]
try:
u = 'http://cms-gwmsmon.cern.ch/prodview/json/history/memoryusage720/%s' % task
print u
perf_data = json.loads(os.popen('curl -s --retry 5 %s' % u).read())
except Exception as e:
print str(e)
return (None, None)
buckets = perf_data['aggregations']["2"]['buckets']
s_m = sum(bucket['key'] * bucket['doc_count'] for bucket in buckets)
w_m = sum(bucket['doc_count'] for bucket in buckets)
m_m = max(bucket['key'] for bucket in buckets) if buckets else None
b_m = None
if w_m > 100:
b_m = m_m
try:
perf_data = json.loads(
os.popen(
'curl -s --retry 5 http://cms-gwmsmon.cern.ch/prodview/json/history/runtime720/%s'
% task).read())
except Exception as e:
print str(e)
return (b_m, None)
buckets = perf_data['aggregations']["2"]['buckets']
s_t = sum(bucket['key'] * bucket['doc_count'] for bucket in buckets)
w_t = sum(bucket['doc_count'] for bucket in buckets)
m_t = max(bucket['key'] for bucket in buckets) if buckets else None
b_t = None
if w_t > 100:
b_t = m_t
return (b_m, b_t)
def getcampaign(task):
taskname = task.pathName.split('/')[-1]
if hasattr(task, 'prepID'):
return task.prepID.split('-')[1]
elif taskname.count('-') >= 1:
return taskname.split('-')[1]
else:
return None
def close(interface):
open('%s/equalizor.json.new' % monitor_dir,
'w').write(json.dumps(interface, indent=2))
os.system('mv %s/equalizor.json.new %s/equalizor.json' %
(monitor_dir, monitor_dir))
os.system('cp %s/equalizor.json %s/logs/equalizor/equalizor.%s.json' %
(monitor_dir, monitor_dir, time.mktime(time.gmtime())))
interface = {'reversed_mapping': reversed_mapping, 'modifications': {}}
if options.augment or options.remove:
interface['modifications'] = json.loads(
open('%s/equalizor.json' % monitor_dir).read())['modifications']
if options.remove:
if specific in interface['modifications']:
print "poping", specific
interface['modifications'].pop(specific)
close(interface)
return
PU_locations = {}
PU_overflow = {}
LHE_overflow = {}
tune_performance = []
pending_HLT = 0
max_HLT = 60000
pending_T0 = 0
max_T0 = 60000
try:
gmon = json.loads(
os.popen(
'curl -s http://cms-gwmsmon.cern.ch/prodview/json/T2_CH_CERN_HLT'
).read())
pending_HLT += gmon["Running"]
pending_HLT += gmon["MatchingIdle"]
except:
pass
stay_within_site_whitelist = False
specific_task = None
if specific and ":" in specific:
specific, specific_task = specific.split(':')
if specific:
wfs = session.query(Workflow).filter(
Workflow.name.contains(specific)).all()
else:
wfs = session.query(Workflow).filter(Workflow.status == 'away').all()
performance = {}
no_routing = []
random.shuffle(wfs)
for wfo in wfs:
if wfo.name in no_routing and not options.augment:
continue
if specific and not specific in wfo.name:
continue
if specific:
wfi = workflowInfo(url, wfo.name)
else:
cached = filter(lambda d: d['RequestName'] == wfo.name, workflows)
if not cached: continue
wfi = workflowInfo(url, wfo.name, request=cached[0])
## only running should get re-routed
if not wfi.request['RequestStatus'] in [
'running-open', 'running-closed'
] and not specific:
continue
tasks_and_campaigns = []
for task in wfi.getWorkTasks():
tasks_and_campaigns.append((task, getcampaign(task)))
_, _, _, sec = wfi.getIO()
## check needs override
needs_overide = False
if not needs_overide and options.augment: needs_overide = True
def overide_from_agent(wfi, needs_overide):
bad_agents = [] #'http://cmssrv219.fnal.gov:5984']
if not bad_agents: return needs_overide
if needs_overide: return True
agents = wfi.getAgents()
wqss = ['Running', 'Acquired']
if any([
agent in agents.get(wqs, {}).keys()
for wqs, agent in itertools.product(wqss, bad_agents)
]):
print "overriding the need for bad agent"
needs_overide = True
return needs_overide
## now parse this for action
for i_task, (task, campaign) in enumerate(tasks_and_campaigns):
if options.augment:
print task.pathName
print campaign
tune = CI.get(campaign, 'tune', options.tune)
if tune and not campaign in tune_performance:
tune_performance.append(campaign)
overflow = CI.get(campaign, 'overflow', {})
if overflow:
if "PU" in overflow and not campaign in PU_overflow:
PU_overflow[campaign] = copy.deepcopy(overflow['PU'])
print "adding", campaign, "to PU overflow rules"
if "LHE" in overflow and not campaign in LHE_overflow:
print "adding", campaign, "to light input overflow rules"
site_list = overflow['LHE']['site_list']
LHE_overflow[campaign] = copy.deepcopy(
getattr(SI, site_list))
### get the task performance, for further massaging.
if campaign in tune_performance or options.tune:
print "performance", task.taskType, task.pathName
if task.taskType in ['Processing', 'Production']:
set_memory, set_time = getPerf(task.pathName)
#print "Performance %s GB %s min"%( set_memory,set_time)
wfi.sendLog(
'equalizor', 'Performance tuning to %s GB %s min' %
(set_memory, set_time))
## get values from gmwsmon
# massage the values : 95% percentile
performance[task.pathName] = {}
if set_memory:
performance[task.pathName]['memory'] = set_memory
if set_time and False:
performance[task.pathName]['time'] = set_time
### rule to avoid the issue of taskchain secondary jobs being stuck at sites processing the initial step
if campaign in LHE_overflow:
if task.taskType in ['Processing']:
needs, task_name, running, idled = needs_action(wfi, task)
needs_overide = overide_from_agent(wfi, needs_overide)
extend_to = list(set(copy.deepcopy(
LHE_overflow[campaign])))
if stay_within_site_whitelist:
extend_to = list(
set(extend_to) & set(wfi.request['SiteWhitelist'])
) ## restrict to stupid-site-whitelist
extend_to = list(
set(extend_to) & set(SI.sites_ready + force_sites))
if extend_to and needs or needs_overide:
modifications[wfo.name][task.pathName] = {
"ReplaceSiteWhitelist": extend_to,
"Running": running,
"Pending": idled,
"Priority": wfi.request['RequestPriority']
}
wfi.sendLog(
'equalizor',
'%s of %s is running %d and pending %d, taking action : ReplaceSiteWhitelist \n %s'
%
(task_name, wfo.name, running, idled,
json.dumps(
sorted(modifications[wfo.name][task.pathName]
['ReplaceSiteWhitelist']))))
altered_tasks.add(task.pathName)
else:
wfi.sendLog(
'equalizor',
'%s of %s is running %d and pending %d' %
(task_name, wfo.name, running, idled))
### overflow the 76 digi-reco to the site holding the pileup
if campaign in PU_overflow:
force = PU_overflow[campaign][
'force'] if 'force' in PU_overflow[campaign] else False
secondary_locations = set(SI.sites_ready + force_sites)
for s in sec:
if not s in PU_locations:
presence = getDatasetPresence(url, s)
#one_secondary_locations = [site for (site,(there,frac)) in presence.items() if there]
one_secondary_locations = [
site for (site, (there, frac)) in presence.items()
if frac > 98.
]
PU_locations[s] = one_secondary_locations
print "secondary is at", sorted(PU_locations[s])
secondary_locations = set(
[SI.SE_to_CE(site)
for site in PU_locations[s]]) & secondary_locations
## we should add all sites that hold the secondary input if any
### given that we have the secondary location available, it is not necessary to use the add-hoc list
##secondary_locations = list(set(PU_overflow[campaign]['sites']) & set( SI.sites_ready ))
if any([
task.pathName.endswith(finish)
for finish in ['_0', 'StepOneProc', 'Production']
]):
needs, task_name, running, idled = needs_action(wfi, task)
## removing the ones in the site whitelist already since they encode the primary input location
if stay_within_site_whitelist:
original_site_in_use = set(
wfi.request['SiteWhitelist'])
else:
original_site_in_use = set(secondary_locations)
## remove the sites that have already running jobs
gmon = wfi.getGlideMon()
if gmon and task_name in gmon and 'Sites' in gmon[
task_name]:
site_in_use = set(gmon[task_name]['Sites'])
print "removing", sorted(site_in_use)
## that determines where you want to run in addition
augment_by = list((set(secondary_locations) -
site_in_use) & original_site_in_use)
else:
print "no existing running site"
augment_by = list(original_site_in_use)
needs_overide = overide_from_agent(wfi, needs_overide)
if augment_by and (
needs or needs_overide
or force) and PU_overflow[campaign][
'pending'] < PU_overflow[campaign]['max']:
PU_overflow[campaign]['pending'] += idled
print "raising overflow to", PU_overflow[campaign][
'pending'], "for", PU_overflow[campaign]['max']
## the step with an input ought to be the digi part : make this one go anywhere
modifications[wfo.name][task.pathName] = {
"AddWhitelist": augment_by,
"Running": running,
"Pending": idled,
"Priority": wfi.request['RequestPriority']
}
altered_tasks.add(task.pathName)
wfi.sendLog(
'equalizor',
'%s of %s is running %d and pending %d, taking action : AddWhitelist \n %s'
% (task_name, wfo.name, running, idled,
json.dumps(sorted(augment_by), indent=2)))
else:
print task_name, "of", wfo.name, "running", running, "and pending", idled
### overflow the skims back to multi-core
if campaign in ['Run2015D', 'Run2015C_25ns'
] and task.taskType == 'Skim':
original_swl = wfi.request['SiteWhitelist']
needs, task_name, running, idled = needs_action(wfi, task)
if (needs or needs_overide):
modifications[wfo.name][task.pathName] = {
'AddWhitelist': original_swl,
"Running": running,
"Pending": idled,
"Priority": wfi.request['RequestPriority']
}
altered_tasks.add(task.pathName)
wfi.sendLog(
'equalizor',
'%s of %s is running %d and pending %d, taking action : AddWhitelist \n %s'
% (task_name, wfo.name, running, idled,
json.dumps(sorted(original_swl), indent=2)))
if options.augment:
print sorted(wfi.request['SiteWhitelist']), i_task, use_HLT
### add the HLT at partner of CERN
if 'T2_CH_CERN' in wfi.request['SiteWhitelist'] and i_task in [
0, 1
] and use_HLT:
needs, task_name, running, idled = needs_action(wfi, task)
if options.augment: needs = True
needs = True
##needs = random.random()<0.40 remove the random, just add up to a limit
if (needs or needs_overide) and pending_HLT < max_HLT:
pending_HLT += idled
if task.pathName in modifications[
wfo.name] and 'AddWhitelist' in modifications[
wfo.name][task.pathName]:
modifications[wfo.name][task.pathName][
"AddWhitelist"].append("T2_CH_CERN_HLT")
print "\t", wfo.name, "adding addHLT up to", pending_HLT, "for", max_HLT
print task.pathName
## this Replace does not work at all for HLT
#elif task.pathName in modifications[wfo.name] and 'ReplaceSiteWhitelist' in modifications[wfo.name][task.pathName]:
#modifications[wfo.name][task.pathName]["ReplaceSiteWhitelist"].append( "T2_CH_CERN_HLT" )
#print "\t",wfo.name,"adding replace HLT up to",pending_HLT,"for",max_HLT
else:
modifications[wfo.name][task.pathName] = {
"AddWhitelist": ["T2_CH_CERN_HLT"],
"Priority": wfi.request['RequestPriority'],
"Running": running,
"Pending": idled
}
wfi.sendLog(
'equalizor',
'adding the HLT in whitelist of %s to %d for %d' %
(task.pathName, pending_HLT, max_HLT))
if i_task == 0 and not sec and use_T0:
needs, task_name, running, idled = needs_action(wfi, task)
if options.augment: needs = True
#needs = True
good_type = wfi.request['RequestType'] in [
'MonteCarlo', 'MonteCarloFromGEN'
]
read_lhe = ((not 'LheInputFiles' in wfi.request)
or bool(wfi.request['LheInputFiles']))
good_type &= not read_lhe
if not good_type and not options.augment: needs = False
##needs = random.random()<0.40 remove the random, just add up to a limit
if (needs or needs_overide):
pending_T0 += idled
if task.pathName in modifications[
wfo.name] and 'AddWhitelist' in modifications[
wfo.name][task.pathName]:
if not "T0_CH_CERN" in modifications[wfo.name][
task.pathName]["AddWhitelist"]:
modifications[wfo.name][task.pathName][
"AddWhitelist"].append("T0_CH_CERN")
wfi, sendLog(
'equalizor',
'adding the T0 for %s to %d for %d' %
(task.pathName, pending_T0, max_T0))
elif task.pathName in modifications[
wfo.
name] and 'ReplaceSiteWhitelist' in modifications[
wfo.name][task.pathName]:
if not "T0_CH_CERN" in modifications[wfo.name][
task.pathName]["ReplaceSiteWhitelist"]:
modifications[wfo.name][task.pathName][
"ReplaceSiteWhitelist"].append("T0_CH_CERN")
wfi, sendLog(
'equalizor',
'adding the T0 to replacement for %s to %d for %d'
% (task.pathName, pending_T0, max_T0))
else:
modifications[wfo.name][task.pathName] = {
"AddWhitelist": ["T0_CH_CERN"],
"Priority": wfi.request['RequestPriority'],
"Running": running,
"Pending": idled
}
wfi, sendLog(
'equalizor', 'adding the T0 for %s to %d for %d' %
(task.pathName, pending_T0, max_T0))
interface['modifications'].update(modifications)
### manage the number of core and job resizing
interface['cores'] = {
'T2_CH_CERN_HLT': {
'min': 4,
'max': 16
},
'default': {
'min': 1,
'max': 4
}
}
interface['resizes'] = ['RunIISpring16DR80']
### manage the modification of the memory and target time
interface['time'] = defaultdict(list)
interface['memory'] = defaultdict(list)
max_N_mem = 10
max_N_time = 10
## discretize the memory to 10 at most values
mems = set([o['memory'] for t, o in performance.items() if 'memory' in o])
times = set([o['time'] for t, o in performance.items() if 'time' in o])
if len(mems) > max_N_mem:
mem_step = int((max(mems) - min(mems)) / float(max_N_mem))
for t in performance:
if not 'memory' in performance[t]: continue
(m, r) = divmod(performance[t]['memory'], mem_step)
performance[t]['memory'] = (m + 1) * mem_step
if len(times) > max_N_time:
time_step = int((max(times) - min(times)) / float(max_N_time))
for t in performance:
if not 'time' in performance[t]: continue
(m, r) = divmod(performance[t]['time'], time_step)
performance[t]['time'] = (m + 1) * time_step
for t, o in performance.items():
if 'time' in o:
interface['time'][str(o['time'])].append(t)
if 'memory' in o:
interface['memory'][str(o['memory'])].append(t)
## close and save
close(interface)
def equalizor(url , specific = None, options=None):
up = componentInfo(mcm=False, soft=['mcm'])
if not up.check(): return
if not specific:
workflows = getWorkflows(url, status='running-closed', details=True)
workflows.extend(getWorkflows(url, status='running-open', details=True))
## start from scratch
modifications = defaultdict(dict)
## define regionality site => fallback allowed. feed on an ssb metric ??
mapping = defaultdict(list)
reversed_mapping = defaultdict(list)
regions = defaultdict(list)
SI = siteInfo()
CI = campaignInfo()
UC = unifiedConfiguration()
for site in SI.sites_ready:
region = site.split('_')[1]
if not region in ['US','DE','IT']: continue
regions[region] = [region]
def site_in_depletion(s):
return True
if s in SI.sites_pressure:
(m, r, pressure) = SI.sites_pressure[s]
if float(m) < float(r):
print s,m,r,"lacking pressure"
return True
else:
print s,m,r,"pressure"
pass
return False
for site in SI.sites_ready:
region = site.split('_')[1]
## fallback to the region, to site with on-going low pressure
mapping[site] = [fb for fb in SI.sites_ready if any([('_%s_'%(reg) in fb and fb!=site and site_in_depletion(fb))for reg in regions[region]]) ]
use_T0 = ('T0_CH_CERN' in UC.get("site_for_overflow"))
if options.t0: use_T0 = True
#if options.augment : use_T0 = True
use_HLT = ('T2_CH_CERN_HLT' in UC.get("site_for_overflow"))
if options.hlt: use_HLT = True
#if options.augment : use_HLT=True
if use_HLT:
mapping['T2_CH_CERN'].append('T2_CH_CERN_HLT')
if use_T0:
mapping['T2_CH_CERN'].append('T0_CH_CERN')
#mapping['T1_FR_CCIN2P3'].append('T0_CH_CERN')
#mapping['T2_IT_Legnaro'].append('T1_IT_CNAF')
for reg in ['IT','DE','UK']:
mapping['T2_CH_CERN'].extend([fb for fb in SI.sites_ready if '_%s_'%reg in fb])
## make them appear as OK to use
force_sites = []
## overflow CERN to underutilized T1s
upcoming = json.loads( open('%s/GQ.json'%monitor_dir).read())
for possible in SI.sites_T1s:
if not possible in upcoming:
mapping['T2_CH_CERN'].append(possible)
## remove add-hoc sites from overflow mapping
prevent_sites = ['T2_US_Purdue']
for prevent in prevent_sites:
if prevent in mapping: mapping.pop( prevent )
for src in mapping:
for prevent in prevent_sites:
if prevent in mapping[src]:
mapping[src].remove( prevent )
## create the reverse mapping for the condor module
for site,fallbacks in mapping.items():
for fb in fallbacks:
reversed_mapping[fb].append(site)
## this is the fallback mapping
print "Direct mapping : site => overflow"
print json.dumps( mapping, indent=2)
print "Reverse mapping : dest <= from origin"
print json.dumps( reversed_mapping, indent=2)
altered_tasks = set()
def running_idle( wfi , task_name):
gmon = wfi.getGlideMon()
#print gmon
if not gmon: return (0,0)
if not task_name in gmon: return (0,0)
return (gmon[task_name]['Running'], gmon[task_name]['Idle'])
def needs_action( wfi, task, min_idled = 100, pressure = 0.2):
task_name = task.pathName.split('/')[-1]
running, idled = running_idle( wfi, task_name)
go = True
if not idled and not running :
go = False
if idled < 100:
go = False
if (not running and idled) or (running and (idled / float(running) > pressure)):
go = True
else:
go = False
return go, task_name, running, idled
def getPerf( task ):
task = task.split('/')[1]+'/'+task.split('/')[-1]
try:
u = 'http://cms-gwmsmon.cern.ch/prodview/json/history/memoryusage720/%s'%task
print u
perf_data = json.loads(os.popen('curl -s --retry 5 %s'%u).read())
except Exception as e:
print str(e)
return (None,None)
buckets = perf_data['aggregations']["2"]['buckets']
s_m = sum( bucket['key']*bucket['doc_count'] for bucket in buckets)
w_m = sum( bucket['doc_count'] for bucket in buckets)
m_m = max( bucket['key'] for bucket in buckets) if buckets else None
b_m = None
if w_m > 100:
b_m = m_m
try:
perf_data = json.loads(os.popen('curl -s --retry 5 http://cms-gwmsmon.cern.ch/prodview/json/history/runtime720/%s'%task).read())
except Exception as e:
print str(e)
return (b_m,None)
buckets = perf_data['aggregations']["2"]['buckets']
s_t = sum( bucket['key']*bucket['doc_count'] for bucket in buckets)
w_t = sum( bucket['doc_count'] for bucket in buckets)
m_t = max( bucket['key'] for bucket in buckets) if buckets else None
b_t = None
if w_t > 100:
b_t = m_t
return (b_m,b_t)
def getcampaign( task ):
taskname = task.pathName.split('/')[-1]
if hasattr( task, 'prepID'):
return task.prepID.split('-')[1]
elif taskname.count('-')>=1:
return taskname.split('-')[1]
else:
return None
def close( interface ):
open('%s/equalizor.json.new'%monitor_dir,'w').write( json.dumps( interface, indent=2))
os.system('mv %s/equalizor.json.new %s/equalizor.json'%(monitor_dir,monitor_dir))
os.system('cp %s/equalizor.json %s/logs/equalizor/equalizor.%s.json'%(monitor_dir,monitor_dir,time.mktime(time.gmtime())))
interface = {
'reversed_mapping' : reversed_mapping,
'modifications' : {}
}
if options.augment or options.remove:
interface['modifications'] = json.loads( open('%s/equalizor.json'%monitor_dir).read())['modifications']
if options.remove:
if specific in interface['modifications']:
print "poping",specific
interface['modifications'].pop(specific)
close( interface )
return
PU_locations = {}
PU_overflow = {}
LHE_overflow = {}
tune_performance = []
pending_HLT = 0
max_HLT = 60000
pending_T0 = 0
max_T0 = 60000
try:
gmon = json.loads(os.popen('curl -s http://cms-gwmsmon.cern.ch/prodview/json/T2_CH_CERN_HLT').read())
pending_HLT += gmon["Running"]
pending_HLT += gmon["MatchingIdle"]
except:
pass
stay_within_site_whitelist = False
specific_task=None
if specific and ":" in specific:
specific,specific_task = specific.split(':')
if specific:
wfs = session.query(Workflow).filter(Workflow.name.contains(specific)).all()
else:
wfs = session.query(Workflow).filter(Workflow.status == 'away').all()
performance = {}
no_routing = [
]
random.shuffle( wfs )
for wfo in wfs:
if wfo.name in no_routing and not options.augment:
continue
if specific and not specific in wfo.name:
continue
if specific:
wfi = workflowInfo(url, wfo.name)
else:
cached = filter(lambda d : d['RequestName']==wfo.name, workflows)
if not cached : continue
wfi = workflowInfo(url, wfo.name, request = cached[0])
## only running should get re-routed
if not wfi.request['RequestStatus'] in ['running-open','running-closed'] and not specific: continue
tasks_and_campaigns = []
for task in wfi.getWorkTasks():
tasks_and_campaigns.append( (task, getcampaign(task) ) )
_,_,_,sec = wfi.getIO()
## check needs override
needs_overide = False
if not needs_overide and options.augment: needs_overide=True
def overide_from_agent( wfi, needs_overide):
bad_agents = []#'http://cmssrv219.fnal.gov:5984']
if not bad_agents: return needs_overide
if needs_overide: return True
agents = wfi.getAgents()
wqss = ['Running','Acquired']
if any([agent in agents.get(wqs,{}).keys() for wqs,agent in itertools.product( wqss, bad_agents)]):
print "overriding the need for bad agent"
needs_overide = True
return needs_overide
## now parse this for action
for i_task,(task,campaign) in enumerate(tasks_and_campaigns):
if options.augment:
print task.pathName
print campaign
tune = CI.get(campaign,'tune',options.tune)
if tune and not campaign in tune_performance:
tune_performance.append( campaign )
overflow = CI.get(campaign,'overflow',{})
if overflow:
if "PU" in overflow and not campaign in PU_overflow:
PU_overflow[campaign] = copy.deepcopy(overflow['PU'])
print "adding",campaign,"to PU overflow rules"
if "LHE" in overflow and not campaign in LHE_overflow:
print "adding",campaign,"to light input overflow rules"
site_list = overflow['LHE']['site_list']
LHE_overflow[campaign] = copy.deepcopy( getattr(SI,site_list) )
### get the task performance, for further massaging.
if campaign in tune_performance or options.tune:
print "performance",task.taskType,task.pathName
if task.taskType in ['Processing','Production']:
set_memory,set_time = getPerf( task.pathName )
#print "Performance %s GB %s min"%( set_memory,set_time)
wfi.sendLog('equalizor','Performance tuning to %s GB %s min'%( set_memory,set_time))
## get values from gmwsmon
# massage the values : 95% percentile
performance[task.pathName] = {}
if set_memory:
performance[task.pathName]['memory']=set_memory
if set_time and False:
performance[task.pathName]['time'] = set_time
### rule to avoid the issue of taskchain secondary jobs being stuck at sites processing the initial step
if campaign in LHE_overflow:
if task.taskType in ['Processing']:
needs, task_name, running, idled = needs_action(wfi, task)
needs_overide = overide_from_agent( wfi, needs_overide)
extend_to = list(set(copy.deepcopy( LHE_overflow[campaign] )))
if stay_within_site_whitelist:
extend_to = list(set(extend_to) & set(wfi.request['SiteWhitelist'])) ## restrict to stupid-site-whitelist
extend_to = list(set(extend_to) & set(SI.sites_ready + force_sites))
if extend_to and needs or needs_overide:
modifications[wfo.name][task.pathName] = { "ReplaceSiteWhitelist" : extend_to ,"Running" : running, "Pending" : idled, "Priority" : wfi.request['RequestPriority']}
wfi.sendLog('equalizor','%s of %s is running %d and pending %d, taking action : ReplaceSiteWhitelist \n %s'%( task_name,
wfo.name,
running,
idled ,
json.dumps( sorted(modifications[wfo.name][task.pathName]['ReplaceSiteWhitelist']))))
altered_tasks.add( task.pathName )
else:
wfi.sendLog('equalizor','%s of %s is running %d and pending %d'%( task_name, wfo.name, running, idled))
### overflow the 76 digi-reco to the site holding the pileup
if campaign in PU_overflow:
force = PU_overflow[campaign]['force'] if 'force' in PU_overflow[campaign] else False
secondary_locations = set(SI.sites_ready + force_sites)
for s in sec:
if not s in PU_locations:
presence = getDatasetPresence( url, s)
#one_secondary_locations = [site for (site,(there,frac)) in presence.items() if there]
one_secondary_locations = [site for (site,(there,frac)) in presence.items() if frac>98.]
PU_locations[s] = one_secondary_locations
print "secondary is at",sorted(PU_locations[s])
secondary_locations = set([SI.SE_to_CE(site) for site in PU_locations[s]]) & secondary_locations
## we should add all sites that hold the secondary input if any
### given that we have the secondary location available, it is not necessary to use the add-hoc list
##secondary_locations = list(set(PU_overflow[campaign]['sites']) & set( SI.sites_ready ))
if any([task.pathName.endswith(finish) for finish in ['_0','StepOneProc','Production']]) :
needs, task_name, running, idled = needs_action(wfi, task)
## removing the ones in the site whitelist already since they encode the primary input location
if stay_within_site_whitelist:
original_site_in_use = set(wfi.request['SiteWhitelist'])
else:
original_site_in_use = set(secondary_locations)
## remove the sites that have already running jobs
gmon = wfi.getGlideMon()
if gmon and task_name in gmon and 'Sites' in gmon[task_name]:
site_in_use = set(gmon[task_name]['Sites'])
print "removing",sorted(site_in_use)
## that determines where you want to run in addition
augment_by = list((set(secondary_locations)- site_in_use) & original_site_in_use)
else:
print "no existing running site"
augment_by = list(original_site_in_use)
needs_overide = overide_from_agent( wfi, needs_overide)
if augment_by and (needs or needs_overide or force) and PU_overflow[campaign]['pending'] < PU_overflow[campaign]['max']:
PU_overflow[campaign]['pending'] += idled
print "raising overflow to",PU_overflow[campaign]['pending'],"for",PU_overflow[campaign]['max']
## the step with an input ought to be the digi part : make this one go anywhere
modifications[wfo.name][task.pathName] = { "AddWhitelist" : augment_by , "Running" : running, "Pending" : idled, "Priority" : wfi.request['RequestPriority']}
altered_tasks.add( task.pathName )
wfi.sendLog('equalizor','%s of %s is running %d and pending %d, taking action : AddWhitelist \n %s'%( task_name, wfo.name,
running, idled,
json.dumps( sorted(augment_by), indent=2 )))
else:
print task_name,"of",wfo.name,"running",running,"and pending",idled
### overflow the skims back to multi-core
if campaign in ['Run2015D','Run2015C_25ns'] and task.taskType =='Skim':
original_swl = wfi.request['SiteWhitelist']
needs, task_name, running, idled = needs_action(wfi, task)
if (needs or needs_overide):
modifications[wfo.name][task.pathName] = { 'AddWhitelist' : original_swl,
"Running" : running, "Pending" : idled, "Priority" : wfi.request['RequestPriority']}
altered_tasks.add( task.pathName )
wfi.sendLog('equalizor','%s of %s is running %d and pending %d, taking action : AddWhitelist \n %s'%( task_name, wfo.name,
running, idled,
json.dumps( sorted(original_swl), indent=2 )))
if options.augment:
print sorted(wfi.request['SiteWhitelist']),i_task,use_HLT
### add the HLT at partner of CERN
if 'T2_CH_CERN' in wfi.request['SiteWhitelist'] and i_task in [0,1] and use_HLT:
needs, task_name, running, idled = needs_action(wfi, task)
if options.augment: needs=True
needs = True
##needs = random.random()<0.40 remove the random, just add up to a limit
if (needs or needs_overide) and pending_HLT < max_HLT:
pending_HLT += idled
if task.pathName in modifications[wfo.name] and 'AddWhitelist' in modifications[wfo.name][task.pathName]:
modifications[wfo.name][task.pathName]["AddWhitelist"].append( "T2_CH_CERN_HLT" )
print "\t",wfo.name,"adding addHLT up to",pending_HLT,"for",max_HLT
print task.pathName
## this Replace does not work at all for HLT
#elif task.pathName in modifications[wfo.name] and 'ReplaceSiteWhitelist' in modifications[wfo.name][task.pathName]:
#modifications[wfo.name][task.pathName]["ReplaceSiteWhitelist"].append( "T2_CH_CERN_HLT" )
#print "\t",wfo.name,"adding replace HLT up to",pending_HLT,"for",max_HLT
else:
modifications[wfo.name][task.pathName] = { "AddWhitelist" : ["T2_CH_CERN_HLT"],
"Priority" : wfi.request['RequestPriority'],
"Running" : running,
"Pending" : idled}
wfi.sendLog('equalizor','adding the HLT in whitelist of %s to %d for %d'%( task.pathName, pending_HLT, max_HLT))
if i_task==0 and not sec and use_T0:
needs, task_name, running, idled = needs_action(wfi, task)
if options.augment: needs=True
#needs = True
good_type = wfi.request['RequestType'] in ['MonteCarlo','MonteCarloFromGEN']
read_lhe = ((not 'LheInputFiles' in wfi.request) or bool(wfi.request['LheInputFiles']))
good_type &= not read_lhe
if not good_type and not options.augment: needs = False
##needs = random.random()<0.40 remove the random, just add up to a limit
if (needs or needs_overide):
pending_T0 += idled
if task.pathName in modifications[wfo.name] and 'AddWhitelist' in modifications[wfo.name][task.pathName]:
if not "T0_CH_CERN" in modifications[wfo.name][task.pathName]["AddWhitelist"]:
modifications[wfo.name][task.pathName]["AddWhitelist"].append( "T0_CH_CERN" )
wfi,sendLog('equalizor','adding the T0 for %s to %d for %d'%( task.pathName, pending_T0, max_T0))
elif task.pathName in modifications[wfo.name] and 'ReplaceSiteWhitelist' in modifications[wfo.name][task.pathName]:
if not "T0_CH_CERN" in modifications[wfo.name][task.pathName]["ReplaceSiteWhitelist"]:
modifications[wfo.name][task.pathName]["ReplaceSiteWhitelist"].append( "T0_CH_CERN" )
wfi,sendLog('equalizor','adding the T0 to replacement for %s to %d for %d'%( task.pathName, pending_T0, max_T0))
else:
modifications[wfo.name][task.pathName] = { "AddWhitelist" : ["T0_CH_CERN"],
"Priority" : wfi.request['RequestPriority'],
"Running" : running,
"Pending" : idled}
wfi,sendLog('equalizor','adding the T0 for %s to %d for %d'%( task.pathName, pending_T0, max_T0))
interface['modifications'].update( modifications )
### manage the number of core and job resizing
interface['cores']={'T2_CH_CERN_HLT': {'min':4,'max':16}, 'default': {'min':1, 'max':4}}
interface['resizes'] = ['RunIISpring16DR80']
### manage the modification of the memory and target time
interface['time'] = defaultdict(list)
interface['memory'] = defaultdict(list)
max_N_mem = 10
max_N_time = 10
## discretize the memory to 10 at most values
mems = set([o['memory'] for t,o in performance.items() if 'memory' in o])
times = set([o['time'] for t,o in performance.items() if 'time' in o])
if len(mems)>max_N_mem:
mem_step = int((max(mems) - min(mems))/ float(max_N_mem))
for t in performance:
if not 'memory' in performance[t]: continue
(m,r) = divmod(performance[t]['memory'], mem_step)
performance[t]['memory'] = (m+1)*mem_step
if len(times)>max_N_time:
time_step = int((max(times) - min(times))/float(max_N_time))
for t in performance:
if not 'time' in performance[t]: continue
(m,r) = divmod(performance[t]['time'], time_step)
performance[t]['time'] = (m+1)*time_step
for t,o in performance.items():
if 'time' in o:
interface['time'][str(o['time'])] .append( t )
if 'memory' in o:
interface['memory'][str(o['memory'])].append( t )
## close and save
close( interface )
def checkor(url, spec=None, options=None):
fDB = falseDB()
wfs=[]
if options.fetch:
#workflows = getWorkflows(url, status='completed')
#for wf in workflows:
# wfo = session.query(Workflow).filter(Workflow.name == wf ).first()
# if wfo:
# if not wfo.status in ['away','assistance']: continue
# wfs.append(wfo )
wfs.extend( session.query(Workflow).filter(Workflow.status == 'away').all() )
wfs.extend( session.query(Workflow).filter(Workflow.status== 'assistance').all() )
else:
## than get all in need for assistance
wfs.extend( session.query(Workflow).filter(Workflow.status.startswith('assistance-')).all() )
custodials = defaultdict(list) #sites : dataset list
transfers = defaultdict(list) #sites : dataset list
invalidations = [] #a list of files
SI = siteInfo()
CI = campaignInfo()
def get_campaign(output, wfi):
campaign = None
try:
campaign = output.split('/')[2].split('-')[0]
except:
if 'Campaign' in wfi.request:
campaign = wfi.request['Campaign']
return campaign
for wfo in wfs:
if spec and not (spec in wfo.name): continue
## get info
wfi = workflowInfo(url, wfo.name)
## make sure the wm status is up to date.
# and send things back/forward if necessary.
wfo.wm_status = wfi.request['RequestStatus']
if wfo.wm_status == 'closed-out':
## manually closed-out
print wfo.name,"is already",wfo.wm_status
wfo.status = 'close'
session.commit()
continue
elif wfo.wm_status in ['failed','aborted','aborted-archived','rejected','rejected-archived']:
## went into trouble
wfo.status = 'trouble'
print wfo.name,"is in trouble",wfo.wm_status
session.commit()
continue
elif wfo.wm_status in ['assigned','acquired']:
## not worth checking yet
print wfo.name,"not running yet"
session.commit()
continue
if wfo.wm_status != 'completed':
## for sure move on with closeout check if in completed
print "no need to check on",wfo.name,"in status",wfo.wm_status
session.commit()
continue
session.commit()
sub_assistance="" # if that string is filled, there will be need for manual assistance
is_closing = True
## do the closed-out checks one by one
# tuck out DQMIO/DQM
wfi.request['OutputDatasets'] = [ out for out in wfi.request['OutputDatasets'] if not '/DQM' in out]
## anything running on acdc
familly = getWorkflowById(url, wfi.request['PrepID'], details=True)
acdc = []
for member in familly:
if member['RequestName'] == wfo.name: continue
if member['RequestDate'] < wfi.request['RequestDate']: continue
if member['RequestType'] != 'Resubmission': continue
if member['RequestStatus'] in ['running-opened','running-closed','assignment-approved','assigned','acquired']:
print wfo.name,"still has an ACDC running",member['RequestName']
acdc.append( member['RequestName'] )
#print json.dumps(member,indent=2)
## hook for just waiting ...
is_closing = False
## completion check
percent_completions = {}
event_expected,lumi_expected = wfi.request['TotalInputEvents'],wfi.request['TotalInputLumis']
fractions_pass = {}
for output in wfi.request['OutputDatasets']:
event_count,lumi_count = getDatasetEventsAndLumis(dataset=output)
percent_completions[output] = 0.
if lumi_expected:
percent_completions[output] = lumi_count / float( lumi_expected )
fractions_pass[output] = 0.95
c = get_campaign(output, wfi)
if c in CI.campaigns and 'fractionpass' in CI.campaigns[c]:
fractions_pass[output] = CI.campaigns[c]['fractionpass']
print "overriding fraction to",fractions_pass[output],"for",output
if options.fractionpass:
fractions_pass[output] = options.fractionpass
print "overriding fraction to",fractions_pass[output],"by command line for",output
if not all([percent_completions[out] > fractions_pass[out] for out in fractions_pass]):
print wfo.name,"is not completed"
print json.dumps(percent_completions, indent=2)
print json.dumps(fractions_pass, indent=2)
## hook for creating automatically ACDC ?
sub_assistance+='-recovery'
is_closing = False
## correct lumi < 300 event per lumi
events_per_lumi = {}
for output in wfi.request['OutputDatasets']:
events_per_lumi[output] = getDatasetEventsPerLumi( output )
lumi_upper_limit = {}
for output in wfi.request['OutputDatasets']:
upper_limit = 300.
campaign = get_campaign(output, wfi)
if campaign in CI.campaigns and 'lumisize' in CI.campaigns[campaign]:
upper_limit = CI.campaigns[campaign]['lumisize']
print "overriding the upper lumi size to",upper_limit,"for",campaign
if options.lumisize:
upper_limit = options.lumisize
print "overriding the upper lumi size to",upper_limit,"by command line"
lumi_upper_limit[output] = upper_limit
if any([ events_per_lumi[out] > lumi_upper_limit[out] for out in events_per_lumi]):
print wfo.name,"has big lumisections"
print json.dumps(events_per_lumi, indent=2)
## hook for rejecting the request ?
sub_assistance+='-biglumi'
is_closing = False
any_presence = {}
for output in wfi.request['OutputDatasets']:
any_presence[output] = getDatasetPresence(url, output, vetoes=[])
## custodial copy
custodial_locations = {}
custodial_presences = {}
for output in wfi.request['OutputDatasets']:
custodial_presences[output] = [s for s in any_presence[output] if 'MSS' in s]
custodial_locations[output] = phedexClient.getCustodialSubscriptionRequestSite(output)
if not custodial_locations[output]:
custodial_locations[output] = []
vetoed_custodial_tier = ['MINIAODSIM']
out_worth_checking = [out for out in custodial_locations.keys() if out.split('/')[-1] not in vetoed_custodial_tier]
if not all(map( lambda sites : len(sites)!=0, [custodial_locations[out] for out in out_worth_checking])):
print wfo.name,"has not all custodial location"
print json.dumps(custodial_locations, indent=2)
##########
## hook for making a custodial replica ?
custodial = None
## get from other outputs
for output in out_worth_checking:
if len(custodial_locations[output]):
custodial = custodial_locations[output][0]
## try to get it from campaign configuration
if not custodial:
for output in out_worth_checking:
campaign = get_campaign(output, wfi)
if campaign in CI.campaigns and 'custodial' in CI.campaigns[campaign]:
custodial = CI.campaigns[campaign]['custodial']
print "Setting custodial to",custodial,"from campaign configuration"
break
## get from the parent
if not custodial and 'InputDataset' in wfi.request:
parents_custodial = findCustodialLocation(url, wfi.request['InputDataset'])
if len(parents_custodial):
custodial = parents_custodial[0]
else:
print "the input dataset",wfi.request['InputDataset'],"does not have custodial in the first place. abort"
continue
if not custodial:
## pick one at random
custodial = SI.pick_SE()
if custodial and not sub_assistance and not acdc:
## register the custodial request, if there are no other big issues
for output in out_worth_checking:
if not len(custodial_locations[output]):
custodials[custodial].append( output )
else:
print "cannot find a custodial for",wfo.name
is_closing = False
## disk copy
disk_copies = {}
for output in wfi.request['OutputDatasets']:
disk_copies[output] = [s for s in any_presence[output] if (not 'MSS' in s) and (not 'Buffer' in s)]
if not all(map( lambda sites : len(sites)!=0, disk_copies.values())):
print wfo.name,"has not all output on disk"
print json.dumps(disk_copies, indent=2)
## presence in dbs
dbs_presence = {}
dbs_invalid = {}
for output in wfi.request['OutputDatasets']:
dbs_presence[output] = dbs3Client.getFileCountDataset( output )
dbs_invalid[output] = dbs3Client.getFileCountDataset( output, onlyInvalid=True)
## presence in phedex
phedex_presence ={}
for output in wfi.request['OutputDatasets']:
phedex_presence[output] = phedexClient.getFileCountDataset(url, output )
fraction_invalid = 0.01
if not all([dbs_presence[out] == (dbs_invalid[out]+phedex_presence[out]) for out in wfi.request['OutputDatasets']]) and not options.ignorefiles:
print wfo.name,"has a dbs,phedex mismatch"
print json.dumps(dbs_presence, indent=2)
print json.dumps(dbs_invalid, indent=2)
print json.dumps(phedex_presence, indent=2)
## hook for just waiting ...
is_closing = False
if not all([(dbs_invalid[out] <= int(fraction_invalid*dbs_presence[out])) for out in wfi.request['OutputDatasets']]) and not options.ignorefiles:
print wfo.name,"has a dbs invalid file level too high"
print json.dumps(dbs_presence, indent=2)
print json.dumps(dbs_invalid, indent=2)
print json.dumps(phedex_presence, indent=2)
## need to be going and taking an eye
sub_assistance+="-invalidfiles"
is_closing = False
## put that heavy part at the end
## duplication check
duplications = {}
if is_closing:
for output in wfi.request['OutputDatasets']:
try:
duplications[output] = dbs3Client.duplicateRunLumi( output )
except:
try:
duplications[output] = dbs3Client.duplicateRunLumi( output )
except:
print "was not possible to get the duplicate count for",output
is_closing=False
if any(duplications.values()) and not options.ignoreduplicates:
print wfo.name,"has duplicates"
print json.dumps(duplications,indent=2)
## hook for making file invalidation ?
sub_assistance+='-duplicates'
is_closing = False
## for visualization later on
if not wfo.name in fDB.record:
#print "adding",wfo.name,"to close out record"
fDB.record[wfo.name] = {
'datasets' :{},
'name' : wfo.name,
'closeOutWorkflow' : None,
}
fDB.record[wfo.name]['closeOutWorkflow'] = is_closing
for output in wfi.request['OutputDatasets']:
if not output in fDB.record[wfo.name]['datasets']: fDB.record[wfo.name]['datasets'][output] = {}
rec = fDB.record[wfo.name]['datasets'][output]
rec['percentage'] = float('%.2f'%(percent_completions[output]*100))
rec['duplicate'] = duplications[output] if output in duplications else 'N/A'
rec['phedexReqs'] = float('%.2f'%any_presence[output][custodial_presences[output][0]][1]) if len(custodial_presences[output])!=0 else 'N/A'
rec['closeOutDataset'] = is_closing
rec['transPerc'] = float('%.2f'%any_presence[output][ disk_copies[output][0]][1]) if len(disk_copies[output])!=0 else 'N/A'
rec['correctLumis'] = int(events_per_lumi[output]) if (events_per_lumi[output] > lumi_upper_limit[output]) else True
rec['missingSubs'] = False if len(custodial_locations[output])==0 else ','.join(list(set(custodial_locations[output])))
rec['dbsFiles'] = dbs_presence[output]
rec['dbsInvFiles'] = dbs_invalid[output]
rec['phedexFiles'] = phedex_presence[output]
rec['acdc'] = len(acdc)
## and move on
if is_closing:
## toggle status to closed-out in request manager
print "setting",wfo.name,"closed-out"
if not options.test:
reqMgrClient.closeOutWorkflowCascade(url, wfo.name)
# set it from away/assistance* to close
wfo.status = 'close'
session.commit()
else:
print wfo.name,"needs assistance"
## that means there is something that needs to be done acdc, lumi invalidation, custodial, name it
wfo.status = 'assistance'+sub_assistance
if not options.test:
print "setting",wfo.name,"to",wfo.status
session.commit()
fDB.summary()
## custodial requests
print "Custodials"
print json.dumps(custodials, indent=2)
for site in custodials:
print ','.join(custodials[site]),'=>',site
if not options.test:
result = makeReplicaRequest(url, site, list(set(custodials[site])),"custodial copy at production close-out",custodial='y',priority='low')
print result
print "Transfers"
print json.dumps(transfers, indent=2)
## replicas requests
for site in transfers:
print ','.join(transfers[site]),'=>',site
if not options.test:
result = None
#result = makeReplicaRequest(url, site, list(set(transfers[site])),"copy to disk at production close-out")
print result
print "File Invalidation"
print invalidations
def closor(url, specific=None, options=None):
if userLock(): return
if duplicateLock(): return
if not componentInfo().check(): return
UC = unifiedConfiguration()
CI = campaignInfo()
all_late_files = []
check_fullcopy_to_announce = UC.get('check_fullcopy_to_announce')
jump_the_line = options.announce if options else False
if jump_the_line:
print "announce option is on. Checking on things on-going ready to be announced"
wfs = session.query(Workflow).filter(
Workflow.status.contains('announce')).filter(
sqlalchemy.not_(Workflow.status.contains('announced'))).all()
else:
print "regular option. Checking on things done and to be announced"
wfs = session.query(Workflow).filter(Workflow.status == 'close').all()
wfs_n = [w.name for w in wfs]
print "unique names?"
print len(set(wfs_n)) == len(wfs_n)
held = set()
print len(wfs), "closing"
random.shuffle(wfs)
max_per_round = UC.get('max_per_round').get('closor', None)
if options.limit: max_per_round = options.limit
if max_per_round:
## order them by priority
all_closedout = sorted(getWorkflows(url, 'closed-out', details=True),
key=lambda r: r['RequestPriority'])
all_closedout = [r['RequestName'] for r in all_closedout]
def rank(wfn):
return all_closedout.index(wfn) if wfn in all_closedout else 0
wfs = sorted(wfs, key=lambda wfo: rank(wfo.name), reverse=True)
wfs = wfs[:max_per_round]
batch_go = {}
batch_warnings = defaultdict(set)
batch_goodness = UC.get("batch_goodness")
for iwfo, wfo in enumerate(wfs):
if specific and not specific in wfo.name: continue
print "Progress [%d/%d]" % (iwfo, len(wfs))
## what is the expected #lumis
wfi = workflowInfo(url, wfo.name)
wfo.wm_status = wfi.request['RequestStatus']
if wfi.isRelval():
has_batch_go = False
batch_name = wfi.getCampaign()
if not batch_name in batch_go:
## do the esimatation whethere this can be announced : only once per batch
in_batches = getWorkflowByCampaign(url,
batch_name,
details=True)
batch_go[batch_name] = all(
map(
lambda s: not s in [
'completed', 'running-open', 'running-closed',
'acquired', 'assigned', 'assignment-approved'
], [r['RequestStatus'] for r in in_batches]))
## already verified
has_batch_go = batch_go[batch_name]
if not has_batch_go:
wfi.sendLog(
'closor',
'Cannot close for now because the batch <a href=https://dmytro.web.cern.ch/dmytro/cmsprodmon/workflows.php?campaign=%s>%s</a> is not all close'
% (batch_name, batch_name))
continue
if wfi.request['RequestStatus'] in ['announced', 'normal-archived'
] and not options.force:
## manually announced ??
wfo.status = 'done'
wfo.wm_status = wfi.request['RequestStatus']
wfi.sendLog(
'closor',
'%s is announced already : %s' % (wfo.name, wfo.wm_status))
session.commit()
if jump_the_line:
wfi.sendLog('closor', 'Announcing while completing')
expected_lumis = 1
if not 'TotalInputLumis' in wfi.request:
print wfo.name, "has not been assigned yet, or the database is corrupted"
elif wfi.request['TotalInputLumis'] == 0:
print wfo.name, "is corrupted with 0 expected lumis"
else:
expected_lumis = wfi.request['TotalInputLumis']
## what are the outputs
outputs = wfi.request['OutputDatasets']
## check whether the number of lumis is as expected for each
all_OK = defaultdict(lambda: False)
stats = defaultdict(int)
#print outputs
if len(outputs):
print wfo.name, wfi.request['RequestStatus']
for out in outputs:
event_count, lumi_count = getDatasetEventsAndLumis(dataset=out)
odb = session.query(Output).filter(
Output.datasetname == out).first()
if not odb:
print "adding an output object", out
odb = Output(datasetname=out)
odb.workflow = wfo
session.add(odb)
odb.nlumis = lumi_count
odb.nevents = event_count
odb.workfow_id = wfo.id
if odb.expectedlumis < expected_lumis:
odb.expectedlumis = expected_lumis
else:
expected_lumis = odb.expectedlumis
odb.date = time.mktime(time.gmtime())
session.commit()
fraction = lumi_count / float(expected_lumis) * 100.
completion_line = "%60s %d/%d = %3.2f%%" % (
out, lumi_count, expected_lumis, fraction)
wfi.sendLog('closor', "\t%s" % completion_line)
if wfi.isRelval() and fraction < batch_goodness:
batch_warnings[wfi.getCampaign()].add(completion_line)
stats[out] = lumi_count
all_OK[out] = True
## check for at least one full copy prior to moving on
in_full = {}
for out in outputs:
in_full[out] = []
presence = getDatasetPresence(url, out)
where = [site for site, info in presence.items() if info[0]]
if where:
all_OK[out] = True
print out, "is in full at", ",".join(where)
in_full[out] = copy.deepcopy(where)
else:
going_to = wfi.request['NonCustodialSites'] + wfi.request[
'CustodialSites']
wfi.sendLog(
'closor', "%s is not in full anywhere. send to %s" %
(out, ",".join(sorted(going_to))))
at_destination = dict([(k, v) for (k, v) in presence.items()
if k in going_to])
else_where = dict([(k, v) for (k, v) in presence.items()
if not k in going_to])
print json.dumps(at_destination)
print json.dumps(else_where, indent=2)
## do the full stuck transfer study, missing files and shit !
for there in going_to:
late_info = findLateFiles(url, out, going_to=there)
for l in late_info:
l.update({"workflow": wfo.name, "dataset": out})
all_late_files.extend(late_info)
if check_fullcopy_to_announce:
## only set this false if the check is relevant
all_OK[out] = False
## verify if we have to do harvesting
if not options.no_harvest and not jump_the_line:
(OK, requests) = spawn_harvesting(url, wfi, in_full)
all_OK.update(OK)
## only that status can let me go into announced
if all(all_OK.values()) and (
(wfi.request['RequestStatus'] in ['closed-out']) or options.force
or jump_the_line):
print wfo.name, "to be announced"
results = []
if not results:
for out in outputs:
if out in stats and not stats[out]:
continue
_, dsn, process_string, tier = out.split('/')
if all_OK[out]:
results.append(setDatasetStatus(out, 'VALID'))
if all_OK[out] and wfi.isRelval():
## make the specific relval rules and the replicas
## figure the destination(s) out
destinations = set()
if tier != "RECO" and tier != "ALCARECO":
destinations.add('T2_CH_CERN')
if tier == "GEN-SIM":
destinations.add('T1_US_FNAL_Disk')
if tier == "GEN-SIM-DIGI-RAW":
destinations.add('T1_US_FNAL_Disk')
if tier == "GEN-SIM-RECO":
destinations.add('T1_US_FNAL_Disk')
if "RelValTTBar" in dsn and "TkAlMinBias" in process_string and tier != "ALCARECO":
destinations.add('T2_CH_CERN')
if "MinimumBias" in dsn and "SiStripCalMinBias" in process_string and tier != "ALCARECO":
destinations.add('T2_CH_CERN')
if destinations:
wfi.sendLog(
'closor', '%s to go to %s' %
(out, ', '.join(sorted(destinations))))
## call to makereplicarequest under relval => done
for site in destinations:
result = makeReplicaRequest(
url,
site, [out],
'Copy for release validation consumption',
priority='normal',
approve=True,
mail=False,
group='RelVal')
try:
request_id = result['phedex'][
'request_created'][0]['id']
results.append(True)
except:
results.append('Failed relval transfer')
elif all_OK[out]:
campaign = None
try:
campaign = out.split('/')[2].split('-')[0]
except:
if 'Campaign' in wfi.request and wfi.request[
'Campaign']:
campaign = wfi.request['Campaign']
to_DDM = False
## campaign override
if campaign and campaign in CI.campaigns and 'toDDM' in CI.campaigns[
campaign] and tier in CI.campaigns[campaign][
'toDDM']:
to_DDM = True
## by typical enabling
if tier in UC.get("tiers_to_DDM"):
to_DDM = True
## check for unitarity
if not tier in UC.get("tiers_no_DDM") + UC.get(
"tiers_to_DDM"):
print "tier", tier, "neither TO or NO DDM for", out
results.append('Not recognitized tier %s' % tier)
#sendEmail("failed DDM injection","could not recognize %s for injecting in DDM"% out)
sendLog(
'closor',
"could not recognize %s for injecting in DDM" %
out,
level='critical')
continue
n_copies = 1
destinations = []
if to_DDM and campaign and campaign in CI.campaigns and 'DDMcopies' in CI.campaigns[
campaign]:
ddm_instructions = CI.campaigns[campaign][
'DDMcopies']
if type(ddm_instructions) == int:
n_copies = CI.campaigns[campaign]['DDMcopies']
elif type(ddm_instructions) == dict:
## a more fancy configuration
for ddmtier, indication in ddm_instructions.items(
):
if ddmtier == tier or ddmtier in [
'*', 'all'
]:
## this is for us
if 'N' in indication:
n_copies = indication['N']
if 'host' in indication:
destinations = indication['host']
destination_spec = ""
if destinations:
destination_spec = "--destination=" + ",".join(
destinations)
group_spec = "" ## not used yet
### should make this a campaign configuration
## inject to DDM when necessary
if to_DDM:
print "Sending", out, " to DDM"
status = pass_to_dynamo(
[out],
N=n_copies,
sites=destinations if destinations else None,
group=group_spec if group_spec else None)
results.append(status)
if status == True:
wfi.sendLog(
'closor',
'%s is send to dynamo in %s copies %s %s' %
(out, n_copies, sorted(destinations),
group_spec))
else:
sendLog('closor',
"could not add " + out +
" to dynamo pool. check closor logs.",
level='critical')
wfi.sendLog(
'closor', "could not add " + out +
" to dynamo pool. check closor logs.")
else:
print wfo.name, "no stats for announcing", out
results.append('No Stats')
if all(
map(lambda result: result in ['None', None, True],
results)):
if not jump_the_line:
## only announce if all previous are fine
res = reqMgrClient.announceWorkflowCascade(
url, wfo.name)
if not res in ['None', None]:
## check the status again, it might well have toggled
wl_bis = workflowInfo(url, wfo.name)
wfo.wm_status = wl_bis.request['RequestStatus']
session.commit()
if wl_bis.request['RequestStatus'] in [
'announced', 'normal-archived'
]:
res = None
else:
## retry ?
res = reqMgrClient.announceWorkflowCascade(
url, wfo.name)
results.append(res)
#print results
if all(map(lambda result: result in ['None', None, True],
results)):
if jump_the_line:
if not 'announced' in wfo.status:
wfo.status = wfo.status.replace(
'announce', 'announced')
else:
wfo.status = 'done'
session.commit()
wfi.sendLog('closor', "workflow outputs are announced")
else:
wfi.sendLog(
'closor', "Error with %s to be announced \n%s" %
(wfo.name, json.dumps(results)))
elif wfi.request['RequestStatus'] in [
'failed', 'aborted', 'aborted-archived', 'rejected',
'rejected-archived', 'aborted-completed'
]:
if wfi.isRelval():
wfo.status = 'forget'
wfo.wm_status = wfi.request['RequestStatus']
wfi.sendLog(
'closor',
"%s is %s, but will not be set in trouble to find a replacement."
% (wfo.name, wfo.wm_status))
else:
wfo.status = 'trouble'
wfo.wm_status = wfi.request['RequestStatus']
session.commit()
else:
print wfo.name, "not good for announcing:", wfi.request[
'RequestStatus']
wfi.sendLog('closor', "cannot be announced")
held.add(wfo.name)
days_late = 0.
retries_late = 10
really_late_files = [
info for info in all_late_files if info['retries'] >= retries_late
]
really_late_files = [
info for info in really_late_files
if info['delay'] / (60 * 60 * 24.) >= days_late
]
if really_late_files:
subject = 'These %d files are lagging for %d days and %d retries announcing dataset \n%s' % (
len(really_late_files), days_late, retries_late,
json.dumps(really_late_files, indent=2))
#sendEmail('waiting for files to announce', subject)
sendLog('closor', subject, level='warning')
sendLog('closor', subject)
print subject
open('%s/stuck_files.json' % monitor_dir,
'w').write(json.dumps(really_late_files, indent=2))
if held:
sendLog('closor',
"the workflows below are held up \n%s" %
("\n".join(sorted(held))),
level='critical')
#batches = json.loads(open('batches.json').read())
for bname, go in batch_go.items():
if go:
subject = "Release Validation Samples Batch %s" % bname
issues = ""
if batch_warnings[bname]:
issues = "The following datasets have outstanding completion (<%d%%) issues:\n\n" % batch_goodness
issues += "\n".join(sorted(batch_warnings[bname]))
issues += "\n\n"
text = """
Dear all,
a batch of release validation workflows has finished.
Batch ID:
%s
Detail of the workflows
https://dmytro.web.cern.ch/dmytro/cmsprodmon/requests.php?campaign=%s
%s
This is an automated message.
""" % (bname, bname, issues)
to = ['*****@*****.**']
sendEmail(subject, text, destination=to)
def stagor(url,specific =None, options=None):
if not componentInfo().check(): return
SI = siteInfo()
CI = campaignInfo()
UC = unifiedConfiguration()
done_by_wf_id = {}
done_by_input = {}
completion_by_input = {}
good_enough = 100.0
lost_blocks = json.loads(open('%s/lost_blocks_datasets.json'%monitor_dir).read())
lost_files = json.loads(open('%s/lost_files_datasets.json'%monitor_dir).read())
known_lost_blocks = {}
known_lost_files = {}
for dataset in set(lost_blocks.keys()+lost_files.keys()):
b,f = findLostBlocksFiles(url, dataset)
if dataset in lost_blocks and not b:
print dataset,"has no really lost blocks"
else:
known_lost_blocks[dataset] = [i['name'] for i in b]
if dataset in lost_files and not f:
print dataset,"has no really lost files"
else:
known_lost_files[dataset] = [i['name'] for i in f]
try:
cached_transfer_statuses = json.loads(open('cached_transfer_statuses.json').read())
except:
print "inexisting transfer statuses. starting fresh"
cached_transfer_statuses = {}
transfer_statuses = {}
## pop all that are now in negative values
for phedexid in cached_transfer_statuses.keys():
transfers = session.query(Transfer).filter(Transfer.phedexid==int(phedexid)).all()
if not transfers:
print phedexid,"does not look relevant to be in cache anymore. poping"
print cached_transfer_statuses.pop( phedexid )
## collect all datasets that are needed for wf in staging, correcting the status of those that are not really in staging
wfois = []
needs = defaultdict(list)
for wfo in session.query(Workflow).filter(Workflow.status == 'staging').all():
wfi = workflowInfo(url, wfo.name)
if wfi.request['RequestStatus'] in ['running-open','running-closed','completed','assigned','acquired']:
wfi.sendLog('stagor', "is in status %s"%wfi.request['RequestStatus'])
wfi.status='away'
session.commit()
continue
if not wfi.request['RequestStatus'] in ['assignment-approved']:
## should be setting 'away' too
print wfo.name,"is",wfi.request['RequestStatus']
sendEmail("wrong status in staging. debug","%s is in %s, should set away."%(wfo.name,wfi.request['RequestStatus']))
wfois.append( (wfo,wfi) )
_,primaries,_,secondaries = wfi.getIO()
for dataset in list(primaries)+list(secondaries):
needs[wfo.name].append( dataset)
done_by_input[dataset] = {}
completion_by_input[dataset] = {}
wfi.sendLog('stagor', '%s needs %s'%( wfo.name, dataset))
open('%s/dataset_requirements.json'%monitor_dir,'w').write( json.dumps( needs, indent=2))
dataset_endpoints = defaultdict(set)
endpoint_in_downtime = defaultdict(set)
#endpoint_completed = defaultdict(set)
endpoint_incompleted = defaultdict(set)
#endpoint = defaultdict(set)
send_back_to_considered = set()
## phedexid are set negative when not relevant anymore
# probably there is a db schema that would allow much faster and simpler query
for transfer in session.query(Transfer).filter(Transfer.phedexid>0).all():
if specific and str(transfer.phedexid)!=str(specific): continue
skip=True
for wfid in transfer.workflows_id:
tr_wf = session.query(Workflow).get(wfid)
if tr_wf:
if tr_wf.status == 'staging':
sendLog('stagor',"\t%s is staging for %s"%(transfer.phedexid, tr_wf.name))
skip=False
if skip:
sendLog('stagor',"setting %s to negative value"%transfer.phedexid)
transfer.phedexid = -transfer.phedexid
session.commit()
continue
if transfer.phedexid<0: continue
## check the status of transfers
checks = checkTransferApproval(url, transfer.phedexid)
approved = all(checks.values())
if not approved:
sendLog('stagor', "%s is not yet approved"%transfer.phedexid)
approveSubscription(url, transfer.phedexid)
continue
## check on transfer completion
if str(transfer.phedexid) in cached_transfer_statuses:
### use a cache for transfer that already looked done
sendLog('stagor',"read %s from cache"%transfer.phedexid)
checks = cached_transfer_statuses[str(transfer.phedexid)]
else:
checks = checkTransferStatus(url, transfer.phedexid, nocollapse=True)
## just write this out
transfer_statuses[str(transfer.phedexid)] = copy.deepcopy(checks)
if not specific:
for dsname in checks:
if not dsname in done_by_input: done_by_input[dsname]={}
if not dsname in completion_by_input: completion_by_input[dsname] = {}
done_by_input[dsname][transfer.phedexid]=all(map(lambda i:i>=good_enough, checks[dsname].values()))
completion_by_input[dsname][transfer.phedexid]=checks[dsname].values()
if checks:
sendLog('stagor',"Checks for %s are %s"%( transfer.phedexid, [node.values() for node in checks.values()]))
done = all(map(lambda i:i>=good_enough,list(itertools.chain.from_iterable([node.values() for node in checks.values()]))))
else:
## it is empty, is that a sign that all is done and away ?
print "ERROR with the scubscriptions API of ",transfer.phedexid
print "Most likely something else is overiding the transfer request. Need to work on finding the replacement automatically, if the replacement exists"
done = False
## the thing above is NOT giving the right number
#done = False
for wfid in transfer.workflows_id:
tr_wf = session.query(Workflow).get(wfid)
if tr_wf:# and tr_wf.status == 'staging':
if not tr_wf.id in done_by_wf_id: done_by_wf_id[tr_wf.id]={}
done_by_wf_id[tr_wf.id][transfer.phedexid]=done
## for those that are in staging, and the destination site is in drain
#if not done and tr_wf.status == 'staging':
for ds in checks:
for s,v in checks[ds].items():
dataset_endpoints[ds].add( s )
if done:
## transfer.status = 'done'
sendLog('stagor',"%s is done"%transfer.phedexid)
cached_transfer_statuses[str(transfer.phedexid)] = copy.deepcopy(checks)
else:
sendLog('stagor',"%s is not finished %s"%(transfer.phedexid, pprint.pformat( checks )))
pprint.pprint( checks )
## check if the destination is in down-time
for ds in checks:
sites_incomplete = [SI.SE_to_CE(s) for s,v in checks[ds].items() if v<good_enough]
sites_incomplete_down = [s for s in sites_incomplete if not s in SI.sites_ready]
if sites_incomplete_down:
sendLog('stagor',"%s are in downtime, while waiting for %s to get there"%( ",".join(sites_incomplete_down), ds))
#sites_complete = [SI.SE_to_CE(s) for s,v in checks[ds].items() if v>=good_enough]
#endpoint[ds].update( sites_complete )
#endpoint[ds].update( sites_incomplete )
#endpoint_completed[ds].update( sites_complete )
endpoint_incompleted[ds].update( sites_incomplete )
endpoint_in_downtime[ds].update( sites_incomplete_down )
print "End point in down time"
for k in endpoint_in_downtime: endpoint_in_downtime[k] = list(endpoint_in_downtime[k])
for k in dataset_endpoints: dataset_endpoints[k] = list(dataset_endpoints[k])
print json.dumps( endpoint_in_downtime , indent=2)
open('cached_transfer_statuses.json','w').write( json.dumps( cached_transfer_statuses, indent=2))
open('%s/transfer_statuses.json'%monitor_dir,'w').write( json.dumps( transfer_statuses, indent=2))
open('%s/dataset_endpoints.json'%monitor_dir,'w').write( json.dumps(dataset_endpoints, indent=2))
already_stuck = json.loads( open('%s/stuck_transfers.json'%monitor_dir).read() )
missing_in_action = defaultdict(list)
print "-"*10,"Checking on workflows in staging","-"*10
#forget_about = ['/MinBias_TuneCUETP8M1_13TeV-pythia8/RunIISummer15GS-MCRUN2_71_V1-v2/GEN-SIM']
#for what in forget_about:
# if not done_by_input[what]:
# done_by_input[what] = {'fake':True}
## come back to workflows and check if they can go
available_cache = defaultdict(lambda : defaultdict(float))
presence_cache = defaultdict(dict)
for wfo,wfi in wfois:
print "#"*30
## the site white list takes site, campaign, memory and core information
(_,primaries,_,secondaries,sites_allowed) = wfi.getSiteWhiteList(verbose=False)
se_allowed = [SI.CE_to_SE(site) for site in sites_allowed]
se_allowed.sort()
se_allowed_key = ','.join(se_allowed)
readys={}
for need in list(primaries)+list(secondaries):
if not need in done_by_input:
wfi.sendLog('stagor',"missing transfer report for %s"%need)
readys[need] = False
## should warn someone about this !!!
## it cannot happen, by construction
sendEmail('missing transfer report','%s does not have a transfer report'%(need))
continue
if not done_by_input[need] and need in list(secondaries):
wfi.sendLog('stagor',"assuming it is OK for secondary %s to have no attached transfers"% need)
readys[need] = True
done_by_input[need] = { "fake" : True }
continue
if len(done_by_input[need]) and all(done_by_input[need].values()):
wfi.sendLog('stagor',"%s is ready"%need)
print json.dumps( done_by_input[need] , indent=2)
readys[need] = True
else:
wfi.sendLog('stagor',"%s is not ready"%need)
print json.dumps( done_by_input[need] , indent=2)
readys[need] = False
if readys and all(readys.values()):
if wfo.status == 'staging':
wfi.sendLog('stagor',"all needs are fullfilled, setting staged")
wfo.status = 'staged'
session.commit()
else:
wfi.sendLog('stagor',"all needs are fullfilled, already")
print json.dumps( readys, indent=2 )
else:
wfi.sendLog('stagor',"missing requirements")
copies_needed,_ = wfi.getNCopies()
jump_ahead = False
re_transfer = False
## there is missing input let's do something more elaborated
for need in list(primaries):#+list(secondaries):
if endpoint_in_downtime[need] == endpoint_incompleted[need]:
#print need,"is going to an end point in downtime"
wfi.sendLog('stagor',"%s has only incomplete endpoint in downtime"%need)
re_transfer=True
if not se_allowed_key in available_cache[need]:
available_cache[need][se_allowed_key] = getDatasetBlocksFraction( url , need, sites=se_allowed )
if available_cache[need][se_allowed_key] >= copies_needed:
wfi.sendLog('stagor',"assuming it is OK to move on like this already for %s"%need)
jump_ahead = True
## compute a time since staging to filter jump starting ?
# check whether the inputs is already in the stuck list ...
for need in list(primaries)+list(secondaries):
if need in already_stuck:
wfi.sendLog('stagor',"%s is stuck, so try to jump ahead"%need)
jump_ahead = True
if jump_ahead or re_transfer:
details_text = "checking on availability for %s to jump ahead"%wfo.name
details_text += '\n%s wants %s copies'%(wfo.name,copies_needed)
copies_needed = max(1,copies_needed-1)
details_text += '\nlowering by one unit to %s'%copies_needed
wfi.sendLog('stagor', details_text)
all_check = True
prim_where = set()
for need in list(primaries):
if not se_allowed_key in presence_cache[need]:
presence_cache[need][se_allowed_key] = getDatasetPresence( url, need , within_sites=se_allowed)
presence = presence_cache[need][se_allowed_key]
prim_where.update( presence.keys() )
available = available_cache[need][se_allowed_key]
this_check = (available >= copies_needed)
wfi.sendLog('stagor', "%s is available %s times %s"%( need, available, this_check))
all_check &= this_check
if not all_check: break
for need in list(secondaries):
## I do not want to check on the secon
this_check = all(done_by_input[need].values())
wfi.sendLog('stagor',"%s is all transfered %s"%(need, json.dumps(done_by_input[need], indent=2)))
all_check&= this_check
#if not se_allowed_key in presence_cache[need]:
# presence_cache[need][se_allowed_key] = getDatasetPresence( url, need , within_sites=se_allowed)
## restrict to where the primary is
#presence = dict([(k,v) for (k,v) in presence_cache[need][se_allowed_key].items() if k in prim_where])
#this_check = all([there for (there,frac) in presence.values()])
#print need,"is present at all sites:",this_check
#all_check&= this_check
if all_check:
wfi.sendLog('stagor',"needs are sufficiently fullfilled, setting staged")
wfo.status = 'staged'
session.commit()
else:
print wfo.name,"has to wait a bit more"
wfi.sendLog('stagor',"needs to wait a bit more")
else:
wfi.sendLog('stagor',"not checking availability")
if re_transfer:
wfi.sendLog('stagor',"Sending back to considered because of endpoint in downtime")
if wfo.status == 'staging':
wfo.status = 'considered'
session.commit()
send_back_to_considered.add( wfo.name )
if send_back_to_considered:
#sendEmail("transfer to endpoint in downtime","sending back to considered the following workflows \n%s"%('\n'.join( send_back_to_considered)))
sendLog('stagor', "sending back to considered the following workflows \n%s"%('\n'.join( send_back_to_considered)), level='critical')
print "-"*10,"Checking on non-available datasets","-"*10
## now check on those that are not fully available
for dsname in available_cache.keys():
## squash the se_allowed_key key
available_cache[dsname] = min( available_cache[dsname].values() )
for dsname,available in available_cache.items():
using_its = getWorkflowByInput(url, dsname)
#print using_its
using_wfos = []
for using_it in using_its:
wf = session.query(Workflow).filter(Workflow.name == using_it).first()
if wf:
using_wfos.append( wf )
if not len(done_by_input[dsname]):
print "For dataset",dsname,"there are no transfer report. That's an issue."
for wf in using_wfos:
if wf.status == 'staging':
if UC.get("stagor_sends_back"):
print "sending",wf.name,"back to considered"
wf.status = 'considered'
session.commit()
#sendEmail( "send back to considered","%s was send back and might be trouble"% wf.name)
sendLog('stagor', "%s was send back and might be trouble"% wf.name, level='critical')
else:
print "would send",wf.name,"back to considered"
#sendEmail( "subscription lagging behind","susbscriptions to get %s running are not appearing in phedex. I would have send it back to considered but that's not good."% wf.name)
sendLog('stagor', "susbscriptions to get %s running are not appearing in phedex. I would have send it back to considered but that's not good."% wf.name, level='critical')
continue
## not compatible with checking on secondary availability
#if all([wf.status != 'staging' for wf in using_wfos]):
# ## means despite all checks that input is not needed
# continue
if available < 1.:
print "incomplete",dsname
## there is a problem in the method below that it does not account for files stuck in T1*Buffer only
lost_blocks,lost_files = findLostBlocksFiles( url, dsname )
lost_block_names = [item['name'] for item in lost_blocks]
lost_file_names = [item['name'] for item in lost_files]
if lost_blocks:
#print json.dumps( lost , indent=2 )
## estimate for how much !
fraction_loss,_,n_missing = getDatasetBlockFraction(dsname, lost_block_names)
print "We have lost",len(lost_block_names),"blocks",lost_block_names,"for %f%%"%(100.*fraction_loss)
if fraction_loss > 0.05: ## 95% completion mark
#sendEmail('we have lost too many blocks','%s is missing %d blocks, for %d events, %f %% loss'%(dsname, len(lost_block_names), n_missing, fraction_loss))
sendLog('stagor', '%s is missing %d blocks, for %d events, %3.2f %% loss'%(dsname, len(lost_block_names), n_missing, 100*fraction_loss), level='warning')
## the workflow should be rejected !
for wf in using_wfos:
if wf.status == 'staging':
print wf.name,"is doomed. setting to trouble"
wf.status = 'trouble'
session.commit()
#sendEmail('doomed workflow','%s has too much loss on the input dataset %s. please check on stagor logs https://cmst2.web.cern.ch/cmst2/unified/logs/stagor/last.log'%(wf.name, dsname))
sendLog('stagor', '%s has too much loss on the input dataset %s. Missing %d blocks, for %d events, %3.2f %% loss'%(wf.name, dsname, len(lost_block_names), n_missing, 100*fraction_loss), level='critical')
else:
## probably enough to make a ggus and remove
if not dsname in known_lost_blocks:
#sendEmail('we have lost a few blocks', '%s is missing %d blocks, for %d events, %f %% loss\n\n%s'%(dsname, len(lost_block_names), n_missing, fraction_loss, '\n'.join( lost_block_names ) ))
sendLog('stagor', '%s is missing %d blocks, for %d events, %f %% loss\n\n%s'%(dsname, len(lost_block_names), n_missing, fraction_loss, '\n'.join( lost_block_names ) ), level='critical')
known_lost_blocks[dsname] = [i['name'] for i in lost_blocks]
if lost_files:
fraction_loss,_,n_missing = getDatasetFileFraction(dsname, lost_file_names)
print "We have lost",len(lost_file_names),"files",lost_file_names,"for %f%%"%fraction_loss
if fraction_loss > 0.05:
#sendEmail('we have lost too many files','%s is missing %d files, for %d events, %f %% loss'%(dsname, len(lost_file_names),n_missing, fraction_loss))
sendLog('stagor', '%s is missing %d files, for %d events, %f %% loss'%(dsname, len(lost_file_names),n_missing, fraction_loss), level='critical')
for wf in using_wfos:
if wf.status == 'staging':
print wf.name,"is doomed. setting to trouble"
wf.status = 'trouble'
session.commit()
else:
## probably enough to make a ggus and remove
if not dsname in known_lost_files:
#sendEmail('we have lost a few files','%s is missing %d files, for %d events, %f %% loss\n\n%s'%(dsname, len(lost_file_names),n_missing, fraction_loss, '\n'.join(lost_file_names)))
sendLog('stagor', '%s is missing %d files, for %d events, %f %% loss\n\n%s'%(dsname, len(lost_file_names),n_missing, fraction_loss, '\n'.join(lost_file_names)), level='critical')
known_lost_files[dsname] = [i['name'] for i in lost_files]
## should the status be change to held-staging and pending on a ticket
missings = [pid for (pid,d) in done_by_input[dsname].items() if d==False]
print "\t",done_by_input[dsname]
print "\tneeds",len(done_by_input[dsname])
print "\tgot",done_by_input[dsname].values().count(True)
print "\tmissing",missings
missing_in_action[dsname].extend( missings )
rr= open('%s/lost_blocks_datasets.json'%monitor_dir,'w')
rr.write( json.dumps( known_lost_blocks, indent=2))
rr.close()
rr= open('%s/lost_files_datasets.json'%monitor_dir,'w')
rr.write( json.dumps( known_lost_files, indent=2))
rr.close()
open('%s/incomplete_transfers.json'%monitor_dir,'w').write( json.dumps(missing_in_action, indent=2) )
print "Stuck transfers and datasets"
print json.dumps( missing_in_action, indent=2 )
print "Going further and make a report of stuck transfers"
datasets_by_phid = defaultdict(set)
for dataset in missing_in_action:
for phid in missing_in_action[dataset]:
#print dataset,"stuck through",phid
datasets_by_phid[phid].add( dataset )
bad_destinations = defaultdict(set)
bad_sources = defaultdict(set)
report = ""
really_stuck_dataset = set()
transfer_timeout = UC.get("transfer_timeout")
transfer_lowrate = UC.get("transfer_lowrate")
for phid,datasets in datasets_by_phid.items():
issues = checkTransferLag( url, phid , datasets=list(datasets) )
for dataset in issues:
for block in issues[dataset]:
for destination in issues[dataset][block]:
(block_size,destination_size,delay,rate,dones) = issues[dataset][block][destination]
## count x_Buffer and x_MSS as one source
redones=[]
for d in dones:
if d.endswith('Buffer') or d.endswith('Export'):
if d.replace('Buffer','MSS').replace('Export','MSS') in dones:
continue
else:
redones.append( d )
else:
redones.append( d )
dones = list(set( redones ))
#dones = filter(lambda s : (s.endswith('Buffer') and not s.replace('Buffer','MSS') in dones) or (not s.endswith('Buffer')) , dones)
if delay>transfer_timeout and rate<transfer_lowrate:
if len(dones)>1:
## its the destination that sucks
bad_destinations[destination].add( block )
else:
dum=[bad_sources[d].add( block ) for d in dones]
really_stuck_dataset.add( dataset )
print "add",dataset,"to really stuck"
report += "%s is not getting to %s, out of %s faster than %f [GB/s] since %f [d]\n"%(block,destination,", ".join(dones), rate, delay)
print "\n"*2
## create tickets right away ?
report+="\nbad sources "+",".join(bad_sources.keys())+"\n"
for site,blocks in bad_sources.items():
report+="\n\n%s:"%site+"\n\t".join(['']+list(blocks))
report+="\nbad destinations "+",".join(bad_destinations.keys())+"\n"
for site,blocks in bad_destinations.items():
report+="\n\n%s:"%site+"\n\t".join(['']+list(blocks))
print '\n'*2,"Datasets really stuck"
print '\n'.join( really_stuck_dataset )
print '\n'*2,"report written at https://cmst2.web.cern.ch/cmst2/unified/logs/incomplete_transfers.log"
print report
stuck_transfers = dict([(k,v) for (k,v) in missing_in_action.items() if k in really_stuck_dataset])
print '\n'*2,'Stuck dataset transfers'
print json.dumps(stuck_transfers , indent=2)
open('%s/stuck_transfers.json'%monitor_dir,'w').write( json.dumps(stuck_transfers , indent=2) )
open('%s/logs/incomplete_transfers.log'%monitor_dir,'w').write( report )
unlock = True
bad_ds = True
time_point("Checked status", sub_lap=True)
if not bad_ds:
## get a chance at unlocking if custodial is existing
(_,dsn,ps,tier) = dataset.split('/')
no_tape = (tier in tier_no_custodial)
if no_tape:
for c in custodial_override:
if c in ps and tier in custodial_override[c]:
no_tape=False
break
if no_tape:
## could add a one-full copy consistency check
presence = getDatasetPresence(url, dataset)
if any([there for _,(there,_) in presence.items()]):
unlock = True
else:
newly_locking.add(dataset)
unlock = False
time_point("Checked presence", sub_lap=True)
else:
custodials,info = findCustodialCompletion(url, dataset)
if len(custodials) == 0:
## add it back for that reason
newly_locking.add(dataset)
if not ds_status: ds_status = getDatasetStatus( dataset )
ds_size = getDatasetSize( dataset )
print "Can't unlock",dataset," of size", ds_size,"[GB] because it is not custodial yet",ds_status
unlock = False
def closor(url, specific=None):
if userLock(): return
if duplicateLock(): return
if not componentInfo().check(): return
UC = unifiedConfiguration()
CI = campaignInfo()
all_late_files = []
check_fullcopy_to_announce = UC.get('check_fullcopy_to_announce')
## manually closed-out workflows should get to close with checkor
if specific:
wfs = session.query(Workflow).filter(
Workflow.name.contains(specific)).all()
else:
wfs = session.query(Workflow).filter(Workflow.status == 'close').all()
held = set()
max_per_round = UC.get('max_per_round').get('closor', None)
random.shuffle(wfs)
if max_per_round: wfs = wfs[:max_per_round]
for wfo in wfs:
if specific and not specific in wfo.name: continue
## what is the expected #lumis
wfi = workflowInfo(url, wfo.name)
wfo.wm_status = wfi.request['RequestStatus']
if wfi.request['RequestStatus'] in ['announced', 'normal-archived']:
## manually announced ??
wfo.status = 'done'
wfo.wm_status = wfi.request['RequestStatus']
wfi.sendLog(
'closor',
'%s is announced already : %s' % (wfo.name, wfo.wm_status))
session.commit()
expected_lumis = 1
if not 'TotalInputLumis' in wfi.request:
print wfo.name, "has not been assigned yet, or the database is corrupted"
else:
expected_lumis = wfi.request['TotalInputLumis']
## what are the outputs
outputs = wfi.request['OutputDatasets']
## check whether the number of lumis is as expected for each
all_OK = defaultdict(lambda: False)
#print outputs
if len(outputs):
print wfo.name, wfi.request['RequestStatus']
for out in outputs:
event_count, lumi_count = getDatasetEventsAndLumis(dataset=out)
odb = session.query(Output).filter(
Output.datasetname == out).first()
if not odb:
print "adding an output object", out
odb = Output(datasetname=out)
odb.workflow = wfo
session.add(odb)
odb.nlumis = lumi_count
odb.nevents = event_count
odb.workfow_id = wfo.id
if odb.expectedlumis < expected_lumis:
odb.expectedlumis = expected_lumis
else:
expected_lumis = odb.expectedlumis
odb.date = time.mktime(time.gmtime())
session.commit()
wfi.sendLog(
'closor', "\t%60s %d/%d = %3.2f%%" %
(out, lumi_count, expected_lumis,
lumi_count / float(expected_lumis) * 100.))
#print wfo.fraction_for_closing, lumi_count, expected_lumis
#fraction = wfo.fraction_for_closing
#fraction = 0.0
#all_OK.append((float(lumi_count) > float(expected_lumis*fraction)))
all_OK[out] = True
## check for at least one full copy prior to moving on
in_full = {}
for out in outputs:
in_full[out] = []
presence = getDatasetPresence(url, out)
where = [site for site, info in presence.items() if info[0]]
if where:
all_OK[out] = True
print out, "is in full at", ",".join(where)
in_full[out] = copy.deepcopy(where)
else:
going_to = wfi.request['NonCustodialSites'] + wfi.request[
'CustodialSites']
wfi.sendLog(
'closor', "%s is not in full anywhere. send to %s" %
(out, ",".join(sorted(going_to))))
at_destination = dict([(k, v) for (k, v) in presence.items()
if k in going_to])
else_where = dict([(k, v) for (k, v) in presence.items()
if not k in going_to])
print json.dumps(at_destination)
print json.dumps(else_where, indent=2)
## do the full stuck transfer study, missing files and shit !
for there in going_to:
late_info = findLateFiles(url, out, going_to=there)
for l in late_info:
l.update({"workflow": wfo.name, "dataset": out})
all_late_files.extend(late_info)
if check_fullcopy_to_announce:
## only set this false if the check is relevant
all_OK[out] = False
## verify if we have to do harvesting
(OK, requests) = spawn_harvesting(url, wfi, in_full)
all_OK.update(OK)
## only that status can let me go into announced
if all(all_OK.values()) and wfi.request['RequestStatus'] in [
'closed-out'
]:
print wfo.name, "to be announced"
results = [] #'dummy']
if not results:
for out in outputs:
if all_OK[out]:
results.append(setDatasetStatus(out, 'VALID'))
tier = out.split('/')[-1]
campaign = None
try:
campaign = out.split('/')[2].split('-')[0]
except:
if 'Campaign' in wfi.request and wfi.request[
'Campaign']:
campaign = wfi.request['Campaign']
to_DDM = False
## campaign override
if campaign and campaign in CI.campaigns and 'toDDM' in CI.campaigns[
campaign] and tier in CI.campaigns[campaign][
'toDDM']:
to_DDM = True
## by typical enabling
if tier in UC.get("tiers_to_DDM"):
to_DDM = True
## check for unitarity
if not tier in UC.get("tiers_no_DDM") + UC.get(
"tiers_to_DDM"):
print "tier", tier, "neither TO or NO DDM for", out
results.append('Not recognitized tier %s' % tier)
#sendEmail("failed DDM injection","could not recognize %s for injecting in DDM"% out)
sendLog(
'closor',
"could not recognize %s for injecting in DDM" %
out,
level='critical')
continue
n_copies = 2
destinations = []
if to_DDM and campaign and campaign in CI.campaigns and 'DDMcopies' in CI.campaigns[
campaign]:
ddm_instructions = CI.campaigns[campaign][
'DDMcopies']
if type(ddm_instructions) == int:
n_copies = CI.campaigns[campaign]['DDMcopies']
elif type(ddm_instructions) == dict:
## a more fancy configuration
for ddmtier, indication in ddm_instructions.items(
):
if ddmtier == tier or ddmtier in [
'*', 'all'
]:
## this is for us
if 'N' in indication:
n_copies = indication['N']
if 'host' in indication:
destinations = indication['host']
destination_spec = ""
if destinations:
destination_spec = "--destination=" + ",".join(
destinations)
## inject to DDM when necessary
if to_DDM:
#print "Sending",out," to DDM"
p = os.popen(
'python assignDatasetToSite.py --nCopies=%d --dataset=%s %s --exec'
% (n_copies, out, destination_spec))
print p.read()
status = p.close()
if status != None:
print "Failed DDM, retrying a second time"
p = os.popen(
'python assignDatasetToSite.py --nCopies=%d --dataset=%s %s --exec'
% (n_copies, out, destination_spec))
print p.read()
status = p.close()
if status != None:
#sendEmail("failed DDM injection","could not add "+out+" to DDM pool. check closor logs.")
sendLog('closor',
"could not add " + out +
" to DDM pool. check closor logs.",
level='critical')
results.append(status)
if status == None:
wfi.sendLog(
'closor',
'%s is send to AnalysisOps DDM pool in %s copies %s'
% (n_copies, out, destination_spec))
else:
print wfo.name, "no stats for announcing", out
results.append('No Stats')
if all(
map(lambda result: result in ['None', None, True],
results)):
## only announce if all previous are fine
res = reqMgrClient.announceWorkflowCascade(url, wfo.name)
if not res in ['None', None]:
## check the status again, it might well have toggled
wl_bis = workflowInfo(url, wfo.name)
wfo.wm_status = wl_bis.request['RequestStatus']
session.commit()
if wl_bis.request['RequestStatus'] in [
'announced', 'normal-archived'
]:
res = None
else:
## retry ?
res = reqMgrClient.announceWorkflowCascade(
url, wfo.name)
results.append(res)
#print results
if all(map(lambda result: result in ['None', None, True],
results)):
wfo.status = 'done'
session.commit()
wfi.sendLog('closor', "workflow is announced")
else:
print "ERROR with ", wfo.name, "to be announced", json.dumps(
results)
else:
print wfo.name, "not good for announcing:", wfi.request[
'RequestStatus']
wfi.sendLog('closor', "cannot be announced")
held.add(wfo.name)
days_late = 0.
retries_late = 10
really_late_files = [
info for info in all_late_files if info['retries'] >= retries_late
]
really_late_files = [
info for info in really_late_files
if info['delay'] / (60 * 60 * 24.) >= days_late
]
if really_late_files:
subject = 'These %d files are lagging for %d days and %d retries announcing dataset \n%s' % (
len(really_late_files), days_late, retries_late,
json.dumps(really_late_files, indent=2))
#sendEmail('waiting for files to announce', subject)
sendLog('closor', subject, level='warning')
sendLog('closor', subject)
print subject
open('%s/stuck_files.json' % monitor_dir,
'w').write(json.dumps(really_late_files, indent=2))
if held:
#sendEmail("held from announcing","the workflows below are held up, please check the logs https://cmst2.web.cern.ch/cmst2/unified/logs/closor/last.log \n%s"%("\n".join( held )))
sendLog('closor',
"the workflows below are held up \n%s" % ("\n".join(held)),
level='critical')
def checkor(url, spec=None, options=None):
fDB = closeoutInfo()
if userLock():
return
if duplicateLock():
return
UC = unifiedConfiguration()
use_mcm = True
up = componentInfo(mcm=use_mcm, soft=["mcm"])
if not up.check():
return
use_mcm = up.status["mcm"]
wfs = []
if options.fetch:
## get all in running and check
wfs.extend(session.query(Workflow).filter(Workflow.status == "away").all())
wfs.extend(session.query(Workflow).filter(Workflow.status == "assistance").all())
if options.nofetch:
## than get all in need for assistance
wfs.extend(session.query(Workflow).filter(Workflow.status.startswith("assistance-")).all())
custodials = defaultdict(list) # sites : dataset list
transfers = defaultdict(list) # sites : dataset list
invalidations = [] # a list of files
SI = global_SI
CI = campaignInfo()
mcm = McMClient(dev=False)
def get_campaign(output, wfi):
campaign = None
try:
campaign = output.split("/")[2].split("-")[0]
except:
if "Campaign" in wfi.request:
campaign = wfi.request["Campaign"]
return campaign
by_passes = []
holdings = []
for bypassor, email in [
("jbadillo", "*****@*****.**"),
("vlimant", "*****@*****.**"),
("jen_a", "*****@*****.**"),
]:
bypass_file = "/afs/cern.ch/user/%s/%s/public/ops/bypass.json" % (bypassor[0], bypassor)
if not os.path.isfile(bypass_file):
print "no file", bypass_file
continue
try:
by_passes.extend(json.loads(open(bypass_file).read()))
except:
print "cannot get by-passes from", bypass_file, "for", bypassor
sendEmail("malformated by-pass information", "%s is not json readable" % (bypass_file), destination=[email])
holding_file = "/afs/cern.ch/user/%s/%s/public/ops/onhold.json" % (bypassor[0], bypassor)
if not os.path.isfile(holding_file):
print "no file", holding_file
continue
try:
holdings.extend(json.loads(open(holding_file).read()))
except:
print "cannot get holdings from", holding_file, "for", bypassor
sendEmail(
"malformated by-pass information", "%s is not json readable" % (holding_file), destination=[email]
)
total_running_time = 5.0 * 60.0
sleep_time = max(0.5, total_running_time / len(wfs))
for wfo in wfs:
if spec and not (spec in wfo.name):
continue
time.sleep(sleep_time)
print "checking on", wfo.name
## get info
wfi = workflowInfo(url, wfo.name)
## make sure the wm status is up to date.
# and send things back/forward if necessary.
wfo.wm_status = wfi.request["RequestStatus"]
if wfo.wm_status == "closed-out":
## manually closed-out
print wfo.name, "is already", wfo.wm_status
wfo.status = "close"
session.commit()
continue
elif wfo.wm_status in [
"failed",
"aborted",
"aborted-archived",
"rejected",
"rejected-archived",
"aborted-completed",
]:
## went into trouble
wfo.status = "trouble"
print wfo.name, "is in trouble", wfo.wm_status
session.commit()
continue
elif wfo.wm_status in ["assigned", "acquired"]:
## not worth checking yet
print wfo.name, "not running yet"
session.commit()
continue
if "-onhold" in wfo.status:
if wfo.name in holdings and wfo.name not in by_passes:
print wfo.name, "on hold"
continue
if wfo.name in holdings and wfo.name not in by_passes:
wfo.status = "assistance-onhold"
print "setting", wfo.name, "on hold"
session.commit()
continue
if wfo.wm_status != "completed" and not wfo.name in by_passes:
## for sure move on with closeout check if in completed
print "no need to check on", wfo.name, "in status", wfo.wm_status
session.commit()
continue
session.commit()
sub_assistance = "" # if that string is filled, there will be need for manual assistance
is_closing = True
## get it from somewhere
by_pass_checks = False
if wfo.name in by_passes:
print "we can bypass checks on", wfo.name
by_pass_checks = True
for bypass in by_passes:
if bypass in wfo.name:
print "we can bypass", wfo.name, "because of keyword", bypass
by_pass_checks = True
break
if not CI.go(wfi.request["Campaign"]) and not by_pass_checks:
print "No go for", wfo.name
continue
# tuck out DQMIO/DQM
wfi.request["OutputDatasets"] = [out for out in wfi.request["OutputDatasets"] if not "/DQM" in out]
## anything running on acdc
familly = getWorkflowById(url, wfi.request["PrepID"], details=True)
acdc = []
acdc_inactive = []
has_recovery_going = False
had_any_recovery = False
for member in familly:
if member["RequestType"] != "Resubmission":
continue
if member["RequestName"] == wfo.name:
continue
if member["RequestDate"] < wfi.request["RequestDate"]:
continue
if member["RequestStatus"] in [
"running-open",
"running-closed",
"assignment-approved",
"assigned",
"acquired",
]:
print wfo.name, "still has an ACDC running", member["RequestName"]
acdc.append(member["RequestName"])
# print json.dumps(member,indent=2)
## hook for just waiting ...
is_closing = False
has_recovery_going = True
elif member["RequestStatus"] == None:
print member["RequestName"], "is not real"
pass
else:
acdc_inactive.append(member["RequestName"])
had_any_recovery = True
## completion check
percent_completions = {}
# print "let's see who is crashing", wfo.name
# print wfi.request['TotalInputEvents'],wfi.request['TotalInputLumis']
if not "TotalInputEvents" in wfi.request:
event_expected, lumi_expected = 0, 0
if not "recovery" in wfo.status:
sendEmail(
"missing member of the request",
"TotalInputEvents is missing from the workload of %s" % wfo.name,
destination=["*****@*****.**"],
)
else:
event_expected, lumi_expected = wfi.request["TotalInputEvents"], wfi.request["TotalInputLumis"]
if "RequestNumEvents" in wfi.request:
event_expected = int(wfi.request["RequestNumEvents"])
elif "Task1" in wfi.request and "RequestNumEvents" in wfi.request["Task1"]:
event_expected = int(wfi.request["Task1"]["RequestNumEvents"])
fractions_pass = {}
for output in wfi.request["OutputDatasets"]:
event_count, lumi_count = getDatasetEventsAndLumis(dataset=output)
percent_completions[output] = 0.0
if lumi_expected:
percent_completions[output] = lumi_count / float(lumi_expected)
if event_expected:
percent_completions[output] = max(percent_completions[output], event_count / float(event_expected))
fractions_pass[output] = 0.95
c = get_campaign(output, wfi)
if c in CI.campaigns and "fractionpass" in CI.campaigns[c]:
fractions_pass[output] = CI.campaigns[c]["fractionpass"]
print "overriding fraction to", fractions_pass[output], "for", output
if options.fractionpass:
fractions_pass[output] = options.fractionpass
print "overriding fraction to", fractions_pass[output], "by command line for", output
if not all([percent_completions[out] >= fractions_pass[out] for out in fractions_pass]):
print wfo.name, "is not completed"
print json.dumps(percent_completions, indent=2)
print json.dumps(fractions_pass, indent=2)
## hook for creating automatically ACDC ?
if has_recovery_going:
sub_assistance += "-recovering"
elif had_any_recovery:
## we want to have this looked at
sub_assistance += "-manual"
else:
sub_assistance += "-recovery"
is_closing = False
## correct lumi < 300 event per lumi
events_per_lumi = {}
for output in wfi.request["OutputDatasets"]:
events_per_lumi[output] = getDatasetEventsPerLumi(output)
lumi_upper_limit = {}
for output in wfi.request["OutputDatasets"]:
upper_limit = 301.0
campaign = get_campaign(output, wfi)
# if 'EventsPerLumi' in wfi.request and 'FilterEfficiency' in wfi.request:
# upper_limit = 1.5*wfi.request['EventsPerLumi']*wfi.request['FilterEfficiency']
# print "setting the upper limit of lumisize to",upper_limit,"by request configuration"
if campaign in CI.campaigns and "lumisize" in CI.campaigns[campaign]:
upper_limit = CI.campaigns[campaign]["lumisize"]
print "overriding the upper lumi size to", upper_limit, "for", campaign
if options.lumisize:
upper_limit = options.lumisize
print "overriding the upper lumi size to", upper_limit, "by command line"
lumi_upper_limit[output] = upper_limit
if any([events_per_lumi[out] >= lumi_upper_limit[out] for out in events_per_lumi]):
print wfo.name, "has big lumisections"
print json.dumps(events_per_lumi, indent=2)
## hook for rejecting the request ?
sub_assistance += "-biglumi"
is_closing = False
any_presence = {}
for output in wfi.request["OutputDatasets"]:
any_presence[output] = getDatasetPresence(url, output, vetoes=[])
## custodial copy
custodial_locations = {}
custodial_presences = {}
for output in wfi.request["OutputDatasets"]:
custodial_presences[output] = [s for s in any_presence[output] if "MSS" in s]
custodial_locations[output] = phedexClient.getCustodialSubscriptionRequestSite(output)
if not custodial_locations[output]:
custodial_locations[output] = []
## presence in phedex
phedex_presence = {}
for output in wfi.request["OutputDatasets"]:
phedex_presence[output] = phedexClient.getFileCountDataset(url, output)
vetoed_custodial_tier = UC.get("tiers_with_no_custodial")
out_worth_checking = [
out for out in custodial_locations.keys() if out.split("/")[-1] not in vetoed_custodial_tier
]
size_worth_checking = sum(
[getDatasetSize(out) / 1023.0 for out in out_worth_checking]
) ## size in TBs of all outputs
if not all(map(lambda sites: len(sites) != 0, [custodial_locations[out] for out in out_worth_checking])):
print wfo.name, "has not all custodial location"
print json.dumps(custodial_locations, indent=2)
##########
## hook for making a custodial replica ?
custodial = None
## get from other outputs
for output in out_worth_checking:
if len(custodial_locations[output]):
custodial = custodial_locations[output][0]
if custodial and float(SI.storage[custodial]) < size_worth_checking:
print "cannot use the other output custodial:", custodial, "because of limited space"
custodial = None
## try to get it from campaign configuration
if not custodial:
for output in out_worth_checking:
campaign = get_campaign(output, wfi)
if campaign in CI.campaigns and "custodial" in CI.campaigns[campaign]:
custodial = CI.campaigns[campaign]["custodial"]
print "Setting custodial to", custodial, "from campaign configuration"
break
if custodial and float(SI.storage[custodial]) < size_worth_checking:
print "cannot use the unified configuration custodial:", custodial, "because of limited space"
custodial = None
## get from the parent
pick_custodial = True
if not custodial and "InputDataset" in wfi.request:
## this is terribly dangerous to assume only
parents_custodial = phedexClient.getCustodialSubscriptionRequestSite(wfi.request["InputDataset"])
###parents_custodial = findCustodialLocation(url, wfi.request['InputDataset'])
if not parents_custodial:
parents_custodial = []
if len(parents_custodial):
custodial = parents_custodial[0]
else:
print "the input dataset", wfi.request[
"InputDataset"
], "does not have custodial in the first place. abort"
sendEmail(
"dataset has no custodial location",
"Please take a look at %s in the logs of checkor" % wfi.request["InputDataset"],
)
is_closing = False
pick_custodial = False
if custodial and float(SI.storage[custodial]) < size_worth_checking:
print "cannot use the parent custodial:", custodial, "because of limited space"
custodial = None
if not custodial and pick_custodial:
## pick one at random
custodial = SI.pick_SE(size=size_worth_checking)
if not custodial:
print "cannot find a custodial for", wfo.name
sendEmail(
"cannot find a custodial",
"cannot find a custodial for %s probably because of the total output size %d"
% (wfo.name, size_worth_checking),
)
if custodial and ((not sub_assistance and not acdc) or by_pass_checks):
## remember how much you added this round already ; this stays locally
SI.storage[custodial] -= size_worth_checking
## register the custodial request, if there are no other big issues
for output in out_worth_checking:
if not len(custodial_locations[output]):
if phedex_presence[output] >= 1:
custodials[custodial].append(output)
else:
print "no file in phedex for", output, " not good to add to custodial requests"
is_closing = False
## disk copy
disk_copies = {}
for output in wfi.request["OutputDatasets"]:
disk_copies[output] = [s for s in any_presence[output] if (not "MSS" in s) and (not "Buffer" in s)]
if not all(map(lambda sites: len(sites) != 0, disk_copies.values())):
print wfo.name, "has not all output on disk"
print json.dumps(disk_copies, indent=2)
## presence in dbs
dbs_presence = {}
dbs_invalid = {}
for output in wfi.request["OutputDatasets"]:
dbs_presence[output] = dbs3Client.getFileCountDataset(output)
dbs_invalid[output] = dbs3Client.getFileCountDataset(output, onlyInvalid=True)
fraction_invalid = 0.01
if (
not all(
[
dbs_presence[out] == (dbs_invalid[out] + phedex_presence[out])
for out in wfi.request["OutputDatasets"]
]
)
and not options.ignorefiles
):
print wfo.name, "has a dbs,phedex mismatch"
print json.dumps(dbs_presence, indent=2)
print json.dumps(dbs_invalid, indent=2)
print json.dumps(phedex_presence, indent=2)
## hook for just waiting ...
is_closing = False
if (
not all(
[
(dbs_invalid[out] <= int(fraction_invalid * dbs_presence[out]))
for out in wfi.request["OutputDatasets"]
]
)
and not options.ignorefiles
):
print wfo.name, "has a dbs invalid file level too high"
print json.dumps(dbs_presence, indent=2)
print json.dumps(dbs_invalid, indent=2)
print json.dumps(phedex_presence, indent=2)
## need to be going and taking an eye
sub_assistance += "-invalidfiles"
is_closing = False
## put that heavy part at the end
## duplication check
duplications = {}
if is_closing:
print "starting duplicate checker for", wfo.name
for output in wfi.request["OutputDatasets"]:
print "\tchecking", output
duplications[output] = True
try:
duplications[output] = dbs3Client.duplicateRunLumi(output)
except:
try:
duplications[output] = dbs3Client.duplicateRunLumi(output)
except:
print "was not possible to get the duplicate count for", output
is_closing = False
if any(duplications.values()) and not options.ignoreduplicates:
print wfo.name, "has duplicates"
print json.dumps(duplications, indent=2)
## hook for making file invalidation ?
sub_assistance += "-duplicates"
is_closing = False
## for visualization later on
if not wfo.name in fDB.record:
# print "adding",wfo.name,"to close out record"
fDB.record[wfo.name] = {"datasets": {}, "name": wfo.name, "closeOutWorkflow": None}
fDB.record[wfo.name]["closeOutWorkflow"] = is_closing
for output in wfi.request["OutputDatasets"]:
if not output in fDB.record[wfo.name]["datasets"]:
fDB.record[wfo.name]["datasets"][output] = {}
rec = fDB.record[wfo.name]["datasets"][output]
rec["percentage"] = float("%.2f" % (percent_completions[output] * 100))
rec["duplicate"] = duplications[output] if output in duplications else "N/A"
rec["phedexReqs"] = (
float("%.2f" % any_presence[output][custodial_presences[output][0]][1])
if len(custodial_presences[output]) != 0
else "N/A"
)
rec["closeOutDataset"] = is_closing
rec["transPerc"] = (
float("%.2f" % any_presence[output][disk_copies[output][0]][1])
if len(disk_copies[output]) != 0
else "N/A"
)
rec["correctLumis"] = (
int(events_per_lumi[output]) if (events_per_lumi[output] > lumi_upper_limit[output]) else True
)
rec["missingSubs"] = (
False if len(custodial_locations[output]) == 0 else ",".join(list(set(custodial_locations[output])))
)
rec["dbsFiles"] = dbs_presence[output]
rec["dbsInvFiles"] = dbs_invalid[output]
rec["phedexFiles"] = phedex_presence[output]
rec["acdc"] = "%d / %d" % (len(acdc), len(acdc + acdc_inactive))
if by_pass_checks:
## force closing
is_closing = True
## and move on
if is_closing:
## toggle status to closed-out in request manager
print "setting", wfo.name, "closed-out"
if not options.test:
if wfo.wm_status in ["closed-out", "announced", "normal-archived"]:
print wfo.name, "is already", wfo.wm_status, "not trying to closed-out and assuming it does"
res = None
else:
res = reqMgrClient.closeOutWorkflowCascade(url, wfo.name)
print "close out answer", res
if not res in ["None", None]:
print "retrying to closing out"
print res
res = reqMgrClient.closeOutWorkflowCascade(url, wfo.name)
if res in [None, "None"]:
wfo.status = "close"
session.commit()
else:
print "could not close out", wfo.name, "will try again next time"
else:
## that means there is something that needs to be done acdc, lumi invalidation, custodial, name it
new_status = "assistance" + sub_assistance
print wfo.name, "needs assistance with", new_status
if sub_assistance and wfo.status != new_status and "PrepID" in wfi.request and not "manual" in wfo.status:
pid = wfi.getPrepIDs()[0].replace("task_", "")
# pid = wfi.request['PrepID'].replace('task_','')
## notify
messages = {
"recovery": "Samples completed with missing statistics:\n%s "
% (
"\n".join(
[
"%.2f %% complete for %s" % (percent_completions[output] * 100, output)
for output in wfi.request["OutputDatasets"]
]
)
),
"biglumi": "Samples completed with large luminosity blocks:\n%s "
% (
"\n".join(
[
"%d > %d for %s" % (events_per_lumi[output], lumi_upper_limit[output], output)
for output in wfi.request["OutputDatasets"]
if (events_per_lumi[output] > lumi_upper_limit[output])
]
)
),
"duplicate": "Samples completed with duplicated luminosity blocks:\n%s"
% (
"\n".join(
[
"%s" % output
for output in wfi.request["OutputDatasets"]
if output in duplications and duplications[output]
]
)
),
}
text = "The request %s (%s) is facing issue in production.\n" % (pid, wfo.name)
content = ""
for case in messages:
if case in new_status:
content += "\n" + messages[case] + "\n"
text += content
text += "You are invited to check, while this is being taken care of by Ops.\n"
text += "This is an automated message."
if use_mcm and content:
print "Sending notification back to requestor"
print text
batches = mcm.getA("batches", query="contains=%s&status=announced" % pid)
if len(batches):
## go notify the batch
bid = batches[-1]["prepid"]
print "batch nofication to", bid
mcm.put("/restapi/batches/notify", {"notes": text, "prepid": bid})
## go notify the request
print "request notification to", pid
mcm.put("/restapi/requests/notify", {"message": text, "prepids": [pid]})
## case where the workflow was in manual from recoveror
if not "manual" in wfo.status or new_status != "assistance-recovery":
wfo.status = new_status
if not options.test:
print "setting", wfo.name, "to", wfo.status
session.commit()
else:
print "current status is", wfo.status, "not changing to anything"
fDB.html()
## custodial requests
print "Custodials"
print json.dumps(custodials, indent=2)
for site in custodials:
print ",".join(custodials[site]), "=>", site
if not options.test:
result = makeReplicaRequest(
url,
site,
list(set(custodials[site])),
"custodial copy at production close-out",
custodial="y",
priority="low",
approve=(site in SI.sites_auto_approve),
)
print result
print "Transfers"
print json.dumps(transfers, indent=2)
## replicas requests
for site in transfers:
print ",".join(transfers[site]), "=>", site
if not options.test:
result = None
# result = makeReplicaRequest(url, site, list(set(transfers[site])),"copy to disk at production close-out")
print result
print "File Invalidation"
print invalidations
def parse_one(url, wfn, options=None):
SI = global_SI()
wfi = workflowInfo( url , wfn)
where_to_run, missing_to_run,missing_to_run_at = wfi.getRecoveryInfo()
all_blocks,needed_blocks,files_in_blocks,files_notin_dbs = wfi.getRecoveryBlocks()
ancestor = workflowInfo( url , wfn)
lhe,prim,_,sec = ancestor.getIO()
high_order_acdc = 0
while ancestor.request['RequestType'] == 'Resubmission':
ancestor = workflowInfo(url, ancestor.request['OriginalRequestName'])
lhe,prim,_,sec = ancestor.getIO()
high_order_acdc += 1
no_input = (not lhe) and len(prim)==0 and len(sec)==0
cache = 0
if options:
cache = options.cache
print "cache timeout", cache
err= wfi.getWMErrors(cache=cache)
stat = wfi.getWMStats(cache=cache)
#adcd = wfi.getRecoveryDoc()
total_by_code_dash = defaultdict( int )
total_by_site_dash = defaultdict( int )
r_dashb =defaultdict( lambda : defaultdict( int ))
dash_board_h = 1
if True :#'pdmvserv_TOP-RunIISummer15wmLHEGS-00103_00183_v0__161005_165048_809' in wfn:
## NB get the since from when the wf has started, not a fixed value
## no dashboard until we get a better api
#dashb = wfi.getFullPicture(since=dash_board_h,cache=cache)
dashb = {}
#print json.dumps( dashb , indent=2)
for site,sinfo in dashb.items():
for s_code,counts in sinfo.items():
d_statuses = ['submitted','pending','app-unknown','done']
total_by_code_dash[str(s_code)]+= counts.get('submitted',0)
total_by_site_dash[site] += counts.get('submitted',0)
r_dashb[str(s_code)][site] += counts.get('submitted',0)
print json.dumps(total_by_code_dash , indent=2)
print json.dumps(total_by_site_dash , indent=2)
status_per_task = defaultdict(lambda : defaultdict(int))
if not 'AgentJobInfo' in stat:
stat['AgentJobInfo'] = {}
#print "bad countent ?"
#print json.dumps( stat, indent=2)
for agent in stat['AgentJobInfo']:
for task in stat['AgentJobInfo'][agent]['tasks']:
if not 'status' in stat['AgentJobInfo'][agent]['tasks'][task]: continue
for status in stat['AgentJobInfo'][agent]['tasks'][task]['status']:
info = stat['AgentJobInfo'][agent]['tasks'][task]['status'][status]
#print status,stat['AgentJobInfo'][agent]['tasks'][task]['status'][status]
if type(info)==dict:
status_per_task[task][status] += sum( stat['AgentJobInfo'][agent]['tasks'][task]['status'][status].values())
else:
status_per_task[task][status] += stat['AgentJobInfo'][agent]['tasks'][task]['status'][status]
#print json.dumps( status_per_task, indent=2)
db_total_per_site = defaultdict(int)
db_total_per_code = defaultdict(int)
## cannot do that since there is no task count in dashboard and we have to take away the submitted
#for site in dashb:
# for error in dashb[site]:
# db_total_per_site[site] += dashb[site][error]
# db_total_per_code[code] += dashb[site][error]
print "ACDC Information"
print json.dumps( where_to_run , indent=2)
print json.dumps(missing_to_run , indent=2)
print json.dumps(missing_to_run_at , indent=2)
task_error_site_count ={}
one_explanation = defaultdict(set)
do_JL = True
do_CL = True
do_all_error_code = False
if options:
do_JL = not options.no_JL
do_CL = not options.no_CL
do_all_error_code = options.all_errors
if high_order_acdc>=1:
print high_order_acdc,"order request, pulling down all logs"
do_all_error_code = True
n_expose = 1
if options:
n_expose = options.expose
expose_archive_code = {'134':defaultdict(lambda : n_expose),#seg fault
'139':defaultdict(lambda : n_expose),# ???
'99109':defaultdict(lambda : n_expose),#stageout
'99303' : defaultdict(lambda : n_expose),#no pkl report. if you are lucky
'60450' : defaultdict(lambda : n_expose),#new
'50513':defaultdict(lambda : n_expose),#new
'8001': defaultdict(lambda : n_expose),# the usual exception in cmsRun
'11003': defaultdict(lambda : n_expose),# job extraction
'73': defaultdict(lambda : n_expose),# job extraction
}
expose_condor_code = {'99109':defaultdict(lambda : n_expose),#stageout
'99303':defaultdict(lambda : n_expose),#no pkl report
'60450':defaultdict(lambda : n_expose),#new
'50513':defaultdict(lambda : n_expose),#new
'11003': defaultdict(lambda : n_expose),
}
tasks = sorted(set(err.keys() + missing_to_run.keys()))
if not tasks:
print "no task to look at"
#return task_error_site_count
html="<html> <center><h1><a href=https://cmsweb.cern.ch/reqmgr2/fetch?rid=%s>%s</a><br><a href=https://dmytro.web.cern.ch/dmytro/cmsprodmon/workflows.php?prep_id=%s>%s</a><br>"%(
wfn,
wfn,
wfi.request['PrepID'],
wfi.request['PrepID']
)
if wfi.request['RequestType'] in ['ReReco']:
html += '<a href=../datalumi/lumi.%s.html>Lumisection Summary</a><br>'% wfi.request['PrepID']
html+= '</center><hr>'
if prim:
html+='Reads in primary<br>'
for dataset in prim:
html +='<b>%s</b>'%dataset
available = getDatasetBlocksFraction(url, dataset)
html +='<br><br>Available %.2f (>1 more than one copy, <1 not in full on disk)<br>'% available
html +='<ul>'
presence = getDatasetPresence(url, dataset)
for site in sorted(presence.keys()):
html += '<li>%s : %.2f %%'%( site, presence[site][1] )
html+='</ul><br>'
if sec:
html+='Reads in secondary<br>'
for dataset in sec:
presence = getDatasetPresence(url, dataset)
html +='<b>%s</b><ul>'%dataset
for site in sorted(presence.keys()):
html += '<li>%s : %.2f %%'%( site, presence[site][1] )
html+='</ul>'
html += "Updated on %s (GMT)" % ( time.asctime(time.gmtime()) )
html += """
<ul>
<li> <b><i>dashboard numbers over %d days</b></i>
<li> ↑ %% with respect to total number of error in the code
<li> → %% with respect to total number of error at the site
</ul>
"""%(dash_board_h)
html += '<hr><br>'
if tasks:
min_rank = min([task.count('/') for task in tasks])
for task in tasks:
#print task
task_rank = task.count('/')
task_short = task.split('/')[-1]
total_per_site = defaultdict(int)
for agent in stat['AgentJobInfo']:
if not task in stat['AgentJobInfo'][agent]['tasks']: continue
if not 'sites' in stat['AgentJobInfo'][agent]['tasks'][task]:continue
for site in stat['AgentJobInfo'][agent]['tasks'][task]['sites']:
info = stat['AgentJobInfo'][agent]['tasks'][task]['sites'][site]
#if site in ['T2_BE_IIHE']: print task,json.dumps( info, indent=2)
#print info.keys()
for s in ['success','failure','cooloff','submitted']:
if not s in info: continue
data = info[s]
#print s,data
if type(data)==dict:
total_per_site[site] += sum( data.values() )
else:
total_per_site[site] += data
#is the task relevant to recover (discard log, cleanup)
if any([v in task.lower() for v in ['logcol','cleanup']]): continue
total_count= defaultdict(int)
error_site_count = defaultdict( lambda : defaultdict(int))
if not task in err:
print task,"has not reported error"
err[task] = {}
#print err[task].keys()
for exittype in err[task]:
#print "\t",err[task][exittype].keys()
for errorcode_s in err[task][exittype]:
if errorcode_s == '0' : continue
#print "\t\t",err[task][exittype][errorcode_s].keys()
for site in err[task][exittype][errorcode_s]:
ce = SI.SE_to_CE(site)
count = err[task][exittype][errorcode_s][site]['errorCount']
total_count[errorcode_s] += count
#error_site_count[errorcode_s][site] += count
error_site_count[errorcode_s][ce] += count
for sample in err[task][exittype][errorcode_s][site]['samples']:
#print sample.keys()
for step in sample['errors']:
for report in sample['errors'][step]:
if report['type'] == 'CMSExeption': continue
#if int(report['exitCode']) == int(errorcode_s):
one_explanation[errorcode_s].add("%s (Exit code: %s) \n%s"%(report['type'], report['exitCode'], report['details']))
#one_explanation[errorcode_s].add( report['details'] )
#else:
#one_explanation[
agent = sample['agent_name']
wmbs = sample['wmbsid']
workflow = sample['workflow']
if do_CL and ((errorcode_s in expose_condor_code and expose_condor_code[errorcode_s][agent]) or do_all_error_code) and 'cern' in agent:
os.system('ssh %s %s/WmAgentScripts/Unified/exec_expose.sh %s %s %s %s %s %s'%( agent, base_dir, workflow, wmbs, errorcode_s, base_dir, monitor_dir, task_short))
if errorcode_s in expose_condor_code:
expose_condor_code[errorcode_s][agent]-=1
for out in sample['output']:
#print out
if out['type'] == 'logArchive':
if do_JL and ((errorcode_s in expose_archive_code and expose_archive_code[errorcode_s][agent]) or (do_all_error_code)):
if errorcode_s in expose_archive_code:
expose_archive_code[errorcode_s][agent]-=1
os.system('mkdir -p /tmp/%s'%(os.getenv('USER')))
local = '/tmp/%s/%s'%(os.getenv('USER'),out['lfn'].split('/')[-1])
command = 'xrdcp root://cms-xrd-global.cern.ch/%s %s'%( out['lfn'], local)
## get the file
os.system( command )
## if this actually fail, let's get the file from eos using the new log mapping
## expose the content
label=out['lfn'].split('/')[-1].split('.')[0]
m_dir = '%s/joblogs/%s/%s/%s/%s'%(monitor_dir,
wfn,
errorcode_s,
task_short,
label)
os.system('mkdir -p %s'%(m_dir))
os.system('tar zxvf %s -C %s'%(local,m_dir))
## truncate the content ??
for fn in os.popen('find %s -type f'%(m_dir)).read().split('\n'):
if not fn: continue
if any([p in fn for p in ['stdout.log']]):
trunc = '/tmp/%s/%s'%(os.getenv('USER'), label)
#print fn
#print trunc
head = tail = 1000
os.system('(head -%d ; echo;echo;echo "<snip>";echo;echo ; tail -%d ) < %s > %s'%(head, tail, fn, trunc))
os.system('mv %s %s'%(trunc, fn))
#print task
#print json.dumps( total_count, indent=2)
#print json.dumps( explanations , indent=2)
all_sites = set()
all_codes = set()
for code in error_site_count:
for site in error_site_count[code]:
all_sites.add( site )
if code != '0':
all_codes.add( code)
## parse the dashboard data
for site in total_by_site_dash:
## no. cannot discriminate by task in dashboard...
#all_sites.add( site )
pass
## parse the acdc data
notreported='NotReported'
all_missing_stats = set()
for site in missing_to_run_at[task]:
if not missing_to_run_at[task][site]: continue
ce = SI.SE_to_CE( site )
#all_sites.add( ce )
all_missing_stats.add( ce )
error_site_count[notreported][ce] = 0
all_codes.add(notreported)
## no error code at that point
all_missing_stats = all_missing_stats &set(SI.all_sites)
all_not_reported = all_missing_stats - all_sites
#print task
#print "site with no report",sorted(all_not_reported)
#print sorted(all_sites)
#print sorted(all_missing_stats)
all_sites = all_missing_stats | all_sites
all_sites = all_sites & set(SI.all_sites)
#success = total_count['0']
#total_jobs = sum(total_count.values())
#print total_jobs,"jobs in total,",success,"successes"
#miss = "{:,}".format(missing_to_run[task]) if task in missing_to_run else "N/A"
## show the total
s_per_code =defaultdict(int)
for site in all_sites:
for code in sorted(all_codes):
s_per_code[code] += error_site_count[code][site]
#no_error = (sum(s_per_code.values())==0)
no_error = len(all_not_reported)!=0
if not no_error and notreported in all_codes:
all_codes.remove( notreported )
missing_events = missing_to_run[task] if task in missing_to_run else 0
html += "<b>%s</b>"%task.split('/')[-1]
if missing_events:
html += " is missing <b>%s events</b>"%( "{:,}".format(missing_events) )
if no_error:
html +="<br><b><font color=red> and has UNreported error</font></b>"
html += "<br><table border=1><thead><tr><th>Sites/Errors</th>"
#for site in all_sites:
# html+='<th>%s</th>'%site
for code in sorted(all_codes):
html+='<th><a href="#%s">%s</a>'%(code,code)
if str(code) in expose_archive_code or do_all_error_code:
html += ' <a href=../joblogs/%s/%s/%s>, JL</a>'%( wfn, code, task_short )
if str(code) in expose_condor_code or do_all_error_code:
html += ' <a href=../condorlogs/%s/%s/%s>, CL</a>'%( wfn, code, task_short )
html += '</th>'
html+='<th>Total jobs</th><th>Site Ready</th>'
html+='</tr></thead>\n'
html+='<tr><td>Total</td>'
for code in sorted(all_codes):
html += '<td bgcolor=orange width=100>%d'%(s_per_code[code])
if code in total_by_code_dash:
html += ' (<b><i>%d</i></b>)'% total_by_code_dash[code]
html += '</td>'
ulist='<ul>'
grand=0
for status in sorted(status_per_task[task].keys()):
ulist+='<li> %s %d'%( status, status_per_task[task][status])
grand+= status_per_task[task][status]
ulist+='<li><b> Total %d </b>'%grand
ulist+='</ul>'
#html += '<td bgcolor=orange> %.2f%% </td>'% (100.*(float(sum(s_per_code.values()))/sum(total_per_site.values())) if sum(total_per_site.values()) else 0.)
html += '<td bgcolor=orange> → %.2f%% ← </td>'% (100.*(float(sum(s_per_code.values()))/ grand) if grand else 0.)
html += '<td bgcolor=orange> %s </td>'% ulist
html+='</tr>'
def palette(frac):
_range = {
0.0 : 'green',
0.5 : 'green',
0.6 : 'darkgreen',
0.7 : 'orange',
0.8 : 'salmon',
0.9 : 'red'
}
which = [k for k in _range.keys() if k<=frac]
if which:
there = max(which)
else:
there=max(_range.keys())
return _range[there]
for site in sorted(all_sites):
site_in = 'Yes'
color = 'bgcolor=lightblue'
if not site in SI.sites_ready:
color = 'bgcolor=indianred'
site_in ='<b>No</b>'
if missing_to_run_at[task][SI.CE_to_SE(site)] == 0 or min_rank == task_rank:
color = 'bgcolor=aquamarine'
site_in = '<b>No</b> but fine'
if not no_error:
site_in +=" (%s events)"%"{:,}".format(missing_to_run_at[task][SI.CE_to_SE(site)])
html+='<tr><td %s>%s</td>'%(color,site)
for code in sorted(all_codes):
if code == notreported:
html += '<td %s width=200>%s events </td>' %(color, "{:,}".format(missing_to_run_at[task][SI.CE_to_SE(site)]))
else:
if error_site_count[code][site]:
er_frac = float(error_site_count[code][site])/s_per_code[code] if s_per_code[code] else 0.
si_frac = float(error_site_count[code][site])/total_per_site[site] if total_per_site[site] else 0.
html += '<td %s width=200>%d'%(color, error_site_count[code][site])
if code in r_dashb and site in r_dashb[code]:
html += ' (<b><i>%d</i></b>)'%( r_dashb[code][site] )
html += ', <font color=%s>↑ %.1f%%</font>, <font color=%s>→ %.1f%%</font></td>'% (
palette(er_frac),100.*er_frac,
palette(si_frac), 100.*si_frac
)
else:
html += '<td %s>0</td>'% color
html += '<td bgcolor=orange>%d</td>'% total_per_site[site]
html += '<td %s>%s</td>'% (color, site_in)
html +='</tr>\n'
html+='</table><br>'
task_error_site_count[task] = error_site_count
html += '<hr><br>'
html += "<b>Blocks (%d/%d) needed for recovery</b><br>"%( len(needed_blocks), len(all_blocks))
for block in sorted(needed_blocks):
html +='%s<br>'%block
html += "<br><b>Files in no block</b><br>"
for f in sorted(files_notin_dbs):
html +='%s<br>'%f
html += '<hr><br>'
html += '<table border=1>'
for code in one_explanation:
html +='<tr><td><a name="%s">%s</a></td><td>%s</td></tr>'% ( code, code, '<br><br>'.join(one_explanation[code]).replace('\n','<br>' ))
#explanations[code].update( one_explanation[code] )
html+='</table>'
html+=('<br>'*30)
html +='</html>'
wfi.sendLog( 'error', html, show=False)
fn = '%s'% wfn
open('%s/report/%s'%(monitor_dir,fn),'w').write( html )
return task_error_site_count, one_explanation
#!/usr/bin/env python
from utils import getDatasetPresence, getDatasetEventsAndLumis
import json
import sys
url = 'cmsweb.cern.ch'
ev, lumi = getDatasetEventsAndLumis(sys.argv[1])
if lumi:
print ev, lumi, ev / float(lumi)
presence = getDatasetPresence(url, sys.argv[1])
print json.dumps(presence, indent=2)
def assignor(url, specific=None, talk=True, options=None):
if userLock(): return
if duplicateLock(): return
#if notRunningBefore( 'stagor' ): return
if not componentInfo().check(): return
UC = unifiedConfiguration()
CI = campaignInfo()
SI = global_SI
#LI = lockInfo()
NLI = newLockInfo()
n_assigned = 0
n_stalled = 0
wfos = []
if specific or options.early:
wfos.extend(
session.query(Workflow).filter(
Workflow.status == 'considered').all())
wfos.extend(
session.query(Workflow).filter(Workflow.status == 'staging').all())
if specific:
wfos.extend(
session.query(Workflow).filter(
Workflow.status == 'considered-tried').all())
wfos.extend(
session.query(Workflow).filter(Workflow.status == 'staged').all())
#if specific:
# #wfos = session.query(Workflow).filter(Workflow.name==specific).all()
# wfos = session.query(Workflow).filter(Workflow.name.contains(specific)).all()
#if not wfos:
# if specific:
# wfos = session.query(Workflow).filter(Workflow.status=='considered').all()
# wfos.extend( session.query(Workflow).filter(Workflow.status=='staging').all())
# wfos.extend(session.query(Workflow).filter(Workflow.status=='staged').all())
random.shuffle(wfos)
for wfo in wfos:
if options.limit and (n_stalled + n_assigned) > options.limit:
break
if specific:
if not any(map(lambda sp: sp in wfo.name, specific.split(','))):
continue
#if not specific in wfo.name: continue
print "\n\n"
wfh = workflowInfo(url, wfo.name)
wfh.sendLog('assignor', "%s to be assigned" % wfo.name)
## check if by configuration we gave it a GO
if not CI.go(wfh.request['Campaign']) and not options.go:
wfh.sendLog('assignor', "No go for %s" % wfh.request['Campaign'])
n_stalled += 1
continue
## check on current status for by-passed assignment
if wfh.request['RequestStatus'] != 'assignment-approved':
if not options.test:
wfh.sendLog('assignor',
"setting %s away and skipping" % wfo.name)
## the module picking up from away will do what is necessary of it
wfo.wm_status = wfh.request['RequestStatus']
wfo.status = 'away'
session.commit()
continue
else:
print wfo.name, wfh.request['RequestStatus']
## retrieve from the schema, dbs and reqMgr what should be the next version
version = wfh.getNextVersion()
if not version:
if options and options.ProcessingVersion:
version = options.ProcessingVersion
else:
wfh.sendLog('assignor', "cannot decide on version number")
n_stalled += 1
wfo.status = 'trouble'
session.commit()
continue
## the site whitelist takes into account siteInfo, campaignInfo, memory and cores
(lheinput, primary, parent, secondary,
sites_allowed) = wfh.getSiteWhiteList()
original_sites_allowed = copy.deepcopy(sites_allowed)
wfh.sendLog('assignor', "Site white list %s" % sorted(sites_allowed))
override_sec_location = CI.get(wfh.request['Campaign'],
'SecondaryLocation', [])
blocks = []
if 'BlockWhitelist' in wfh.request:
blocks = wfh.request['BlockWhitelist']
if 'RunWhitelist' in wfh.request and wfh.request['RunWhitelist']:
## augment with run white list
for dataset in primary:
blocks = list(
set(blocks + getDatasetBlocks(
dataset, runs=wfh.request['RunWhitelist'])))
wfh.sendLog('assignor', "Allowed %s" % sorted(sites_allowed))
secondary_locations = None
for sec in list(secondary):
if override_sec_location:
print "We don't care where the secondary is"
print "Cannot pass for now"
sendEmail("tempting to pass sec location check",
"but we cannot yet IMO")
#pass
presence = getDatasetPresence(url, sec)
print sec
print json.dumps(presence, indent=2)
one_secondary_locations = [
site for (site, (there, frac)) in presence.items()
if frac > 98.
]
#one_secondary_locations = [site for (site,(there,frac)) in presence.items() if there]
if secondary_locations == None:
secondary_locations = one_secondary_locations
else:
secondary_locations = list(
set(secondary_locations) & set(one_secondary_locations))
## reduce the site white list to site with secondary only
#sites_allowed = [site for site in sites_allowed if any([osite.startswith(site) for osite in one_secondary_locations])]
sites_allowed = [
site for site in sites_allowed
if SI.CE_to_SE(site) in one_secondary_locations
]
wfh.sendLog(
'assignor', "From secondary requirement, now Allowed%s" %
sorted(sites_allowed))
initial_sites_allowed = copy.deepcopy(
sites_allowed
) ## keep track of this, after secondary input location restriction : that's how you want to operate it
sites_all_data = copy.deepcopy(sites_allowed)
sites_with_data = copy.deepcopy(sites_allowed)
sites_with_any_data = copy.deepcopy(sites_allowed)
primary_locations = None
available_fractions = {}
set_lfn = '/store/mc' ## by default
for prim in list(primary):
set_lfn = getLFNbase(prim)
presence = getDatasetPresence(url, prim, only_blocks=blocks)
if talk:
print prim
print json.dumps(presence, indent=2)
available_fractions[prim] = getDatasetBlocksFraction(
url,
prim,
sites=[SI.CE_to_SE(site) for site in sites_allowed],
only_blocks=blocks)
#sites_all_data = [site for site in sites_with_data if any([osite.startswith(site) for osite in [psite for (psite,(there,frac)) in presence.items() if there]])]
#sites_with_data = [site for site in sites_with_data if any([osite.startswith(site) for osite in [psite for (psite,frac) in presence.items() if frac[1]>90.]])]
sites_all_data = [
site for site in sites_with_data if SI.CE_to_SE(site) in [
psite for (psite, (there, frac)) in presence.items()
if there
]
]
sites_with_data = [
site for site in sites_with_data if SI.CE_to_SE(site) in
[psite for (psite, frac) in presence.items() if frac[1] > 90.]
]
sites_with_any_data = [
site for site in sites_with_any_data
if SI.CE_to_SE(site) in presence.keys()
]
wfh.sendLog(
'assignor', "Holding the data but not allowed %s" % sorted(
list(
set([
se_site for se_site in presence.keys()
if not SI.SE_to_CE(se_site) in sites_allowed
]))))
if primary_locations == None:
primary_locations = presence.keys()
else:
primary_locations = list(
set(primary_locations) & set(presence.keys()))
sites_with_data = list(set(sites_with_data))
sites_with_any_data = list(set(sites_with_any_data))
opportunistic_sites = []
down_time = False
## opportunistic running where any piece of data is available
if secondary_locations or primary_locations:
## intersection of both any pieces of the primary and good IO
#opportunistic_sites = [SI.SE_to_CE(site) for site in list((set(secondary_locations) & set(primary_locations) & set(SI.sites_with_goodIO)) - set(sites_allowed))]
if secondary_locations and primary_locations:
opportunistic_sites = [
SI.SE_to_CE(site) for site in
list((set(secondary_locations) & set(primary_locations)) -
set([SI.CE_to_SE(site) for site in sites_allowed]))
]
elif primary_locations:
opportunistic_sites = [
SI.SE_to_CE(site) for site in list(
set(primary_locations) -
set([SI.CE_to_SE(site) for site in sites_allowed]))
]
else:
opportunistic_sites = []
wfh.sendLog(
'assignor', "We could be running in addition at %s" %
sorted(opportunistic_sites))
if any(
[osite in SI.sites_not_ready
for osite in opportunistic_sites]):
wfh.sendLog(
'assignor', "One of the usable site is in downtime %s" % ([
osite in SI.sites_not_ready
for osite in opportunistic_sites
]))
down_time = True
## should this be send back to considered ?
"""
if available_fractions and not all([available>=1. for available in available_fractions.values()]):
print "The input dataset is not located in full over sites"
print json.dumps(available_fractions)
if not options.test and not options.go:
known = []
try:
known = json.loads(open('cannot_assign.json').read())
except:
pass
if not wfo.name in known:
sendEmail( "cannot be assigned","%s is not full over sites \n %s"%(wfo.name,json.dumps(available_fractions)))
known.append( wfo.name )
open('cannot_assign.json','w').write(json.dumps( known, indent=2))
n_stalled+=1
continue ## skip skip skip
"""
## should be 2 but for the time-being let's lower it to get things going
copies_wanted, cpuh = wfh.getNCopies()
if 'Campaign' in wfh.request and wfh.request[
'Campaign'] in CI.campaigns and 'maxcopies' in CI.campaigns[
wfh.request['Campaign']]:
copies_needed_from_campaign = CI.campaigns[
wfh.request['Campaign']]['maxcopies']
copies_wanted = min(copies_needed_from_campaign, copies_wanted)
if not options.early:
less_copies_than_requested = UC.get("less_copies_than_requested")
copies_wanted = max(
1, copies_wanted -
less_copies_than_requested) # take one out for the efficiency
wfh.sendLog('assignor',
"needed availability fraction %s" % copies_wanted)
## should also check on number of sources, if large enough, we should be able to overflow most, efficiently
if available_fractions and not all([
available >= copies_wanted
for available in available_fractions.values()
]):
not_even_once = not all([
available >= 1. for available in available_fractions.values()
])
wfh.sendLog(
'assignor',
"The input dataset is not available %s times, only %s" %
(copies_wanted, available_fractions.values()))
if down_time and not options.go and not options.early:
wfo.status = 'considered'
session.commit()
wfh.sendLog(
'assignor',
"sending back to considered because of site downtime, instead of waiting"
)
sendEmail(
"cannot be assigned due to downtime",
"%s is not sufficiently available, due to down time of a site in the whitelist. check the assignor logs. sending back to considered."
% wfo.name)
continue
#pass
print json.dumps(available_fractions)
if (options.go and not_even_once) or not options.go:
known = []
try:
known = json.loads(open('cannot_assign.json').read())
except:
pass
if not wfo.name in known and not options.limit and not options.go and not options.early:
wfh.sendLog(
'assignor',
"cannot be assigned, %s is not sufficiently available.\n %s"
% (wfo.name, json.dumps(available_fractions)))
sendEmail(
"cannot be assigned",
"%s is not sufficiently available.\n %s" %
(wfo.name, json.dumps(available_fractions)))
known.append(wfo.name)
open('cannot_assign.json',
'w').write(json.dumps(known, indent=2))
n_stalled += 1
if options.early:
if wfo.status == 'considered':
wfh.sendLog('assignor', "setting considered-tried")
wfo.status = 'considered-tried'
session.commit()
else:
print "tried but status is", wfo.status
continue
## default back to white list to original white list with any data
print "Allowed", sites_allowed
if options.primary_aaa:
sites_allowed = initial_sites_allowed
#options.useSiteListAsLocation = True
options.TrustSitelists = True
else:
sites_allowed = sites_with_any_data
wfh.sendLog('assignor',
"Selected for any data %s" % sorted(sites_allowed))
if options.restrict:
print "Allowed", sites_allowed
sites_allowed = sites_with_any_data
print "Selected", sites_allowed
else:
if set(sites_with_data) != set(sites_allowed):
## the data is not everywhere we wanted to run at : enable aaa
print "Sites with 90% data not matching site white list (block choping!)"
print "Resorting to AAA reading for", list(
set(sites_allowed) - set(sites_with_data)), "?"
print "Whitelist site with any data", list(
set(sites_allowed) - set(sites_with_any_data))
#options.useSiteListAsLocation = True
#print "Not commissioned yet"
#continue
#print "We could be running at",opportunistic_sites,"in addition"
##sites_allowed = list(set(sites_allowed+ opportunistic_sites))
if not len(sites_allowed):
wfh.sendLog('assignor', "cannot be assign with no matched sites")
sendEmail("cannot be assigned", "%s has no whitelist" % (wfo.name))
n_stalled += 1
continue
t1_only = [ce for ce in sites_allowed if ce.startswith('T1')]
if t1_only:
# try to pick from T1 only first
sites_out = [SI.pick_dSE([SI.CE_to_SE(ce) for ce in t1_only])]
else:
# then pick any otherwise
sites_out = [
SI.pick_dSE([SI.CE_to_SE(ce) for ce in sites_allowed])
]
## one last modification now that we know we can assign, and to make sure all ressource can be used by the request : set all ON sites to whitelist
###sites_allowed = original_sites_allowed ## not needed, afterall as secondary jobs go their own ways
wfh.sendLog('assignor', "Placing the output on %s" % sites_out)
parameters = {
'SiteWhitelist': sites_allowed,
#'CustodialSites' : sites_custodial,
'NonCustodialSites': sites_out,
'AutoApproveSubscriptionSites': list(set(sites_out)),
'AcquisitionEra': wfh.acquisitionEra(),
'ProcessingString': wfh.processingString(),
'MergedLFNBase': set_lfn,
'ProcessingVersion': version,
}
## plain assignment here
team = 'production'
if os.getenv('UNIFIED_TEAM'): team = os.getenv('UNIFIED_TEAM')
if options and options.team:
team = options.team
## high priority team agent
#if wfh.request['RequestPriority'] >= 100000 and (wfh.request['TimePerEvent']*int(wfh.getRequestNumEvents()))/(8*3600.) < 10000:
# team = 'highprio'
# sendEmail("sending work with highprio team","%s"% wfo.name, destination=['*****@*****.**'])
## SDSC redirection
#if "T2_US_UCSD" in sites_with_data and random.random() < -0.5 and wfh.request['Campaign']=='RunIISpring15DR74' and int(wfh.getRequestNumEvents()) < 600000 and not any([out.endswith('RAW') for out in wfh.request['OutputDatasets']]):
# ## consider SDSC
# parameters['SiteWhitelist'] = ['T2_US_UCSD','T3_US_SDSC']
# parameters['useSiteListAsLocation'] = True
# team = 'allocation-based'
# sendEmail("sending work to SDSC","%s was assigned to SDSC/UCSD"% wfo.name, destination=['*****@*****.**'])
## SDSC redirection
#if wfh.request['Campaign']==R'unIIWinter15GS' and random.random() < -1.0:
# parameters['SiteWhitelist'] = ['T3_US_SDSC']
# team = 'allocation-based'
# sendEmail("sending work to SDSC","%s was assigned to SDSC"% wfo.name, destination=['*****@*****.**'])
if False and 'T2_CH_CERN' in parameters['SiteWhitelist']:
## add some check on
### the amount pending to HLT
### the size of the request
### the priority of the request (maybe not if we decide to overflow during runs)
parameters['SiteWhitelist'] = ['T2_CH_CERN_HLT']
team = 'hlt'
## reduce the splitting by factor of 4, regardless of type of splitting
sendEmail("sending work to HLT",
"%s was assigned to HLT" % wfo.name)
##parse options entered in command line if any
if options:
for key in reqMgrClient.assignWorkflow.keys:
v = getattr(options, key)
if v != None:
if type(v) == str and ',' in v:
parameters[key] = filter(None, v.split(','))
else:
parameters[key] = v
## pick up campaign specific assignment parameters
parameters.update(CI.parameters(wfh.request['Campaign']))
if not options.test:
parameters['execute'] = True
split_check = wfh.checkWorkflowSplitting()
if split_check != True:
parameters.update(split_check)
if 'EventBased' in split_check.values():
wfh.sendLog('assignor', "Falling back to event splitting.")
sendEmail(
"Fallback to EventBased",
"the workflow %s is too heavy to be processed as it is. Fallback to EventBased splitting"
% wfo.name)
elif 'EventsPerJob' in split_check.values():
wfh.sendLog('assignor',
"Modifying the number of job per event")
sendEmail(
"Modifying the job per events",
"the workflow %s is too heavy in number of jobs explosion"
% wfo.name)
# Handle run-dependent MC
pstring = wfh.processingString()
if 'PU_RD' in pstring:
numEvents = wfh.getRequestNumEvents()
eventsPerLumi = [getDatasetEventsPerLumi(prim) for prim in primary]
eventsPerLumi = sum(eventsPerLumi) / float(len(eventsPerLumi))
reqJobs = 500
if 'PU_RD2' in pstring:
reqJobs = 2000
eventsPerJob = int(numEvents / (reqJobs * 1.4))
lumisPerJob = int(eventsPerJob / eventsPerLumi)
if lumisPerJob == 0:
sendEmail(
"issue with event splitting for run-dependent MC",
"%s needs to be split by event with %s per job" %
(wfo.name, eventsPerJob))
wfh.sendLog(
'assignor',
"%s needs to be split by event with %s per job" %
(wfo.name, eventsPerJob))
parameters['EventsPerJob'] = eventsPerJob
else:
spl = wfh.getSplittings()[0]
eventsPerJobEstimated = spl[
'events_per_job'] if 'events_per_job' in spl else None
eventsPerJobEstimated = spl[
'avg_events_per_job'] if 'avg_events_per_job' in spl else None
if eventsPerJobEstimated and eventsPerJobEstimated > eventsPerJob:
sendEmail(
"setting lumi splitting for run-dependent MC",
"%s was assigned with %s lumis/job" %
(wfo.name, lumisPerJob))
wfh.sendLog(
'assignor', "%s was assigned with %s lumis/job" %
(wfo.name, lumisPerJob))
parameters['LumisPerJob'] = lumisPerJob
else:
sendEmail("leaving splitting untouched for PU_RD*",
"please check on " + wfo.name)
wfh.sendLog(
'assignor',
"leaving splitting untouched for PU_RD*, please check."
)
result = reqMgrClient.assignWorkflow(url, wfo.name, team, parameters)
# set status
if not options.test:
if result:
wfo.status = 'away'
session.commit()
n_assigned += 1
try:
## refetch information and lock output
new_wfi = workflowInfo(url, wfo.name)
(_, prim, _, sec) = new_wfi.getIO()
for secure in list(prim) + list(
sec) + new_wfi.request['OutputDatasets']:
## lock all outputs flat
NLI.lock(secure)
#for site in [SI.CE_to_SE(site) for site in sites_allowed]:
# for output in new_wfi.request['OutputDatasets']:
# LI.lock( output, site, 'dataset in production')
# for primary in prim:
# LI.lock( primary, site, 'dataset used in input')
# for secondary in sec:
# LI.lock( secondary, site, 'required for mixing' )
except Exception as e:
print "fail in locking output"
print str(e)
sendEmail("failed locking of output", str(e))
else:
print "ERROR could not assign", wfo.name
else:
pass
print "Assignment summary:"
sendLog('assignor', "Assigned %d Stalled %s" % (n_assigned, n_stalled))
def checkor(url, spec=None, options=None):
if userLock(): return
if duplicateLock(): return
fDB = closeoutInfo()
UC = unifiedConfiguration()
use_mcm = True
up = componentInfo(mcm=use_mcm, soft=['mcm'])
if not up.check(): return
use_mcm = up.status['mcm']
wfs=[]
if options.new:
## get all in running and check
## you want to intersect with what is completed !
if options.strict:
completed_wfi = getWorkflows(url, status='completed')
for wfo in session.query(Workflow).filter(Workflow.status == 'away').all():
if wfo.name in completed_wfi:
wfs.append( wfo )
else:
print wfo.name,"is not completed"
sendLog('checkor','%s is not completed'%( wfo.name))
else:
wfs.extend( session.query(Workflow).filter(Workflow.status == 'away').all() )
if options.current:
## recheck those already there, probably to just pass them along
wfs.extend( session.query(Workflow).filter(Workflow.status== 'assistance').all() )
if options.old:
## than get all in need for assistance
wfs.extend( session.query(Workflow).filter(Workflow.status.startswith('assistance-')).all() )
custodials = defaultdict(list) #sites : dataset list
transfers = defaultdict(list) #sites : dataset list
invalidations = [] #a list of files
SI = global_SI
CI = campaignInfo()
mcm = McMClient(dev=False)
def get_campaign(output, wfi):
campaign = None
try:
campaign = output.split('/')[2].split('-')[0]
except:
if 'Campaign' in wfi.request:
campaign = wfi.request['Campaign']
return campaign
## retrieve bypass and onhold configuration
bypasses = []
holdings = []
#try:
# already_notified = json.loads(open('already_notifified.json').read())
#except:
# print "no record of already notified workflow. starting fresh"
# already_notified = []
for bypassor,email in [('vlimant','*****@*****.**'),('jen_a','*****@*****.**')]:
bypass_file = '/afs/cern.ch/user/%s/%s/public/ops/bypass.json'%(bypassor[0],bypassor)
if not os.path.isfile(bypass_file):
#sendLog('checkor','no file %s',bypass_file)
continue
try:
bypasses.extend( json.loads(open(bypass_file).read()))
except:
sendLog('checkor',"cannot get by-passes from %s for %s"%(bypass_file ,bypassor))
sendEmail("malformated by-pass information","%s is not json readable"%(bypass_file), destination=[email])
holding_file = '/afs/cern.ch/user/%s/%s/public/ops/onhold.json'%(bypassor[0],bypassor)
if not os.path.isfile(holding_file):
#sendLog('checkor',"no file %s"%holding_file)
continue
try:
holdings.extend( json.loads(open(holding_file).read()))
except:
sendLog('checkor',"cannot get holdings from %s for %s"%(holding_file, bypassor))
sendEmail("malformated by-pass information","%s is not json readable"%(holding_file), destination=[email])
## once this was force-completed, you want to bypass
for rider,email in [('vlimant','*****@*****.**'),('jen_a','*****@*****.**'),('srimanob','*****@*****.**')]:
rider_file = '/afs/cern.ch/user/%s/%s/public/ops/forcecomplete.json'%(rider[0],rider)
if not os.path.isfile(rider_file):
print "no file",rider_file
#sendLog('checkor',"no file %s"%rider_file)
continue
try:
bypasses.extend( json.loads(open( rider_file ).read() ) )
except:
sendLog('checkor',"cannot get force complete list from %s"%rider)
sendEmail("malformated force complet file","%s is not json readable"%rider_file, destination=[email])
if use_mcm:
mcm_force = mcm.get('/restapi/requests/forcecomplete')
bypasses.extend( mcm_force )
pattern_fraction_pass = UC.get('pattern_fraction_pass')
total_running_time = 5.*60.
sleep_time = 1
if len(wfs):
sleep_time = min(max(0.5, total_running_time / len(wfs)), 10)
random.shuffle( wfs )
print len(wfs),"to consider, pausing for",sleep_time
for wfo in wfs:
if spec and not (spec in wfo.name): continue
time.sleep( sleep_time )
## get info
wfi = workflowInfo(url, wfo.name)
wfi.sendLog('checkor',"checking on %s %s"%( wfo.name,wfo.status))
## make sure the wm status is up to date.
# and send things back/forward if necessary.
wfo.wm_status = wfi.request['RequestStatus']
if wfo.wm_status == 'closed-out':
## manually closed-out
wfi.sendLog('checkor',"%s is already %s, setting close"%( wfo.name , wfo.wm_status))
wfo.status = 'close'
session.commit()
continue
elif wfo.wm_status in ['failed','aborted','aborted-archived','rejected','rejected-archived','aborted-completed']:
## went into trouble
wfo.status = 'trouble'
wfi.sendLog('checkor',"%s is in trouble %s"%(wfo.name, wfo.wm_status))
session.commit()
continue
elif wfo.wm_status in ['assigned','acquired']:
## not worth checking yet
wfi.sendLog('checkor',"%s is not running yet"%wfo.name)
session.commit()
continue
if '-onhold' in wfo.status:
if wfo.name in holdings and wfo.name not in bypasses:
wfi.sendLog('checkor',"%s is on hold"%wfo.name)
continue
if wfo.wm_status != 'completed': #and not wfo.name in bypasses:
## for sure move on with closeout check if in completed
wfi.sendLog('checkor',"no need to check on %s in status %s"%(wfo.name, wfo.wm_status))
session.commit()
continue
if wfo.name in holdings and wfo.name not in bypasses:
wfo.status = 'assistance-onhold'
wfi.sendLog('checkor',"setting %s on hold"%wfo.name)
session.commit()
continue
session.commit()
#sub_assistance="" # if that string is filled, there will be need for manual assistance
existing_assistance_tags = set(wfo.status.split('-')[1:]) #[0] should be assistance
assistance_tags = set()
is_closing = True
## get it from somewhere
bypass_checks = False
pids = wfi.getPrepIDs()
bypass_by_mcm = False
for bypass in bypasses:
if bypass in wfo.name:
wfi.sendLog('checkor',"we can bypass checks on %s because of keyword %s "%( wfo.name, bypass))
bypass_checks = True
break
if bypass in pids:
wfi.sendLog('checkor',"we can bypass checks on %s because of prepid %s "%( wfo.name, bypass))
bypass_checks = True
bypass_by_mcm = True
break
#if not CI.go( wfi.request['Campaign'] ) and not bypass_checks:
# print "No go for",wfo.name
# wfi.sendLog('checkor',"No go for %s"%wfi.request['Campaign'])
# continue
tiers_with_no_check = copy.deepcopy(UC.get('tiers_with_no_check')) # dqm*
vetoed_custodial_tier = copy.deepcopy(UC.get('tiers_with_no_custodial')) #dqm*, reco
campaigns = {}
for out in wfi.request['OutputDatasets']:
c = get_campaign(out, wfi)
campaigns[out] = c
if c in CI.campaigns and 'custodial_override' in CI.campaigns[c]:
vetoed_custodial_tier = list(set(vetoed_custodial_tier) - set(CI.campaigns[c]['custodial_override']))
## add those that we need to check for custodial copy
tiers_with_no_check = list(set(tiers_with_no_check) - set(CI.campaigns[c]['custodial_override'])) ## would remove DQM from the vetoed check
check_output_text = "Initial outputs:"+",".join(sorted(wfi.request['OutputDatasets'] ))
wfi.request['OutputDatasets'] = [ out for out in wfi.request['OutputDatasets'] if not any([out.split('/')[-1] == veto_tier for veto_tier in tiers_with_no_check])]
check_output_text += "\nWill check on:"+",".join(sorted(wfi.request['OutputDatasets'] ))
check_output_text += "\ntiers out:"+",".join( sorted(tiers_with_no_check ))
check_output_text += "\ntiers no custodial:"+",".join( sorted(vetoed_custodial_tier) )
wfi.sendLog('checkor', check_output_text )
## anything running on acdc : getting the real prepid is not worth it
familly = getWorkflowById(url, wfi.request['PrepID'], details=True)
acdc = []
acdc_inactive = []
for member in familly:
if member['RequestType'] != 'Resubmission': continue
if member['RequestName'] == wfo.name: continue
if member['RequestDate'] < wfi.request['RequestDate']: continue
if member['RequestStatus'] in ['running-open','running-closed','assigned','acquired']:
print wfo.name,"still has an ACDC running",member['RequestName']
acdc.append( member['RequestName'] )
## cannot be bypassed!
is_closing = False
assistance_tags.add('recovering')
elif member['RequestStatus']==None:
print member['RequestName'],"is not real"
pass
else:
acdc_inactive.append( member['RequestName'] )
assistance_tags.add('recovered')
## completion check
percent_completions = {}
if not 'TotalInputEvents' in wfi.request:
event_expected,lumi_expected = 0,0
if not 'recovery' in wfo.status:
#sendEmail("missing member of the request","TotalInputEvents is missing from the workload of %s"% wfo.name, destination=['*****@*****.**'])
sendLog('checkor',"TotalInputEvents is missing from the workload of %s"% wfo.name, level='critical')
else:
event_expected,lumi_expected = wfi.request['TotalInputEvents'],wfi.request['TotalInputLumis']
if 'RequestNumEvents' in wfi.request and int(wfi.request['RequestNumEvents']):
event_expected = int(wfi.request['RequestNumEvents'])
elif 'Task1' in wfi.request and 'RequestNumEvents' in wfi.request['Task1']:
event_expected = int(wfi.request['Task1']['RequestNumEvents'])
fractions_pass = {}
over_100_pass = False
(lhe,prim,_,_) = wfi.getIO()
if lhe or prim: over_100_pass = False
for output in wfi.request['OutputDatasets']:
event_count,lumi_count = getDatasetEventsAndLumis(dataset=output)
percent_completions[output] = 0.
if lumi_expected:
percent_completions[output] = lumi_count / float( lumi_expected )
if event_expected:
wfi.sendLog('checkor', "event completion real %s expected %s"%(event_count, event_expected ))
percent_completions[output] = max(percent_completions[output], float(event_count) / float( event_expected ) )
fractions_pass[output] = 0.95
c = campaigns[output]
if c in CI.campaigns and 'fractionpass' in CI.campaigns[c]:
fractions_pass[output] = CI.campaigns[c]['fractionpass']
wfi.sendLog('checkor', "overriding fraction to %s for %s by campaign requirement"%( fractions_pass[output], output))
if options.fractionpass:
fractions_pass[output] = options.fractionpass
print "overriding fraction to",fractions_pass[output],"by command line for",output
for key in pattern_fraction_pass:
if key in output:
fractions_pass[output] = pattern_fraction_pass[key]
print "overriding fraction to",fractions_pass[output],"by dataset key",key
if not all([percent_completions[out] >= fractions_pass[out] for out in fractions_pass]):
print wfo.name,"is not completed"
print json.dumps(percent_completions, indent=2)
print json.dumps(fractions_pass, indent=2)
## hook for creating automatically ACDC ?
if not bypass_checks:
assistance_tags.add('recovery')
is_closing = False
if over_100_pass and any([percent_completions[out] >100 for out in fractions_pass]):
print wfo.name,"is over completed"
print json.dumps(percent_completions, indent=2)
if not bypass_checks:
assistance_tags.add('over100')
is_closing = False
## correct lumi < 300 event per lumi
events_per_lumi = {}
for output in wfi.request['OutputDatasets']:
events_per_lumi[output] = getDatasetEventsPerLumi( output )
lumi_upper_limit = {}
for output in wfi.request['OutputDatasets']:
upper_limit = 301.
campaign = campaigns[output]
#if 'EventsPerLumi' in wfi.request and 'FilterEfficiency' in wfi.request:
# upper_limit = 1.5*wfi.request['EventsPerLumi']*wfi.request['FilterEfficiency']
# print "setting the upper limit of lumisize to",upper_limit,"by request configuration"
if campaign in CI.campaigns and 'lumisize' in CI.campaigns[campaign]:
upper_limit = CI.campaigns[campaign]['lumisize']
print "overriding the upper lumi size to",upper_limit,"for",campaign
if options.lumisize:
upper_limit = options.lumisize
print "overriding the upper lumi size to",upper_limit,"by command line"
lumi_upper_limit[output] = upper_limit
if wfi.request['RequestType'] in ['ReDigi']: lumi_upper_limit[output] = -1
if any([ (lumi_upper_limit[out]>0 and events_per_lumi[out] >= lumi_upper_limit[out]) for out in events_per_lumi]):
print wfo.name,"has big lumisections"
print json.dumps(events_per_lumi, indent=2)
## hook for rejecting the request ?
if not bypass_checks:
assistance_tags.add('biglumi')
is_closing = False
any_presence = {}
for output in wfi.request['OutputDatasets']:
any_presence[output] = getDatasetPresence(url, output, vetoes=[])
## custodial copy
custodial_locations = {}
custodial_presences = {}
for output in wfi.request['OutputDatasets']:
custodial_presences[output] = [s for s in any_presence[output] if 'MSS' in s]
custodial_locations[output] = phedexClient.getCustodialSubscriptionRequestSite(output)
if not custodial_locations[output]:
custodial_locations[output] = []
## presence in phedex
phedex_presence ={}
for output in wfi.request['OutputDatasets']:
phedex_presence[output] = phedexClient.getFileCountDataset(url, output )
out_worth_checking = [out for out in custodial_locations.keys() if out.split('/')[-1] not in vetoed_custodial_tier]
size_worth_checking = sum([getDatasetSize(out)/1023. for out in out_worth_checking ]) ## size in TBs of all outputs
if not all(map( lambda sites : len(sites)!=0, [custodial_locations[out] for out in out_worth_checking])):
print wfo.name,"has not all custodial location"
print json.dumps(custodial_locations, indent=2)
##########
## hook for making a custodial replica ?
custodial = None
## get from other outputs
for output in out_worth_checking:
if len(custodial_locations[output]):
custodial = custodial_locations[output][0]
if custodial and float(SI.storage[custodial]) < size_worth_checking:
print "cannot use the other output custodial:",custodial,"because of limited space"
custodial = None
## try to get it from campaign configuration
if not custodial:
for output in out_worth_checking:
campaign = campaigns[output]
if campaign in CI.campaigns and 'custodial' in CI.campaigns[campaign]:
custodial = CI.campaigns[campaign]['custodial']
print "Setting custodial to",custodial,"from campaign configuration"
if custodial and float(SI.storage[custodial]) < size_worth_checking:
print "cannot use the campaign configuration custodial:",custodial,"because of limited space"
custodial = None
## get from the parent
pick_custodial = True
use_parent_custodial = UC.get('use_parent_custodial')
_,prim,_,_ = wfi.getIO()
if not custodial and prim and use_parent_custodial:
parent_dataset = prim.pop()
## this is terribly dangerous to assume only
parents_custodial = phedexClient.getCustodialSubscriptionRequestSite( parent_dataset )
###parents_custodial = findCustodialLocation(url, parent_dataset)
if not parents_custodial:
parents_custodial = []
if len(parents_custodial):
custodial = parents_custodial[0]
else:
print "the input dataset",parent_dataset,"does not have custodial in the first place. abort"
#sendEmail( "dataset has no custodial location", "Please take a look at %s in the logs of checkor"%parent_dataset)
## does not work for RAWOADSIM
sendLog('checkor',"Please take a look at %s for missing custodial location"% parent_dataset)
## cannot be bypassed, this is an issue to fix
is_closing = False
pick_custodial = False
assistance_tags.add('parentcustodial')
if custodial and float(SI.storage[custodial]) < size_worth_checking:
print "cannot use the parent custodial:",custodial,"because of limited space"
custodial = None
if not custodial and pick_custodial:
## pick one at random
custodial = SI.pick_SE(size=size_worth_checking)
if not custodial:
print "cannot find a custodial for",wfo.name
wfi.sendLog('checkor',"cannot find a custodial for %s probably because of the total output size %d"%( wfo.name, size_worth_checking))
#sendEmail( "cannot find a custodial","cannot find a custodial for %s probably because of the total output size %d"%( wfo.name, size_worth_checking))
sendLog('checkor',"cannot find a custodial for %s probably because of the total output size %d"%( wfo.name, size_worth_checking), level='critical')
if custodial and (is_closing or bypass_checks):
print "picked",custodial,"for tape copy"
## remember how much you added this round already ; this stays locally
SI.storage[custodial] -= size_worth_checking
## register the custodial request, if there are no other big issues
for output in out_worth_checking:
if not len(custodial_locations[output]):
if phedex_presence[output]>=1:
custodials[custodial].append( output )
## let's wait and see if that's needed
assistance_tags.add('custodial')
else:
print "no file in phedex for",output," not good to add to custodial requests"
#cannot be bypassed
is_closing = False
## disk copy
disk_copies = {}
for output in wfi.request['OutputDatasets']:
disk_copies[output] = [s for s in any_presence[output] if (not 'MSS' in s) and (not 'Buffer' in s)]
if not all(map( lambda sites : len(sites)!=0, disk_copies.values())):
print wfo.name,"has not all output on disk"
print json.dumps(disk_copies, indent=2)
## presence in dbs
dbs_presence = {}
dbs_invalid = {}
for output in wfi.request['OutputDatasets']:
dbs_presence[output] = dbs3Client.getFileCountDataset( output )
dbs_invalid[output] = dbs3Client.getFileCountDataset( output, onlyInvalid=True)
fraction_invalid = 0.01
if not all([dbs_presence[out] == (dbs_invalid[out]+phedex_presence[out]) for out in wfi.request['OutputDatasets']]) and not options.ignorefiles:
print wfo.name,"has a dbs,phedex mismatch"
print json.dumps(dbs_presence, indent=2)
print json.dumps(dbs_invalid, indent=2)
print json.dumps(phedex_presence, indent=2)
if not 'recovering' in assistance_tags:
assistance_tags.add('filemismatch')
#print this for show and tell if no recovery on-going
for out in dbs_presence:
_,_,missing_phedex,missing_dbs = getDatasetFiles(url, out)
if missing_phedex:
print "These %d files are missing in phedex"%(len(missing_phedex))
print "\n".join( missing_phedex )
if missing_dbs:
print "These %d files are missing in dbs"%(len(missing_dbs))
print "\n".join( missing_dbs )
#if not bypass_checks:
## I don't think we can by pass this
is_closing = False
if not all([(dbs_invalid[out] <= int(fraction_invalid*dbs_presence[out])) for out in wfi.request['OutputDatasets']]) and not options.ignorefiles:
print wfo.name,"has a dbs invalid file level too high"
print json.dumps(dbs_presence, indent=2)
print json.dumps(dbs_invalid, indent=2)
print json.dumps(phedex_presence, indent=2)
## need to be going and taking an eye
assistance_tags.add('invalidfiles')
if not bypass_checks:
#sub_assistance+="-invalidfiles"
is_closing = False
## put that heavy part at the end
## duplication check
duplications = {}
if is_closing or bypass_checks:
print "starting duplicate checker for",wfo.name
for output in wfi.request['OutputDatasets']:
print "\tchecking",output
duplications[output] = True
try:
duplications[output] = dbs3Client.duplicateRunLumi( output , skipInvalid=True, verbose=True)
except:
try:
duplications[output] = dbs3Client.duplicateRunLumi( output , skipInvalid=True, verbose=True)
except:
print "was not possible to get the duplicate count for",output
is_closing=False
if any(duplications.values()) and not options.ignoreduplicates:
print wfo.name,"has duplicates"
print json.dumps(duplications,indent=2)
## hook for making file invalidation ?
## it shouldn't be allowed to bypass it
assistance_tags.add('duplicates')
is_closing = False
## for visualization later on
if not wfo.name in fDB.record:
#print "adding",wfo.name,"to close out record"
fDB.record[wfo.name] = {
'datasets' :{},
'name' : wfo.name,
'closeOutWorkflow' : None,
}
fDB.record[wfo.name]['closeOutWorkflow'] = is_closing
fDB.record[wfo.name]['priority'] = wfi.request['RequestPriority']
fDB.record[wfo.name]['prepid'] = wfi.request['PrepID']
for output in wfi.request['OutputDatasets']:
if not output in fDB.record[wfo.name]['datasets']: fDB.record[wfo.name]['datasets'][output] = {}
rec = fDB.record[wfo.name]['datasets'][output]
rec['percentage'] = float('%.2f'%(percent_completions[output]*100))
rec['duplicate'] = duplications[output] if output in duplications else 'N/A'
rec['phedexReqs'] = float('%.2f'%any_presence[output][custodial_presences[output][0]][1]) if len(custodial_presences[output])!=0 else 'N/A'
rec['closeOutDataset'] = is_closing
rec['transPerc'] = float('%.2f'%any_presence[output][ disk_copies[output][0]][1]) if len(disk_copies[output])!=0 else 'N/A'
rec['correctLumis'] = int(events_per_lumi[output]) if (events_per_lumi[output] > lumi_upper_limit[output]) else True
rec['missingSubs'] = False if len(custodial_locations[output])==0 else ','.join(list(set(custodial_locations[output])))
rec['dbsFiles'] = dbs_presence[output]
rec['dbsInvFiles'] = dbs_invalid[output]
rec['phedexFiles'] = phedex_presence[output]
rec['acdc'] = "%d / %d"%(len(acdc),len(acdc+acdc_inactive))
## and move on
if is_closing:
## toggle status to closed-out in request manager
print "setting",wfo.name,"closed-out"
if not options.test:
if wfo.wm_status in ['closed-out','announced','normal-archived']:
print wfo.name,"is already",wfo.wm_status,"not trying to closed-out and assuming it does"
res = None
else:
res = reqMgrClient.closeOutWorkflowCascade(url, wfo.name)
print "close out answer",res
if not res in ["None",None]:
print "try to get the current status again"
wfi_bis = workflowInfo(url, wfo.name)
if wfi_bis.request['RequestStatus'] == 'closed-out':
print "the request did toggle to closed-out"
res = None
if not res in ["None",None]:
print "retrying to closing out"
print res
res = reqMgrClient.closeOutWorkflowCascade(url, wfo.name)
if res in [None,"None"]:
wfo.status = 'close'
session.commit()
if use_mcm and bypass_by_mcm:
## shoot large on all prepids
for pid in pids:
mcm.delete('/restapi/requests/forcecomplete/%s'%pid)
else:
print "could not close out",wfo.name,"will try again next time"
else:
## full known list
#recovering # has active ACDC
##OUT #recovered #had inactive ACDC
#recovery #not over the pass bar
#over100 # over 100%
#biglumi # has a big lumiblock
#parentcustodial # the parent does not have a valid subscription yet
#custodial # has had the transfer made, is waiting for a valid custodial subscription to appear
#filemismatch # there is a dbs/phedex mismatch
#duplicates #a lumi section is there twice
## manual is not added yet, and should be so by recoveror
print wfo.name,"was tagged with :",list(assistance_tags)
if 'recovering' in assistance_tags:
## if active ACDC, being under threshold, filemismatch do not matter
assistance_tags = assistance_tags - set(['recovery','filemismatch'])
if 'recovery' in assistance_tags and 'recovered' in assistance_tags:
## should not set -recovery to anything that add ACDC already
assistance_tags = assistance_tags - set(['recovery','recovered'])
## straight to manual
assistance_tags.add('manual')
## that means there is something that needs to be done acdc, lumi invalidation, custodial, name it
print wfo.name,"needs assistance with",",".join( assistance_tags )
print wfo.name,"existing conditions",",".join( existing_assistance_tags )
#########################################
##### notification to requester #########
go_notify=False
if assistance_tags and not 'manual' in existing_assistance_tags and existing_assistance_tags != assistance_tags:
go_notify=True
if go_notify:
#if wfo.name in already_notified:
# print "double notification"
# sendEmail('double notification','please take a look at %s'%(wfo.name))
#else:
# already_notified.append( wfo.name )
detailslink = 'https://cmsweb.cern.ch/reqmgr/view/details/%s'
perflink = 'https://cmsweb.cern.ch/couchdb/workloadsummary/_design/WorkloadSummary/_show/histogramByWorkflow/%s'%(wfo.name)
splitlink = 'https://cmsweb.cern.ch/reqmgr/view/splitting/%s'%(wfo.name)
## notify templates
messages= {
'recovery': 'Samples completed with missing statistics:\n%s\n%s '%( '\n'.join(['%.2f %% complete for %s'%(percent_completions[output]*100, output) for output in wfi.request['OutputDatasets'] ] ), perflink ),
'biglumi': 'Samples completed with large luminosity blocks:\n%s\n%s '%('\n'.join(['%d > %d for %s'%(events_per_lumi[output], lumi_upper_limit[output], output) for output in wfi.request['OutputDatasets'] if (events_per_lumi[output] > lumi_upper_limit[output])]), splitlink),
'duplicates': 'Samples completed with duplicated luminosity blocks:\n%s\n'%( '\n'.join(['%s'%output for output in wfi.request['OutputDatasets'] if output in duplications and duplications[output] ] ) ),
'filemismatch': 'Samples completed with inconsistency in DBS/Phedex',
#'manual' : 'Workflow completed and requires manual checks by Ops',
}
content = "The request PREPID (WORKFLOW) is facing issue in production.\n"
motive = False
for case in messages:
if case in assistance_tags:
content+= "\n"+messages[case]+"\n"
motive = True
content += "You are invited to check, while this is being taken care of by Comp-Ops.\n"
content += "This is an automated message from Comp-Ops.\n"
items_notified = set()
if use_mcm and motive:
wfi.notifyRequestor( content , mcm = mcm)
#########################################
## logic to set the status further
if assistance_tags:
new_status = 'assistance-'+'-'.join(sorted(assistance_tags) )
else:
new_status = 'assistance'
## case where the workflow was in manual from recoveror
if not 'manual' in wfo.status or new_status!='assistance-recovery':
wfo.status = new_status
if not options.test:
print "setting",wfo.name,"to",wfo.status
session.commit()
else:
print "current status is",wfo.status,"not changing to anything"
#open('already_notifified.json','w').write( json.dumps( already_notified , indent=2))
fDB.html()
if not spec:
#sendEmail("fresh assistance status available","Fresh status are available at https://cmst2.web.cern.ch/cmst2/unified/assistance.html",destination=['*****@*****.**'])
#it's a bit annoying
pass
## custodial requests
print "Custodials"
print json.dumps(custodials, indent=2)
for site in custodials:
print ','.join(custodials[site]),'=>',site
if not options.test:
result = makeReplicaRequest(url, site, list(set(custodials[site])),"custodial copy at production close-out",custodial='y',priority='low', approve = (site in SI.sites_auto_approve) )
print result
print "Transfers"
print json.dumps(transfers, indent=2)
## replicas requests
for site in transfers:
print ','.join(transfers[site]),'=>',site
if not options.test:
result = None
#result = makeReplicaRequest(url, site, list(set(transfers[site])),"copy to disk at production close-out")
print result
print "File Invalidation"
print invalidations
def transferor(url, specific=None, talk=True, options=None):
if userLock(): return
mlock = moduleLock()
if mlock(): return
use_mcm = True
up = componentInfo(soft=['mcm', 'wtc', 'jira'])
if not up.check(): return
use_mcm = up.status['mcm']
if options and options.test:
execute = False
else:
execute = True
SI = siteInfo()
CI = campaignInfo()
#NLI = newLockInfo()
#if not NLI.free(): return
LI = lockInfo()
if not LI.free(): return
mcm = McMClient(dev=False)
dss = DSS()
#allowed_secondary = UC.get('')
print "counting all being handled..."
being_handled = len(
session.query(Workflow).filter(Workflow.status == 'away').all())
being_handled += len(
session.query(Workflow).filter(
Workflow.status.startswith('stag')).all())
being_transfered = len(
session.query(Workflow).filter(Workflow.status == 'staging').all())
#being_handled += len(session.query(Workflow).filter(Workflow.status.startswith('assistance-')).all())
being_handled += len(
session.query(Workflow).filter(
Workflow.status.startswith('assistance-')).filter(
~Workflow.status.contains('custodial')).all())
max_to_handle = options.maxworkflows
max_to_transfer = options.maxstaging
allowed_to_handle = max(0, max_to_handle - being_handled)
allowed_to_transfer = max(0, max_to_transfer - being_transfered)
wf_buffer = 5
if allowed_to_handle <= wf_buffer: ## buffer for having several wf per transfer
print "Not allowed to run more than", max_to_handle, "at a time. Currently", being_handled, "and", wf_buffer, "buffer"
else:
print being_handled, "already being handled", max_to_handle, "max allowed,", allowed_to_handle, "remaining", "and", wf_buffer, "buffer"
if allowed_to_transfer <= wf_buffer:
print "Not allowed to transfer more than", max_to_transfer, "at a time. Currently", being_transfered, "and", wf_buffer, "buffer"
else:
print being_transfered, "already being transfered", max_to_transfer, "max allowed,", allowed_to_transfer, "remaining", "and", wf_buffer, "buffer"
print "... done"
all_transfers = defaultdict(list)
workflow_dependencies = defaultdict(
set) ## list of wf.id per input dataset
wfs_and_wfh = []
max_per_round = UC.get('max_per_round').get('transferor', None)
print "getting all wf to consider ..."
cache = getWorkflows(url, 'assignment-approved', details=True)
all_to_include = session.query(Workflow).filter(
Workflow.status.startswith('considered')).all()
if len(cache) > 2000:
max_to_include = max_per_round
random.shuffle(cache) ## randomize first by wf name
cache = sorted(cache, key=lambda r: r['RequestPriority'],
reverse=True) ## order by prio
highest = [r['RequestName'] for r in cache[:max_to_include]]
all_to_include = [wfo for wfo in all_to_include if wfo.name in highest]
print "limiting what to consider to", max_to_include, "because there is too much stuff going on. Got", len(
all_to_include)
for wfo in all_to_include:
print "\t", wfo.name
if specific and not specific in wfo.name: continue
cache_r = filter(lambda d: d['RequestName'] == wfo.name, cache)
if len(cache_r):
wfs_and_wfh.append((wfo,
workflowInfo(url,
wfo.name,
spec=False,
request=cache_r[0])))
else:
wfs_and_wfh.append((wfo, workflowInfo(url, wfo.name, spec=False)))
print "... done"
transfers_per_sites = defaultdict(int)
input_sizes = defaultdict(float)
ignored_input_sizes = defaultdict(float)
input_cput = {}
input_st = {}
## list the size of those in transfer already
in_transfer_priority = None
min_transfer_priority = None
print "getting all wf in staging ..."
#stucks = json.loads(open('%s/stuck_transfers.json'%monitor_pub_dir).read())
stucks = json.loads(eosRead('%s/stuck_transfers.json' % monitor_pub_dir))
for wfo in session.query(Workflow).filter(
Workflow.status == 'staging').all():
wfh = workflowInfo(url, wfo.name, spec=False)
#(lheinput,primary,parent,secondary) = wfh.getIO()
#sites_allowed = getSiteWhiteList( (lheinput,primary,parent,secondary) )
print wfo.name, "staging"
(lheinput, primary, parent, secondary,
sites_allowed) = wfh.getSiteWhiteList()
for site in sites_allowed: ## we should get the actual transfer destination instead of the full white list
transfers_per_sites[site] += 1
#input_cput[wfo.name] = wfh.getComputingTime()
#input_st[wfo.name] = wfh.getSystemTime()
blocks = wfh.getBlocks()
for prim in primary:
ds_s = dss.get(prim, blocks=blocks)
if prim in stucks:
wfh.sendLog(
'transferor',
"%s appears stuck, so not counting it %s [GB]" %
(prim, ds_s))
ignored_input_sizes[prim] = max(ds_s,
ignored_input_sizes[prim])
else:
input_sizes[prim] = max(ds_s, input_sizes[prim])
wfh.sendLog('transferor',
"%s needs %s [GB]" % (wfo.name, ds_s))
if in_transfer_priority == None:
in_transfer_priority = int(wfh.request['RequestPriority'])
else:
in_transfer_priority = max(in_transfer_priority,
int(wfh.request['RequestPriority']))
if min_transfer_priority == None:
min_transfer_priority = int(wfh.request['RequestPriority'])
else:
min_transfer_priority = min(min_transfer_priority,
int(wfh.request['RequestPriority']))
try:
print "Ignored input sizes"
ignored_values = list(ignored_input_sizes.items())
ignored_values.sort(key=lambda i: i[1])
print "\n".join(map(str, ignored_values))
print "Considered input sizes"
considered_values = list(input_sizes.items())
considered_values.sort(key=lambda i: i[1])
print "\n".join(map(str, considered_values))
except Exception as e:
print "trying to print the summary of input size"
print str(e)
print "... done"
print "Max priority in transfer already", in_transfer_priority
print "Min priority in transfer already", min_transfer_priority
print "transfers per sites"
print json.dumps(transfers_per_sites, indent=2)
in_transfer_already = sum(input_sizes.values())
cput_in_transfer_already = sum(input_cput.values())
st_in_transfer_already = sum(input_st.values())
## list the size of all inputs
primary_input_per_workflow_gb = defaultdict(float)
print "getting all input sizes ..."
input_blocks = {}
for (wfo, wfh) in wfs_and_wfh:
(_, primary, _, _) = wfh.getIO()
blocks = wfh.getBlocks()
input_blocks[wfo.name] = blocks
for prim in primary:
## do not count it if it appears stalled !
prim_size = dss.get(prim, blocks=blocks)
input_sizes[prim] = max(prim_size, input_sizes[prim])
primary_input_per_workflow_gb[wfo.name] += prim_size
print "... done"
# shuffle first by name
random.shuffle(wfs_and_wfh)
# Sort smallest transfers first; allows us to transfer as many as possible workflows.
def prio_and_size(i, j):
if int(i[1].request['RequestPriority']) == int(
j[1].request['RequestPriority']):
return cmp(int(primary_input_per_workflow_gb.get(j[0].name, 0)),
int(primary_input_per_workflow_gb.get(i[0].name, 0)))
else:
return cmp(int(i[1].request['RequestPriority']),
int(j[1].request['RequestPriority']))
#wfs_and_wfh.sort(cmp = prio_and_size, reverse=True)
#wfs_and_wfh.sort(cmp = lambda i,j : cmp(int(primary_input_per_workflow_gb.get(i[0].name, 0)), int(primary_input_per_workflow_gb.get(j[0].name, 0)) ))
#sort by priority higher first
wfs_and_wfh.sort(cmp=lambda i, j: cmp(int(i[1].request[
'RequestPriority']), int(j[1].request['RequestPriority'])),
reverse=True)
if min_transfer_priority == None or in_transfer_priority == None:
print "nothing is lining up for transfer"
sendLog(
"transferor",
"No request in staging, using first request to set priority limit")
if len(wfs_and_wfh):
min_transfer_priority = wfs_and_wfh[0][1].request[
'RequestPriority']
in_transfer_priority = wfs_and_wfh[0][1].request['RequestPriority']
else:
return
cput_grand_total = sum(input_cput.values())
cput_to_transfer = cput_grand_total - cput_in_transfer_already
st_grand_total = sum(input_st.values())
st_to_transfer = st_grand_total - st_in_transfer_already
print "%15.4f [CPU h] worth already in transfer" % cput_in_transfer_already
print "%15.4f [CPU h] worth is the current requested transfer load" % cput_to_transfer
print "%15.4f [h] worth of absolute system time in transfer" % (
cput_in_transfer_already / SI.availableSlots())
print "%15.4f [h] worth of absolute system time is the current requested transfer load" % (
cput_to_transfer / SI.availableSlots())
print "%15.4f [h] worth of theoritical system time in transfer" % (
st_in_transfer_already)
print "%15.4f [h] worth of theoritical system time is the current requested transfer load" % (
st_to_transfer)
grand_total = sum(input_sizes.values())
to_transfer = grand_total - in_transfer_already
grand_transfer_limit = options.maxtransfer
#grand_transfer_limit = SI.total_disk()*0.25*1024## half of the free sapce in TB->GB
transfer_limit = grand_transfer_limit - in_transfer_already
print "%15.4f GB already being transfered" % in_transfer_already
print "%15.4f GB is the current requested transfer load" % to_transfer
print "%15.4f GB is the global transfer limit" % grand_transfer_limit
print "%15.4f GB is the available limit" % transfer_limit
max_staging_per_site = options.maxstagingpersite
# the max priority value per dataset.
max_priority = defaultdict(int)
needs_transfer = 0 ## so that we can count'em
passing_along = 0
transfer_sizes = defaultdict(float)
went_over_budget = False
destination_cache = {}
no_goes = set()
if max_per_round and not spec:
wfs_and_wfh = wfs_and_wfh[:max_per_round]
for (wfo, wfh) in wfs_and_wfh:
print wfo.name, "to be transfered with priority", wfh.request[
'RequestPriority']
if wfh.request['RequestStatus'] != 'assignment-approved':
if wfh.request['RequestStatus'] in [
'aborted', 'rejected', 'rejected-archived',
'aborted-archived'
]:
if wfh.isRelval():
wfo.status = 'forget'
else:
wfo.status = 'trouble' ## so that we look or a replacement
else:
wfo.status = 'away'
wfh.sendLog(
'transferor', '%s in status %s, setting %s' %
(wfo.name, wfh.request['RequestStatus'], wfo.status))
continue
(lheinput, primary, parent, secondary,
sites_allowed) = wfh.getSiteWhiteList()
blocks = input_blocks.get(wfo.name, wfh.getBlocks())
if blocks:
print "Reading only", len(blocks), "blocks in input"
this_load = sum([dss.get(prim, blocks=blocks) for prim in primary])
no_budget = False
if (this_load
and (sum(transfer_sizes.values()) + this_load > transfer_limit
or went_over_budget)):
if went_over_budget:
wfh.sendLog('transferor', "Transfer has gone over bubget.")
else:
wfh.sendLog('transferor', "Transfer will go over bubget.")
wfh.sendLog(
'transferor',
"%15.4f GB this load, %15.4f GB already this round, %15.4f GB is the available limit"
% (this_load, sum(transfer_sizes.values()), transfer_limit))
#if sum(transfer_sizes.values()) > transfer_limit:
went_over_budget = True
if in_transfer_priority != None and min_transfer_priority != None:
if int(
wfh.request['RequestPriority']
) >= in_transfer_priority and min_transfer_priority != in_transfer_priority:
wfh.sendLog(
'transferor',
"Higher priority sample %s >= %s go-on over budget" %
(wfh.request['RequestPriority'], in_transfer_priority))
else:
if not options.go:
wfh.sendLog(
'transferor',
"%s minimum priority %s < %s : stop" %
(min_transfer_priority,
wfh.request['RequestPriority'],
in_transfer_priority))
no_budget = True
## throtlle by campaign go
no_go = False
if not wfh.go(log=True) and not options.go:
no_go = True
no_goes.add(wfo.name)
allowed_secondary = {}
overide_parameters = {}
check_secondary = (not wfh.isRelval())
output_tiers = list(
set([o.split('/')[-1] for o in wfh.request['OutputDatasets']]))
for campaign in wfh.getCampaigns():
if campaign in CI.campaigns:
overide_parameters.update(CI.campaigns[campaign])
if campaign in CI.campaigns and 'secondaries' in CI.campaigns[
campaign]:
if CI.campaigns[campaign]['secondaries']:
allowed_secondary.update(
CI.campaigns[campaign]['secondaries'])
check_secondary = True
if campaign in CI.campaigns and 'banned_tier' in CI.campaigns[
campaign]:
banned_tier = list(
set(CI.campaigns[campaign]['banned_tier'])
& set(output_tiers))
if banned_tier:
no_go = True
wfh.sendLog(
'transferor', 'These data tiers %s are not allowed' %
(','.join(banned_tier)))
sendLog('transferor',
'These data tiers %s are not allowed in %s' %
(','.join(banned_tier), wfo.name),
level='critical')
if secondary and check_secondary:
if (set(secondary) & set(allowed_secondary.keys()) !=
set(secondary)):
msg = '%s is not an allowed secondary' % (
', '.join(set(secondary) - set(allowed_secondary.keys())))
wfh.sendLog('transferor', msg)
critical_msg = msg + '\nWorkflow URL: https://dmytro.web.cern.ch/dmytro/cmsprodmon/workflows.php?prep_id=task_{}'.format(
wfh.getPrepIDs()[0])
sendLog('transferor', critical_msg, level='critical')
if not options.go:
no_go = True
for sec in secondary:
if sec in allowed_secondary:
overide_parameters.update(allowed_secondary[sec])
if 'SiteWhitelist' in overide_parameters:
sites_allowed = list(
set(sites_allowed) & set(overide_parameters['SiteWhitelist']))
wfh.sendLog(
'transferor',
'Intersecting with the overriding whitelist parameters, allowed sites become {}'
.format(sites_allowed))
if no_go:
continue
if passing_along >= allowed_to_handle:
#if int(wfh.request['RequestPriority']) >= in_transfer_priority and min_transfer_priority!=in_transfer_priority:
if in_transfer_priority != None and min_transfer_priority != None:
if int(wfh.request['RequestPriority']
) >= in_transfer_priority and int(
wfh.request['RequestPriority']
) != min_transfer_priority:
## higher priority, and not only this priority being transfered
wfh.sendLog(
'transferor',
"Higher priority sample %s >= %s go-on over %s" %
(wfh.request['RequestPriority'], in_transfer_priority,
max_to_handle))
else:
wfh.sendLog(
'transferor',
" Not allowed to pass more than %s at a time. Currently %s handled, and adding %s"
% (max_to_handle, being_handled, passing_along))
if not options.go:
## should not allow to jump that fence
break
if this_load and needs_transfer >= allowed_to_transfer:
if in_transfer_priority != None and min_transfer_priority != None:
if int(wfh.request['RequestPriority']
) >= in_transfer_priority and int(
wfh.request['RequestPriority']
) != min_transfer_priority:
## higher priority, and not only this priority being transfered
wfh.sendLog(
'transferor',
"Higher priority sample %s >= %s go-on over %s" %
(wfh.request['RequestPriority'], in_transfer_priority,
max_to_transfer))
else:
wfh.sendLog(
'transferor',
"Not allowed to transfer more than %s at a time. Currently %s transfering, and adding %s"
% (max_to_transfer, being_transfered, needs_transfer))
if not options.go:
no_budget = True
if no_budget:
continue
# break ## try this for a while to make things faster
## the site white list considers site, campaign, memory and core information
if options and options.tosites:
sites_allowed = options.tosites.split(',')
for dataset in list(primary) + list(parent) + list(secondary):
LI.lock(dataset, reason='staging')
if not sites_allowed:
wfh.sendLog('transferor', "not possible site to run at")
#sendEmail("no possible sites","%s has no possible sites to run at"%( wfo.name ))
sendLog('transferor',
"%s has no possible sites to run at" % (wfo.name),
level='critical')
continue
can_go = True
staging = False
allowed = True
primary_destinations = set()
if primary:
copies_needed_from_CPUh, CPUh = wfh.getNCopies()
if talk:
print wfo.name, 'reads', ', '.join(primary), 'in primary'
## chope the primary dataset
for prim in primary:
## keep track of what needs what
workflow_dependencies[prim].add(wfo.id)
max_priority[prim] = max(max_priority[prim],
int(wfh.request['RequestPriority']))
wfh.sendLog(
'transferor', "Would make %s from cpu requirement %s" %
(copies_needed_from_CPUh, CPUh))
copies_needed = copies_needed_from_CPUh
if 'Campaign' in wfh.request and wfh.request[
'Campaign'] in CI.campaigns and 'maxcopies' in CI.campaigns[
wfh.request['Campaign']]:
copies_needed_from_campaign = CI.campaigns[
wfh.request['Campaign']]['maxcopies']
copies_needed = min(copies_needed_from_campaign,
copies_needed)
wfh.sendLog(
'transferor',
"Maxed to %s by campaign configuration %s" %
(copies_needed, wfh.request['Campaign']))
if blocks:
print "limiting to blocks", "\n".join(sorted(blocks))
### new ways of making the whole thing
destinations, all_block_names = getDatasetDestinations(
url,
prim,
within_sites=[SI.CE_to_SE(site) for site in sites_allowed],
only_blocks=blocks)
print json.dumps(destinations, indent=2)
## get where the dataset is in full and completed
prim_location = [
site for (site, info) in destinations.items()
if info['completion'] == 100 and info['data_fraction'] == 1
]
## the rest is places it is going to be
#prim_destination = [site for site in destinations.keys() if not site in prim_location]
prim_destination = [
site for (site, info) in destinations.items()
if info['data_fraction'] == 1 and info['completion'] != 100
]
## veto the site with no current disk space, for things that are not relval
prim_destination = [
site for site in prim_destination
if (SI.disk[site] or wfh.isRelval())
]
if len(prim_location) >= copies_needed:
wfh.sendLog(
'transferor',
"The input is all fully in place at %s sites %s" %
(len(prim_location), sorted(prim_location)))
continue
copies_needed = max(0, copies_needed - len(prim_location))
wfh.sendLog(
'transferor',
"Counting existing copies ; now need %s" % copies_needed)
copies_being_made = [
sum([
info['blocks'].keys().count(block)
for site, info in destinations.items()
if site in prim_destination
]) for block in all_block_names
]
latching_on_transfers = set()
[
latching_on_transfers.update(info['blocks'].values())
for site, info in destinations.items()
if site in prim_destination
]
latching_on_transfers = list(latching_on_transfers)
#print latching_on_transfers
## figure out where all this is going to go
prim_to_distribute = [
site for site in sites_allowed
if not SI.CE_to_SE(site) in prim_location
]
prim_to_distribute = [
site for site in prim_to_distribute
if not SI.CE_to_SE(site) in prim_destination
]
## take out the ones that cannot receive transfers
potential_destinations = len(prim_to_distribute)
#prim_to_distribute = [site for site in prim_to_distribute if not SI.CE_to_SE(site) in SI.sites_veto_transfer]
prim_to_distribute = [
site for site in prim_to_distribute
if (SI.disk[SI.CE_to_SE(site)] or wfh.isRelval())
]
## do we want to restrict transfers if the amount of site in vetoe are too large ?
wfh.sendLog(
'transferor',
"Could be going to: %s" % sorted(prim_to_distribute))
if not prim_to_distribute or any([
transfers_per_sites[site] < max_staging_per_site
for site in prim_to_distribute
]):
## means there is openings let me go
print "There are transfer slots available:", [
(site, transfers_per_sites[site])
for site in prim_to_distribute
]
else:
if int(
wfh.request['RequestPriority']
) >= in_transfer_priority and min_transfer_priority != in_transfer_priority:
wfh.sendLog(
'transferor',
"Higher priority sample %s >= %s go-on over transfer slots available"
% (wfh.request['RequestPriority'],
in_transfer_priority))
else:
wfh.sendLog(
'transferor',
"Not allowed to transfer more than %s per site at a time. Going overboard for %s"
% (max_staging_per_site,
sorted([
site for site in prim_to_distribute
if transfers_per_sites[site] >=
max_staging_per_site
])))
if not options.go:
allowed = False
break
for latching in latching_on_transfers:
existings = session.query(TransferImp).filter(
TransferImp.phedexid == int(latching)).filter(
TransferImp.workflow_id == wfo.id).all()
if not existings:
tri = TransferImp(phedexid=int(latching), workflow=wfo)
print "adding", wfo.id, "with phedexid", latching
session.add(tri)
else:
for existing in existings:
existing.active = True
session.flush()
can_go = False
transfer_sizes[prim] = max(this_load, transfer_sizes[prim])
staging = True
# reduce the number of copies required by the on-going full transfer : how do we bootstrap on waiting for them ??
#copies_needed = max(0,copies_needed - len(prim_destination))
copies_needed = max(0, copies_needed - min(copies_being_made))
wfh.sendLog(
'transferor',
"Counting the copies being made ; then need %s" %
copies_needed)
if copies_needed == 0:
wfh.sendLog(
'transferor',
"The input is either fully in place or getting in full somewhere with %s"
% latching_on_transfers)
can_go = True
continue
elif len(prim_to_distribute) == 0:
wfh.sendLog(
'transferor',
"We are going to need extra copies of %s, but no destinations seems available"
% (prim))
sendLog(
'transferor',
"We are going to need extra copies of %s, but no destinations seems available"
% (prim),
level='critical')
print json.dumps(prim_to_distribute, indent=2)
print json.dumps(prim_location, indent=2)
print json.dumps(prim_destination, indent=2)
prim_to_distribute = [
site for site in sites_allowed
if not SI.CE_to_SE(site) in prim_location
]
#prim_to_distribute = [site for site in prim_to_distribute if not SI.CE_to_SE(site) in SI.sites_veto_transfer ]
prim_to_distribute = [
site for site in prim_to_distribute
if (SI.disk[SI.CE_to_SE(site)] or wfh.isRelval())
]
print "changed to"
print json.dumps(prim_to_distribute, indent=2)
if len(
prim_to_distribute
) > 0: ## maybe that a parameter we can play with to limit the
if not options or options.chop:
### hard include the tape disk andpoint ?
#tapes = [site for site in getDatasetPresence( url, prim, vetos=['T0','T2','T3','Disk']) if site.endswith('MSS')]
chops, sizes = getDatasetChops(
prim,
chop_threshold=options.chopsize,
only_blocks=blocks)
spreading = distributeToSites(chops,
prim_to_distribute,
n_copies=copies_needed,
weights=SI.cpu_pledges,
sizes=sizes)
## prune the blocks/destination that are already in the making, so that subscription don't overlap
for site in spreading:
for block in list(spreading[site]):
if site in destinations and block in destinations[
site]['blocks'].keys():
## prune it
spreading[site].remove(block)
transfer_sizes[prim] = sum(sizes)
if not spreading:
sendLog(
'transferor',
'cannot send %s to any site, it cannot fit anywhere'
% prim,
level='critical')
wfh.sendLog(
'transferor',
"cannot send to any site. %s cannot seem to fit anywhere"
% (prim))
staging = False
can_go = False
else:
spreading = {}
for site in prim_to_distribute:
if blocks:
spreading[site] = blocks
else:
spreading[site] = [prim]
transfer_sizes[prim] = max(this_load,
transfer_sizes[prim])
can_go = False
wfh.sendLog(
'transferor', "selected CE destinations %s" %
(sorted(spreading.keys())))
for (site, items) in spreading.items():
all_transfers[site].extend(items)
transfers_per_sites[site] += 1
primary_destinations.add(site)
else:
can_go = False
allowed = False
if not allowed:
wfh.sendLog('transferor', "Not allowed to move on with")
continue
if secondary:
override_sec_destination = []
if 'SecondaryLocation' in CI.campaigns[wfh.request['Campaign']]:
override_sec_destination = CI.campaigns[
wfh.request['Campaign']]['SecondaryLocation']
if 'SecondaryLocation' in overide_parameters:
override_sec_destination = overide_parameters[
'SecondaryLocation']
print wfo.name, 'reads', ', '.join(secondary), 'in secondary'
for sec in secondary:
workflow_dependencies[sec].add(wfo.id)
if True:
## new style, failing on minbias
if not sec in destination_cache:
## this is barbbaric, and does not show the correct picture on workflow by workflow with different whitelist
destination_cache[sec], _ = getDatasetDestinations(
url, sec) ## NO SITE WHITE LIST ADDED
#destination_cache[sec],_ = getDatasetDestinations(url, sec, within_sites = [SI.CE_to_SE(site) for site in sites_allowed])
## limit to the site whitelist NOW
se_allowed = set(
[SI.CE_to_SE(site) for site in sites_allowed])
destinations = dict([
(k, v) for (k, v) in destination_cache[sec].items()
if k in se_allowed
])
## truncate location/destination to those making up for >90% of the dataset
bad_destinations = [
destinations.pop(site)
for (site, info) in destinations.items()
if info['data_fraction'] < 0.9
]
print sec, json.dumps(destinations, indent=2)
sec_location = [
site for (site, info) in destinations.items()
if info['completion'] >= 95
]
sec_destination = [
site for site in destinations.keys()
if not site in sec_location
] ## this is in SE
else:
## old style
presence = getDatasetPresence(url, sec)
sec_location = [
site for site, pres in presence.items()
if pres[1] > 90.
] ## more than 90% of the minbias at sites
subscriptions = listSubscriptions(url, sec)
sec_destination = [site for site in subscriptions]
## how to make unified understand that it has to wait for the secondary if the sec_destination and
#sec_to_distribute = [site for site in sites_allowed if not any([osite.startswith(site) for osite in sec_location])]
sec_to_distribute = [
site for site in sites_allowed
if not SI.CE_to_SE(site) in sec_location
]
#sec_to_distribute = [site for site in sec_to_distribute if not any([osite.startswith(site) for osite in sec_destination])]
sec_to_distribute = [
site for site in sec_to_distribute
if not SI.CE_to_SE(site) in sec_destination
]
presitespace_sec_to_distribute = copy.deepcopy(
sec_to_distribute)
#sec_to_distribute = [site for site in sec_to_distribute if not any([osite.startswith(site) for osite in SI.sites_veto_transfer])]
#sec_to_distribute = [site for site in sec_to_distribute if not SI.CE_to_SE(site) in SI.sites_veto_transfer]
sec_to_distribute = [
site for site in sec_to_distribute
if (SI.disk[SI.CE_to_SE(site)] or wfh.isRelval())
]
## at this point you have a problem
if len(sec_to_distribute) == 0 and len(
presitespace_sec_to_distribute):
sendLog(
'transferor',
'%s is getting no possible destinations because of lack of space. To be decided what to do in general'
% (sec),
level='critical')
if override_sec_destination:
## intersect with where we want the PU to be
not_needed_anymore = list(
set(sec_to_distribute) - set(override_sec_destination))
#sendEmail("secondary superfluous","the dataset %s could be removed from %s"%( sec, not_needed_anymore ))
sendLog(
'transferor',
"the dataset %s could be removed from %s" %
(sec, not_needed_anymore))
sec_to_distribute = list(
set(sec_to_distribute) & set(override_sec_destination))
if len(sec_to_distribute) > 0:
print "secondary could go to", sorted(sec_to_distribute)
sec_size = dss.get(sec)
for site in sec_to_distribute:
site_se = SI.CE_to_SE(site)
if (SI.disk[site_se] *
1024.) > sec_size or wfh.isRelval():
wfh.sendLog('transferor',
'Sending %s to %s' % (sec, site))
all_transfers[site].append(sec)
can_go = False
else:
print "could not send the secondary input to", site_se, "because it is too big for the available disk", SI.disk[
site_se] * 1024, "GB need", sec_size
if primary_destinations and site in primary_destinations:
#sendEmail('secondary input too big','%s is too big (%s) for %s (%s)'%( sec, sec_size, site_se, SI.disk[site_se]*1024))
sendLog(
'transferor',
'%s is too big (%s) for %s (%s). %s will not be able to run there.'
% (sec, sec_size, site_se,
SI.disk[site_se] * 1024, wfo.name),
level='critical')
wfh.sendLog(
'transferor',
'%s is too big (%s) for %s (%s). will not be able to run there.'
% (sec, sec_size, site_se,
SI.disk[site_se] * 1024))
else:
## this is bas overall
print "the secondary input does not have to be send to site"
## is that possible to do something more
if can_go:
## no explicit transfer required this time
if staging:
## but using existing ones
wfh.sendLog(
'transferor',
"latches on existing transfers, and nothing else, settin staging"
)
wfo.status = 'staging'
needs_transfer += 1
else:
wfh.sendLog(
'transferor', "should just be assigned now to %s" %
sorted(sites_allowed))
wfo.status = 'staged'
passing_along += 1
wfh.sendLog('transferor',
"setting %s status to %s" % (wfo.name, wfo.status))
#session.commit()
continue
else:
## there is an explicit transfer required
if staging:
## and also using an existing one
wfh.sendLog('transferor', "latches on existing transfers")
if not options.test:
wfo.status = 'staging'
wfh.sendLog(
'transferor',
"setting %s status to %s" % (wfo.name, wfo.status))
#session.commit()
wfh.sendLog('transferor', "needs a transfer")
needs_transfer += 1
passing_along += 1
if no_goes:
#sendEmail("no go for managing","No go for \n"+"\n".join( no_goes ))
sendLog('transferor',
"No go for \n" + "\n".join(sorted(no_goes)),
level='critical')
print "accumulated transfers"
print json.dumps(all_transfers, indent=2)
fake_id = -1
wf_id_in_prestaging = set()
for (site, items_to_transfer) in all_transfers.iteritems():
items_to_transfer = list(set(items_to_transfer))
## convert to storage element
site_se = SI.CE_to_SE(site)
## site that do not want input datasets
#if site in SI.sites_veto_transfer:
# print site,"does not want transfers"
# continue
## throttle the transfer size to T2s ? we'd be screwed by a noPU sample properly configured.
## massage a bit the items
blocks = [it for it in items_to_transfer if '#' in it]
block_datasets = list(set([it.split('#')[0] for it in blocks]))
datasets = [it for it in items_to_transfer if not '#' in it]
details_text = "Making a replica to %s (CE) %s (SE) for" % (site,
site_se)
#print "\t",len(blocks),"blocks"
## remove blocks if full dataset is send out
blocks = [
block for block in blocks if not block.split('#')[0] in datasets
]
#print "\t",len(blocks),"needed blocks for",list(set([block.split('#')[0] for block in blocks]))
#print "\t",len(datasets),"datasets"
#print "\t",datasets
details_text += '\n\t%d blocks' % len(blocks)
details_text += '\n\t%d needed blocks for %s' % (
len(blocks),
sorted(list(set([block.split('#')[0] for block in blocks]))))
details_text += '\n\t%d datasets' % len(datasets)
details_text += '\n\t%s' % sorted(datasets)
items_to_transfer = blocks + datasets
if execute:
sendLog('transferor', details_text)
else:
print "Would make a replica to", site, "(CE)", site_se, "(SE) for"
print details_text
## operate the transfer
if options and options.stop:
## ask to move-on
answer = raw_input('Continue with that ?')
if not answer.lower() in ['y', 'yes', 'go']:
continue
transfered_items = defaultdict(set)
if execute:
priority = 'normal'
cds = [
ds for ds in set(datasets + block_datasets)
if ds in max_priority
]
## bucketize the transfers by priority of workflows
prioritized_items = defaultdict(set)
for item in items_to_transfer:
d = item.split('#')[0]
p = max_priority.get(d, 80000)
q = 'normal'
if p > 100000:
q = 'reserved'
elif p < 70000:
q = 'low'
prioritized_items[q].add(item)
for priority, items in prioritized_items.items():
result = makeReplicaRequest(url,
site_se,
list(items),
'prestaging',
priority=priority,
approve=True)
if result:
these_transfers = [
o['id'] for o in result['phedex']['request_created']
]
#phedexids.extend( these_transfers )
for ph in these_transfers:
transfered_items[ph].update(items)
else:
sendLog(
'transferor',
'Could not make a replica request for items %s to site %s'
% (items, site_se),
level='critical')
#result = makeReplicaRequest(url, site_se, items_to_transfer, 'prestaging', priority=priority, approve=True)
#phedexids = [o['id'] for o in result['phedex']['request_created']]:
#else:
# #result= {'phedex':{'request_created' : []}}
# phedexids = []
# fake_id-=1
if not transfered_items:
sendLog(
'transferor',
'Could not make a replica request for items %s to site %s' %
(items_to_transfer, site),
level='critical')
continue
for phedexid, items in transfered_items.items():
print phedexid, "transfer created"
for transfering in list(
set(map(lambda it: it.split('#')[0], items))):
for wfid in workflow_dependencies[transfering]:
new_transfer = session.query(TransferImp).filter(
TransferImp.phedexid == int(phedexid)).filter(
TransferImp.workflow_id == wfid).first()
if not new_transfer:
new_transfer = TransferImp(
phedexid=phedexid,
workflow=session.query(Workflow).get(wfid))
session.add(new_transfer)
else:
new_transfer.active = True
wf_id_in_prestaging.add(wfid)
#session.commit()
for wfid in wf_id_in_prestaging:
tr_wf = session.query(Workflow).get(wfid)
if tr_wf and tr_wf.status != 'staging':
if execute:
tr_wf.status = 'staging'
if talk:
print "setting", tr_wf.name, "to staging"
#session.commit()
## one big session commit at the end that everything went fine
session.commit()
def checkor(url, spec=None, options=None):
if userLock(): return
if duplicateLock(): return
fDB = closeoutInfo()
UC = unifiedConfiguration()
use_mcm = True
up = componentInfo(mcm=use_mcm, soft=['mcm'])
if not up.check(): return
use_mcm = up.status['mcm']
wfs=[]
if options.new:
## get all in running and check
## you want to intersect with what is completed !
if options.strict:
completed_wfi = getWorkflows(url, status='completed')
for wfo in session.query(Workflow).filter(Workflow.status == 'away').all():
if wfo.name in completed_wfi:
wfs.append( wfo )
else:
print wfo.name,"is not completed"
sendLog('checkor','%s is not completed'%( wfo.name))
else:
wfs.extend( session.query(Workflow).filter(Workflow.status == 'away').all() )
if options.current:
## recheck those already there, probably to just pass them along
wfs.extend( session.query(Workflow).filter(Workflow.status== 'assistance').all() )
if options.old:
## than get all in need for assistance
wfs.extend( session.query(Workflow).filter(Workflow.status.startswith('assistance-')).all() )
custodials = defaultdict(list) #sites : dataset list
transfers = defaultdict(list) #sites : dataset list
invalidations = [] #a list of files
SI = siteInfo()
CI = campaignInfo()
mcm = McMClient(dev=False)
def get_campaign(output, wfi):
campaign = None
try:
campaign = output.split('/')[2].split('-')[0]
except:
if 'Campaign' in wfi.request:
campaign = wfi.request['Campaign']
return campaign
## retrieve bypass and onhold configuration
bypasses = []
forcings = []
overrides = getForceCompletes()
holdings = []
for bypassor,email in [('vlimant','*****@*****.**'),('jen_a','*****@*****.**'),('prozober','*****@*****.**')]:
bypass_file = '/afs/cern.ch/user/%s/%s/public/ops/bypass.json'%(bypassor[0],bypassor)
if not os.path.isfile(bypass_file):
#sendLog('checkor','no file %s',bypass_file)
continue
try:
bypasses.extend( json.loads(open(bypass_file).read()))
except:
sendLog('checkor',"cannot get by-passes from %s for %s"%(bypass_file ,bypassor))
sendEmail("malformated by-pass information","%s is not json readable"%(bypass_file), destination=[email])
holding_file = '/afs/cern.ch/user/%s/%s/public/ops/onhold.json'%(bypassor[0],bypassor)
if not os.path.isfile(holding_file):
#sendLog('checkor',"no file %s"%holding_file)
continue
try:
holdings.extend( json.loads(open(holding_file).read()))
except:
sendLog('checkor',"cannot get holdings from %s for %s"%(holding_file, bypassor))
sendEmail("malformated by-pass information","%s is not json readable"%(holding_file), destination=[email])
## once this was force-completed, you want to bypass
for rider,email in [('vlimant','*****@*****.**'),('jen_a','*****@*****.**'),('srimanob','*****@*****.**')]:
rider_file = '/afs/cern.ch/user/%s/%s/public/ops/forcecomplete.json'%(rider[0],rider)
if not os.path.isfile(rider_file):
print "no file",rider_file
#sendLog('checkor',"no file %s"%rider_file)
continue
try:
bypasses.extend( json.loads(open( rider_file ).read() ) )
except:
sendLog('checkor',"cannot get force complete list from %s"%rider)
sendEmail("malformated force complet file","%s is not json readable"%rider_file, destination=[email])
if use_mcm:
forcings = mcm.get('/restapi/requests/forcecomplete')
if forcings:
sendEmail('force completing mechanism','please check what checkor is doing with %s'%( ','.join(forcings)))
pattern_fraction_pass = UC.get('pattern_fraction_pass')
total_running_time = 5.*60.
sleep_time = 1
if len(wfs):
sleep_time = min(max(0.5, total_running_time / len(wfs)), 10)
random.shuffle( wfs )
print len(wfs),"to consider, pausing for",sleep_time
max_per_round = UC.get('max_per_round').get('checkor',None)
if max_per_round and not spec: wfs = wfs[:max_per_round]
for wfo in wfs:
if spec and not (spec in wfo.name): continue
time.sleep( sleep_time )
## get info
wfi = workflowInfo(url, wfo.name)
wfi.sendLog('checkor',"checking on %s %s"%( wfo.name,wfo.status))
## make sure the wm status is up to date.
# and send things back/forward if necessary.
wfo.wm_status = wfi.request['RequestStatus']
if wfo.wm_status == 'closed-out':
## manually closed-out
wfi.sendLog('checkor',"%s is already %s, setting close"%( wfo.name , wfo.wm_status))
wfo.status = 'close'
session.commit()
continue
elif wfo.wm_status in ['failed','aborted','aborted-archived','rejected','rejected-archived','aborted-completed']:
## went into trouble
wfo.status = 'trouble'
wfi.sendLog('checkor',"%s is in trouble %s"%(wfo.name, wfo.wm_status))
session.commit()
continue
elif wfo.wm_status in ['assigned','acquired']:
## not worth checking yet
wfi.sendLog('checkor',"%s is not running yet"%wfo.name)
session.commit()
continue
if '-onhold' in wfo.status:
if wfo.name in holdings and wfo.name not in bypasses:
wfi.sendLog('checkor',"%s is on hold"%wfo.name)
continue
if wfo.wm_status != 'completed': #and not wfo.name in bypasses:
## for sure move on with closeout check if in completed
wfi.sendLog('checkor',"no need to check on %s in status %s"%(wfo.name, wfo.wm_status))
session.commit()
continue
if wfo.name in holdings and wfo.name not in bypasses:
wfo.status = 'assistance-onhold'
wfi.sendLog('checkor',"setting %s on hold"%wfo.name)
session.commit()
continue
session.commit()
#sub_assistance="" # if that string is filled, there will be need for manual assistance
existing_assistance_tags = set(wfo.status.split('-')[1:]) #[0] should be assistance
assistance_tags = set()
is_closing = True
## get it from somewhere
bypass_checks = False
for bypass in bypasses:
if bypass in wfo.name:
wfi.sendLog('checkor',"we can bypass checks on %s because of keyword %s "%( wfo.name, bypass))
bypass_checks = True
break
pids = wfi.getPrepIDs()
force_by_mcm = False
force_by_user = False
for force in forcings:
if force in pids:
wfi.sendLog('checkor',"we can bypass checks and force complete %s because of prepid %s "%( wfo.name, force))
bypass_checks = True
force_by_mcm = True
break
for user in overrides:
for force in overrides[user]:
if force in wfo.name:
wfi.sendLog('checkor',"we can bypass checks and force complete %s because of keyword %s of user %s"%( wfo.name, force, user))
bypass_checks = True
force_by_user = True
break
tiers_with_no_check = copy.deepcopy(UC.get('tiers_with_no_check')) # dqm*
vetoed_custodial_tier = copy.deepcopy(UC.get('tiers_with_no_custodial')) #dqm*, reco
campaigns = {}
expected_outputs = copy.deepcopy( wfi.request['OutputDatasets'] )
for out in wfi.request['OutputDatasets']:
c = get_campaign(out, wfi)
campaigns[out] = c
if c in CI.campaigns and 'custodial_override' in CI.campaigns[c]:
vetoed_custodial_tier = list(set(vetoed_custodial_tier) - set(CI.campaigns[c]['custodial_override']))
## add those that we need to check for custodial copy
tiers_with_no_check = list(set(tiers_with_no_check) - set(CI.campaigns[c]['custodial_override'])) ## would remove DQM from the vetoed check
check_output_text = "Initial outputs:"+",".join(sorted(wfi.request['OutputDatasets'] ))
wfi.request['OutputDatasets'] = [ out for out in wfi.request['OutputDatasets'] if not any([out.split('/')[-1] == veto_tier for veto_tier in tiers_with_no_check])]
check_output_text += "\nWill check on:"+",".join(sorted(wfi.request['OutputDatasets'] ))
check_output_text += "\ntiers out:"+",".join( sorted(tiers_with_no_check ))
check_output_text += "\ntiers no custodial:"+",".join( sorted(vetoed_custodial_tier) )
wfi.sendLog('checkor', check_output_text )
## anything running on acdc : getting the real prepid is not worth it
familly = getWorkflowById(url, wfi.request['PrepID'], details=True)
acdc = []
acdc_inactive = []
forced_already=False
acdc_bads = []
for member in familly:
if member['RequestType'] != 'Resubmission': continue
if member['RequestName'] == wfo.name: continue
if member['RequestDate'] < wfi.request['RequestDate']: continue
if 'OriginalRequestName' in member and member['OriginalRequestName'] != wfo.name: continue
if member['RequestStatus'] == None: continue
if not set(member['OutputDatasets']).issubset( set(expected_outputs)):
if not member['RequestStatus'] in ['rejected-archived','rejected','aborted','aborted-archived']:
##this is not good at all
wfi.sendLog('checkor','inconsistent ACDC %s'%member['RequestName'] )
acdc_bads.append( member['RequestName'] )
is_closing = False
assistance_tags.add('manual')
continue
if member['RequestStatus'] in ['running-open','running-closed','assigned','acquired']:
print wfo.name,"still has an ACDC running",member['RequestName']
acdc.append( member['RequestName'] )
## cannot be bypassed!
is_closing = False
assistance_tags.add('recovering')
if (force_by_mcm or force_by_user) and not forced_already:
wfi.sendLog('checkor','%s is being forced completed while recovering'%wfo.name)
wfi.notifyRequestor("The workflow %s was force completed"% wfo.name, do_batch=False)
forceComplete(url, wfi)
forced_already=True
else:
acdc_inactive.append( member['RequestName'] )
assistance_tags.add('recovered')
if acdc_bads:
sendEmail('inconsistent ACDC','for %s, ACDC %s is inconsistent, preventing from closing'%( wfo.name, ','.join(acdc_bads) ))
## completion check
percent_completions = {}
if not 'TotalInputEvents' in wfi.request:
event_expected,lumi_expected = 0,0
if not 'recovery' in wfo.status:
#sendEmail("missing member of the request","TotalInputEvents is missing from the workload of %s"% wfo.name, destination=['*****@*****.**'])
sendLog('checkor',"TotalInputEvents is missing from the workload of %s"% wfo.name, level='critical')
else:
event_expected,lumi_expected = wfi.request['TotalInputEvents'],wfi.request['TotalInputLumis']
if 'RequestNumEvents' in wfi.request and int(wfi.request['RequestNumEvents']):
event_expected = int(wfi.request['RequestNumEvents'])
elif 'Task1' in wfi.request and 'RequestNumEvents' in wfi.request['Task1']:
event_expected = wfi.request['Task1']['RequestNumEvents']
for i in range(1,20):
if 'Task%d'%i in wfi.request:
## this is wrong ibsolute
if 'FilterEfficiency' in wfi.request['Task%d'%i]:
event_expected *= float(wfi.request['Task%d'%i]['FilterEfficiency'])
event_expected = int(event_expected)
fractions_pass = {}
over_100_pass = False
(lhe,prim,_,_) = wfi.getIO()
if lhe or prim: over_100_pass = False
for output in wfi.request['OutputDatasets']:
event_count,lumi_count = getDatasetEventsAndLumis(dataset=output)
percent_completions[output] = 0.
if lumi_expected:
percent_completions[output] = lumi_count / float( lumi_expected )
if event_expected:
wfi.sendLog('checkor', "event completion real %s expected %s"%(event_count, event_expected ))
percent_completions[output] = max(percent_completions[output], float(event_count) / float( event_expected ) )
fractions_pass[output] = 0.95
c = campaigns[output]
if c in CI.campaigns and 'fractionpass' in CI.campaigns[c]:
fractions_pass[output] = CI.campaigns[c]['fractionpass']
wfi.sendLog('checkor', "overriding fraction to %s for %s by campaign requirement"%( fractions_pass[output], output))
if options.fractionpass:
fractions_pass[output] = options.fractionpass
print "overriding fraction to",fractions_pass[output],"by command line for",output
for key in pattern_fraction_pass:
if key in output:
fractions_pass[output] = pattern_fraction_pass[key]
print "overriding fraction to",fractions_pass[output],"by dataset key",key
if not all([percent_completions[out] >= fractions_pass[out] for out in fractions_pass]):
possible_recoveries = wfi.getRecoveryDoc()
if possible_recoveries == []:
wfi.sendLog('checkor','%s has missing statistics \n%s \n%s, but nothing is recoverable. passing through to annoucement'%(
wfo.name, json.dumps(percent_completions, indent=2), json.dumps(fractions_pass, indent=2) ))
sendEmail('nothing is recoverable','%s is not completed, but has nothing to be recovered, passing along ?'%wfo.name)
bypass_checks = True
else:
wfi.sendLog('checkor','%s is not completed \n%s \n%s'%(
wfo.name, json.dumps(percent_completions, indent=2), json.dumps(fractions_pass, indent=2) ))
## hook for creating automatically ACDC ?
if not bypass_checks:
assistance_tags.add('recovery')
is_closing = False
if over_100_pass and any([percent_completions[out] >100 for out in fractions_pass]):
print wfo.name,"is over completed"
print json.dumps(percent_completions, indent=2)
if not bypass_checks:
assistance_tags.add('over100')
is_closing = False
## correct lumi < 300 event per lumi
events_per_lumi = {}
for output in wfi.request['OutputDatasets']:
events_per_lumi[output] = getDatasetEventsPerLumi( output )
lumi_upper_limit = {}
for output in wfi.request['OutputDatasets']:
upper_limit = 301.
campaign = campaigns[output]
#if 'EventsPerLumi' in wfi.request and 'FilterEfficiency' in wfi.request:
# upper_limit = 1.5*wfi.request['EventsPerLumi']*wfi.request['FilterEfficiency']
# print "setting the upper limit of lumisize to",upper_limit,"by request configuration"
if campaign in CI.campaigns and 'lumisize' in CI.campaigns[campaign]:
upper_limit = CI.campaigns[campaign]['lumisize']
print "overriding the upper lumi size to",upper_limit,"for",campaign
if options.lumisize:
upper_limit = options.lumisize
print "overriding the upper lumi size to",upper_limit,"by command line"
lumi_upper_limit[output] = upper_limit
if wfi.request['RequestType'] in ['ReDigi']: lumi_upper_limit[output] = -1
if any([ (lumi_upper_limit[out]>0 and events_per_lumi[out] >= lumi_upper_limit[out]) for out in events_per_lumi]):
print wfo.name,"has big lumisections"
print json.dumps(events_per_lumi, indent=2)
## hook for rejecting the request ?
if not bypass_checks:
assistance_tags.add('biglumi')
is_closing = False
any_presence = {}
for output in wfi.request['OutputDatasets']:
any_presence[output] = getDatasetPresence(url, output, vetoes=[])
## custodial copy
custodial_locations = {}
custodial_presences = {}
for output in wfi.request['OutputDatasets']:
custodial_presences[output] = [s for s in any_presence[output] if 'MSS' in s]
custodial_locations[output] = phedexClient.getCustodialSubscriptionRequestSite(output)
if not custodial_locations[output]:
custodial_locations[output] = []
## presence in phedex
phedex_presence ={}
for output in wfi.request['OutputDatasets']:
phedex_presence[output] = phedexClient.getFileCountDataset(url, output )
out_worth_checking = [out for out in custodial_locations.keys() if out.split('/')[-1] not in vetoed_custodial_tier]
size_worth_checking = sum([getDatasetSize(out)/1023. for out in out_worth_checking ]) ## size in TBs of all outputs
if not all(map( lambda sites : len(sites)!=0, [custodial_locations[out] for out in out_worth_checking])):
print wfo.name,"has not all custodial location"
print json.dumps(custodial_locations, indent=2)
##########
## hook for making a custodial replica ?
custodial = None
## get from other outputs
for output in out_worth_checking:
if len(custodial_locations[output]):
custodial = custodial_locations[output][0]
if custodial and float(SI.storage[custodial]) < size_worth_checking:
print "cannot use the other output custodial:",custodial,"because of limited space"
custodial = None
## try to get it from campaign configuration
if not custodial:
for output in out_worth_checking:
campaign = campaigns[output]
if campaign in CI.campaigns and 'custodial' in CI.campaigns[campaign]:
custodial = CI.campaigns[campaign]['custodial']
print "Setting custodial to",custodial,"from campaign configuration"
if custodial and float(SI.storage[custodial]) < size_worth_checking:
print "cannot use the campaign configuration custodial:",custodial,"because of limited space"
custodial = None
## get from the parent
pick_custodial = True
use_parent_custodial = UC.get('use_parent_custodial')
_,prim,_,_ = wfi.getIO()
if not custodial and prim and use_parent_custodial:
parent_dataset = prim.pop()
## this is terribly dangerous to assume only
parents_custodial = phedexClient.getCustodialSubscriptionRequestSite( parent_dataset )
###parents_custodial = findCustodialLocation(url, parent_dataset)
if not parents_custodial:
parents_custodial = []
if len(parents_custodial):
custodial = parents_custodial[0]
else:
print "the input dataset",parent_dataset,"does not have custodial in the first place. abort"
#sendEmail( "dataset has no custodial location", "Please take a look at %s in the logs of checkor"%parent_dataset)
## does not work for RAWOADSIM
sendLog('checkor',"Please take a look at %s for missing custodial location"% parent_dataset)
## cannot be bypassed, this is an issue to fix
is_closing = False
pick_custodial = False
assistance_tags.add('parentcustodial')
if custodial and float(SI.storage[custodial]) < size_worth_checking:
print "cannot use the parent custodial:",custodial,"because of limited space"
custodial = None
if not custodial and pick_custodial:
## pick one at random
custodial = SI.pick_SE(size=size_worth_checking)
if not custodial:
print "cannot find a custodial for",wfo.name
wfi.sendLog('checkor',"cannot find a custodial for %s probably because of the total output size %d"%( wfo.name, size_worth_checking))
#sendEmail( "cannot find a custodial","cannot find a custodial for %s probably because of the total output size %d"%( wfo.name, size_worth_checking))
sendLog('checkor',"cannot find a custodial for %s probably because of the total output size %d"%( wfo.name, size_worth_checking), level='critical')
if custodial and (is_closing or bypass_checks):
print "picked",custodial,"for tape copy"
## remember how much you added this round already ; this stays locally
SI.storage[custodial] -= size_worth_checking
## register the custodial request, if there are no other big issues
for output in out_worth_checking:
if not len(custodial_locations[output]):
if phedex_presence[output]>=1:
custodials[custodial].append( output )
## let's wait and see if that's needed
assistance_tags.add('custodial')
else:
print "no file in phedex for",output," not good to add to custodial requests"
#cannot be bypassed
is_closing = False
## disk copy
disk_copies = {}
for output in wfi.request['OutputDatasets']:
disk_copies[output] = [s for s in any_presence[output] if (not 'MSS' in s) and (not 'Buffer' in s)]
if not all(map( lambda sites : len(sites)!=0, disk_copies.values())):
print wfo.name,"has not all output on disk"
print json.dumps(disk_copies, indent=2)
## presence in dbs
dbs_presence = {}
dbs_invalid = {}
for output in wfi.request['OutputDatasets']:
dbs_presence[output] = dbs3Client.getFileCountDataset( output )
dbs_invalid[output] = dbs3Client.getFileCountDataset( output, onlyInvalid=True)
fraction_invalid = 0.01
if not all([dbs_presence[out] == (dbs_invalid[out]+phedex_presence[out]) for out in wfi.request['OutputDatasets']]) and not options.ignorefiles:
print wfo.name,"has a dbs,phedex mismatch"
print json.dumps(dbs_presence, indent=2)
print json.dumps(dbs_invalid, indent=2)
print json.dumps(phedex_presence, indent=2)
if not 'recovering' in assistance_tags:
assistance_tags.add('filemismatch')
#print this for show and tell if no recovery on-going
for out in dbs_presence:
_,_,missing_phedex,missing_dbs = getDatasetFiles(url, out)
if missing_phedex:
wfi.sendLog('checkor',"These %d files are missing in phedex\n%s"%(len(missing_phedex),
"\n".join( missing_phedex )))
if missing_dbs:
wfi.sendLog('checkor',"These %d files are missing in dbs\n%s"%(len(missing_dbs),
"\n".join( missing_dbs )))
#if not bypass_checks:
## I don't think we can by pass this
is_closing = False
if not all([(dbs_invalid[out] <= int(fraction_invalid*dbs_presence[out])) for out in wfi.request['OutputDatasets']]) and not options.ignorefiles:
print wfo.name,"has a dbs invalid file level too high"
print json.dumps(dbs_presence, indent=2)
print json.dumps(dbs_invalid, indent=2)
print json.dumps(phedex_presence, indent=2)
## need to be going and taking an eye
assistance_tags.add('invalidfiles')
if not bypass_checks:
#sub_assistance+="-invalidfiles"
is_closing = False
## put that heavy part at the end
## duplication check
duplications = {}
if is_closing or bypass_checks:
print "starting duplicate checker for",wfo.name
for output in wfi.request['OutputDatasets']:
print "\tchecking",output
duplications[output] = True
try:
duplications[output] = dbs3Client.duplicateRunLumi( output , skipInvalid=True, verbose=True)
except:
try:
duplications[output] = dbs3Client.duplicateRunLumi( output , skipInvalid=True, verbose=True)
except:
print "was not possible to get the duplicate count for",output
is_closing=False
if any(duplications.values()) and not options.ignoreduplicates:
print wfo.name,"has duplicates"
print json.dumps(duplications,indent=2)
## hook for making file invalidation ?
## it shouldn't be allowed to bypass it
assistance_tags.add('duplicates')
is_closing = False
## for visualization later on
if not wfo.name in fDB.record:
#print "adding",wfo.name,"to close out record"
fDB.record[wfo.name] = {
'datasets' :{},
'name' : wfo.name,
'closeOutWorkflow' : None,
}
fDB.record[wfo.name]['closeOutWorkflow'] = is_closing
fDB.record[wfo.name]['priority'] = wfi.request['RequestPriority']
fDB.record[wfo.name]['prepid'] = wfi.request['PrepID']
for output in wfi.request['OutputDatasets']:
if not output in fDB.record[wfo.name]['datasets']: fDB.record[wfo.name]['datasets'][output] = {}
rec = fDB.record[wfo.name]['datasets'][output]
rec['percentage'] = float('%.2f'%(percent_completions[output]*100))
rec['duplicate'] = duplications[output] if output in duplications else 'N/A'
rec['phedexReqs'] = float('%.2f'%any_presence[output][custodial_presences[output][0]][1]) if len(custodial_presences[output])!=0 else 'N/A'
rec['closeOutDataset'] = is_closing
rec['transPerc'] = float('%.2f'%any_presence[output][ disk_copies[output][0]][1]) if len(disk_copies[output])!=0 else 'N/A'
rec['correctLumis'] = int(events_per_lumi[output]) if (events_per_lumi[output] > lumi_upper_limit[output]) else True
rec['missingSubs'] = False if len(custodial_locations[output])==0 else ','.join(list(set(custodial_locations[output])))
rec['dbsFiles'] = dbs_presence[output]
rec['dbsInvFiles'] = dbs_invalid[output]
rec['phedexFiles'] = phedex_presence[output]
rec['acdc'] = "%d / %d"%(len(acdc),len(acdc+acdc_inactive))
now = time.gmtime()
rec['timestamp'] = time.mktime(now)
rec['updated'] = time.asctime(now)+' (GMT)'
## and move on
if is_closing:
## toggle status to closed-out in request manager
print "setting",wfo.name,"closed-out"
if not options.test:
if wfo.wm_status in ['closed-out','announced','normal-archived']:
print wfo.name,"is already",wfo.wm_status,"not trying to closed-out and assuming it does"
res = None
else:
res = reqMgrClient.closeOutWorkflowCascade(url, wfo.name)
print "close out answer",res
if not res in ["None",None]:
print "try to get the current status again"
wfi_bis = workflowInfo(url, wfo.name)
if wfi_bis.request['RequestStatus'] == 'closed-out':
print "the request did toggle to closed-out"
res = None
if not res in ["None",None]:
print "retrying to closing out"
print res
res = reqMgrClient.closeOutWorkflowCascade(url, wfo.name)
if res in [None,"None"]:
wfo.status = 'close'
session.commit()
if use_mcm and force_by_mcm:
## shoot large on all prepids, on closing the wf
for pid in pids:
mcm.delete('/restapi/requests/forcecomplete/%s'%pid)
else:
print "could not close out",wfo.name,"will try again next time"
else:
## full known list
#recovering # has active ACDC
##OUT #recovered #had inactive ACDC
#recovery #not over the pass bar
#over100 # over 100%
#biglumi # has a big lumiblock
#parentcustodial # the parent does not have a valid subscription yet
#custodial # has had the transfer made, is waiting for a valid custodial subscription to appear
#filemismatch # there is a dbs/phedex mismatch
#duplicates #a lumi section is there twice
## manual is not added yet, and should be so by recoveror
print wfo.name,"was tagged with :",list(assistance_tags)
if 'recovering' in assistance_tags:
## if active ACDC, being under threshold, filemismatch do not matter
assistance_tags = assistance_tags - set(['recovery','filemismatch'])
if 'recovery' in assistance_tags and 'recovered' in assistance_tags:
## should not set -recovery to anything that add ACDC already
assistance_tags = assistance_tags - set(['recovery','recovered'])
## straight to manual
assistance_tags.add('manual')
## that means there is something that needs to be done acdc, lumi invalidation, custodial, name it
print wfo.name,"needs assistance with",",".join( assistance_tags )
print wfo.name,"existing conditions",",".join( existing_assistance_tags )
#########################################
##### notification to requester #########
go_notify=False
if assistance_tags and not 'manual' in existing_assistance_tags and existing_assistance_tags != assistance_tags:
go_notify=True
if go_notify:
#if wfo.name in already_notified:
# print "double notification"
# sendEmail('double notification','please take a look at %s'%(wfo.name))
#else:
# already_notified.append( wfo.name )
detailslink = 'https://cmsweb.cern.ch/reqmgr/view/details/%s'
perflink = 'https://cmsweb.cern.ch/couchdb/workloadsummary/_design/WorkloadSummary/_show/histogramByWorkflow/%s'%(wfo.name)
splitlink = 'https://cmsweb.cern.ch/reqmgr/view/splitting/%s'%(wfo.name)
## notify templates
messages= {
'recovery': 'Samples completed with missing statistics:\n%s\n%s '%( '\n'.join(['%.2f %% complete for %s'%(percent_completions[output]*100, output) for output in wfi.request['OutputDatasets'] ] ), perflink ),
'biglumi': 'Samples completed with large luminosity blocks:\n%s\n%s '%('\n'.join(['%d > %d for %s'%(events_per_lumi[output], lumi_upper_limit[output], output) for output in wfi.request['OutputDatasets'] if (events_per_lumi[output] > lumi_upper_limit[output])]), splitlink),
'duplicates': 'Samples completed with duplicated luminosity blocks:\n%s\n'%( '\n'.join(['%s'%output for output in wfi.request['OutputDatasets'] if output in duplications and duplications[output] ] ) ),
'filemismatch': 'Samples completed with inconsistency in DBS/Phedex',
#'manual' : 'Workflow completed and requires manual checks by Ops',
}
content = "The request PREPID (WORKFLOW) is facing issue in production.\n"
motive = False
for case in messages:
if case in assistance_tags:
content+= "\n"+messages[case]+"\n"
motive = True
content += "You are invited to check, while this is being taken care of by Comp-Ops.\n"
content += "This is an automated message from Comp-Ops.\n"
items_notified = set()
if use_mcm and motive:
wfi.notifyRequestor( content , mcm = mcm)
#########################################
## logic to set the status further
if assistance_tags:
new_status = 'assistance-'+'-'.join(sorted(assistance_tags) )
else:
new_status = 'assistance'
## case where the workflow was in manual from recoveror
if not 'manual' in wfo.status or new_status!='assistance-recovery':
wfo.status = new_status
if not options.test:
print "setting",wfo.name,"to",wfo.status
session.commit()
else:
print "current status is",wfo.status,"not changing to anything"
#open('already_notifified.json','w').write( json.dumps( already_notified , indent=2))
fDB.html()
if not spec:
#sendEmail("fresh assistance status available","Fresh status are available at https://cmst2.web.cern.ch/cmst2/unified/assistance.html",destination=['*****@*****.**'])
#it's a bit annoying
pass
## custodial requests
print "Custodials"
print json.dumps(custodials, indent=2)
for site in custodials:
print ','.join(custodials[site]),'=>',site
if not options.test:
result = makeReplicaRequest(url, site, list(set(custodials[site])),"custodial copy at production close-out",custodial='y',priority='low', approve = (site in SI.sites_auto_approve) )
print result
print "Transfers"
print json.dumps(transfers, indent=2)
## replicas requests
for site in transfers:
print ','.join(transfers[site]),'=>',site
if not options.test:
result = None
#result = makeReplicaRequest(url, site, list(set(transfers[site])),"copy to disk at production close-out")
print result
print "File Invalidation"
print invalidations
def close(self):
if os.path.isfile('.closor_stop'):
print "The closing of workflows is shortened"
return
url = self.url
batch_go = self.batch_go
CI = self.CI
UC = self.UC
wfo = self.wfo
jump_the_line = self.jump_the_line
batch_goodness = self.batch_goodness
check_fullcopy_to_announce = UC.get('check_fullcopy_to_announce')
## what is the expected #lumis
self.wfi = workflowInfo(url, wfo.name)
wfi = self.wfi
wfo.wm_status = wfi.request['RequestStatus']
if wfi.isRelval():
has_batch_go = False
batch_name = wfi.getCampaign()
if not batch_name in batch_go:
## do the esimatation whethere this can be announced : only once per batch
in_batches = getWorkflowByCampaign(url,
batch_name,
details=True)
batch_go[batch_name] = all(
map(
lambda s: not s in [
'completed', 'running-open', 'running-closed',
'acquired', 'staged', 'staging', 'assigned',
'assignment-approved'
], [r['RequestStatus'] for r in in_batches]))
## already verified
has_batch_go = batch_go[batch_name]
if not has_batch_go:
wfi.sendLog(
'closor',
'Cannot close for now because the batch <a href=https://dmytro.web.cern.ch/dmytro/cmsprodmon/workflows.php?campaign=%s>%s</a> is not all close'
% (batch_name, batch_name))
return
if wfi.request['RequestStatus'] in ['announced', 'normal-archived'
] and not options.force:
## manually announced ??
self.to_status = 'done'
self.to_wm_status = wfi.request['RequestStatus']
wfi.sendLog(
'closor',
'%s is announced already : %s' % (wfo.name, self.to_wm_status))
return
if jump_the_line:
wfi.sendLog('closor', 'Announcing while completing')
expected_lumis = 1
if not 'TotalInputLumis' in wfi.request:
print wfo.name, "has not been assigned yet, or the database is corrupted"
elif wfi.request['TotalInputLumis'] == 0:
print wfo.name, "is corrupted with 0 expected lumis"
else:
expected_lumis = wfi.request['TotalInputLumis']
## what are the outputs
outputs = wfi.request['OutputDatasets']
## check whether the number of lumis is as expected for each
all_OK = defaultdict(lambda: False)
stats = defaultdict(int)
#print outputs
if len(outputs):
print wfo.name, wfi.request['RequestStatus']
for out in outputs:
event_count, lumi_count = getDatasetEventsAndLumis(dataset=out)
self.outs.append(Output(datasetname=out))
odb = self.outs[-1]
odb.workflow = wfo
odb.nlumis = lumi_count
odb.nevents = event_count
odb.workfow_id = wfo.id
if odb.expectedlumis < expected_lumis:
odb.expectedlumis = expected_lumis
else:
expected_lumis = odb.expectedlumis
odb.date = time.mktime(time.gmtime())
fraction = lumi_count / float(expected_lumis) * 100.
completion_line = "%60s %d/%d = %3.2f%%" % (
out, lumi_count, expected_lumis, fraction)
wfi.sendLog('closor', "\t%s" % completion_line)
if wfi.isRelval() and fraction < batch_goodness:
self.batch_warnings[wfi.getCampaign()].add(completion_line)
if fraction < 50:
self.batch_extreme_warnings[wfi.getCampaign()].add(
completion_line)
stats[out] = lumi_count
all_OK[out] = True
## check for at least one full copy prior to moving on
in_full = {}
for out in outputs:
in_full[out] = []
presence = getDatasetPresence(url, out)
where = [site for site, info in presence.items() if info[0]]
if where:
all_OK[out] = True
print out, "is in full at", ",".join(where)
in_full[out] = copy.deepcopy(where)
else:
going_to = wfi.request['NonCustodialSites'] + wfi.request[
'CustodialSites']
wfi.sendLog(
'closor', "%s is not in full anywhere. send to %s" %
(out, ",".join(sorted(going_to))))
at_destination = dict([(k, v) for (k, v) in presence.items()
if k in going_to])
else_where = dict([(k, v) for (k, v) in presence.items()
if not k in going_to])
print json.dumps(at_destination)
print json.dumps(else_where, indent=2)
## do the full stuck transfer study, missing files and shit !
for there in going_to:
late_info = findLateFiles(url, out, going_to=there)
for l in late_info:
l.update({"workflow": wfo.name, "dataset": out})
self.all_late_files.extend(late_info)
if check_fullcopy_to_announce:
## only set this false if the check is relevant
all_OK[out] = False
## verify if we have to do harvesting
if not options.no_harvest and not jump_the_line:
(OK, requests) = spawn_harvesting(url, wfi, in_full)
all_OK.update(OK)
## only that status can let me go into announced
if all(all_OK.values()) and (
(wfi.request['RequestStatus'] in ['closed-out']) or options.force
or jump_the_line):
print wfo.name, "to be announced"
results = []
if not results:
for out in outputs:
print "dealing with", out
if out in stats and not stats[out]:
continue
_, dsn, process_string, tier = out.split('/')
if all_OK[out]:
print "setting valid"
results.append(
setDatasetStatus(out, 'VALID', withFiles=False))
if all_OK[out] and wfi.isRelval():
## make the specific relval rules and the replicas
## figure the destination(s) out
destinations = set()
if tier != "RECO" and tier != "ALCARECO":
destinations.add('T2_CH_CERN')
if tier == "GEN-SIM":
destinations.add('T1_US_FNAL_Disk')
if tier == "GEN-SIM-DIGI-RAW":
destinations.add('T1_US_FNAL_Disk')
if tier == "GEN-SIM-RECO":
destinations.add('T1_US_FNAL_Disk')
if "RelValTTBar" in dsn and "TkAlMinBias" in process_string and tier != "ALCARECO":
destinations.add('T2_CH_CERN')
if "MinimumBias" in dsn and "SiStripCalMinBias" in process_string and tier != "ALCARECO":
destinations.add('T2_CH_CERN')
if destinations:
wfi.sendLog(
'closor', '%s to go to %s' %
(out, ', '.join(sorted(destinations))))
## call to makereplicarequest under relval => done
for site in destinations:
result = makeReplicaRequest(
url,
site, [out],
'Copy for release validation consumption',
priority='normal',
approve=True,
mail=False,
group='RelVal')
try:
request_id = result['phedex'][
'request_created'][0]['id']
results.append(True)
except:
results.append('Failed relval transfer')
elif all_OK[out]:
campaign = None
try:
campaign = out.split('/')[2].split('-')[0]
except:
if 'Campaign' in wfi.request and wfi.request[
'Campaign']:
campaign = wfi.request['Campaign']
to_DDM = False
## campaign override
if campaign and campaign in CI.campaigns and 'toDDM' in CI.campaigns[
campaign] and tier in CI.campaigns[campaign][
'toDDM']:
to_DDM = True
## by typical enabling
if tier in UC.get("tiers_to_DDM"):
to_DDM = True
## check for unitarity
if not tier in UC.get("tiers_no_DDM") + UC.get(
"tiers_to_DDM"):
print "tier", tier, "neither TO or NO DDM for", out
results.append('Not recognitized tier %s' % tier)
#sendEmail("failed DDM injection","could not recognize %s for injecting in DDM"% out)
sendLog(
'closor',
"could not recognize %s for injecting in DDM" %
out,
level='critical')
continue
n_copies = 1
destinations = []
if to_DDM and campaign and campaign in CI.campaigns and 'DDMcopies' in CI.campaigns[
campaign]:
ddm_instructions = CI.campaigns[campaign][
'DDMcopies']
if type(ddm_instructions) == int:
n_copies = CI.campaigns[campaign]['DDMcopies']
elif type(ddm_instructions) == dict:
## a more fancy configuration
for ddmtier, indication in ddm_instructions.items(
):
if ddmtier == tier or ddmtier in [
'*', 'all'
]:
## this is for us
if 'N' in indication:
n_copies = indication['N']
if 'host' in indication:
destinations = indication['host']
destination_spec = ""
if destinations:
destination_spec = "--destination=" + ",".join(
destinations)
group_spec = "" ## not used yet
### should make this a campaign configuration
## inject to DDM when necessary
if to_DDM:
print "Sending", out, " to DDM"
status = pass_to_dynamo(
[out],
N=n_copies,
sites=destinations if destinations else None,
group=group_spec if group_spec else None)
results.append(status)
if status == True:
wfi.sendLog(
'closor',
'%s is send to dynamo in %s copies %s %s' %
(out, n_copies, sorted(destinations),
group_spec))
else:
sendLog('closor',
"could not add " + out +
" to dynamo pool. check closor logs.",
level='critical')
wfi.sendLog(
'closor', "could not add " + out +
" to dynamo pool. check closor logs.")
else:
print wfo.name, "no stats for announcing", out
results.append('No Stats')
if all(
map(lambda result: result in ['None', None, True],
results)):
if not jump_the_line:
## only announce if all previous are fine
res = reqMgrClient.announceWorkflowCascade(
url, wfo.name)
if not res in ['None', None]:
## check the status again, it might well have toggled
wl_bis = workflowInfo(url, wfo.name)
self.to_wm_status = wl_bis.request['RequestStatus']
if wl_bis.request['RequestStatus'] in [
'announced', 'normal-archived'
]:
res = None
else:
res = reqMgrClient.announceWorkflowCascade(
url, wfo.name)
results.append(res)
print results
if all(map(lambda result: result in ['None', None, True],
results)):
if jump_the_line:
if not 'announced' in wfo.status:
self.to_status = wfo.status.replace(
'announce', 'announced')
else:
self.to_status = 'done'
self.closing = True
wfi.sendLog('closor', "workflow outputs are announced")
else:
wfi.sendLog(
'closor', "Error with %s to be announced \n%s" %
(wfo.name, json.dumps(results)))
elif wfi.request['RequestStatus'] in [
'failed', 'aborted', 'aborted-archived', 'rejected',
'rejected-archived', 'aborted-completed'
]:
if wfi.isRelval():
self.to_status = 'forget'
self.to_wm_status = wfi.request['RequestStatus']
wfi.sendLog(
'closor',
"%s is %s, but will not be set in trouble to find a replacement."
% (wfo.name, self.to_wm_status))
else:
self.to_status = 'trouble'
self.to_wm_status = wfi.request['RequestStatus']
else:
print wfo.name, "not good for announcing:", wfi.request[
'RequestStatus']
wfi.sendLog('closor', "cannot be announced")
self.held.add(wfo.name)
def assignor(url, specific=None, talk=True, options=None):
if userLock(): return
if duplicateLock(): return
if not componentInfo().check(): return
UC = unifiedConfiguration()
CI = campaignInfo()
SI = siteInfo()
NLI = newLockInfo()
n_assigned = 0
n_stalled = 0
wfos = []
if specific or options.early:
wfos.extend(
session.query(Workflow).filter(
Workflow.status == 'considered').all())
wfos.extend(
session.query(Workflow).filter(Workflow.status == 'staging').all())
if specific:
wfos.extend(
session.query(Workflow).filter(
Workflow.status == 'considered-tried').all())
wfos.extend(
session.query(Workflow).filter(Workflow.status == 'staged').all())
dataset_endpoints = json.loads(
open('%s/dataset_endpoints.json' % monitor_dir).read())
max_per_round = UC.get('max_per_round').get('assignor', None)
max_cpuh_block = UC.get('max_cpuh_block')
random.shuffle(wfos)
for wfo in wfos:
if options.limit and (n_stalled + n_assigned) > options.limit:
break
if max_per_round and (n_stalled + n_assigned) > max_per_round:
break
if specific:
if not any(map(lambda sp: sp in wfo.name, specific.split(','))):
continue
#if not specific in wfo.name: continue
print "\n\n"
wfh = workflowInfo(url, wfo.name)
wfh.sendLog('assignor', "%s to be assigned" % wfo.name)
## the site whitelist takes into account siteInfo, campaignInfo, memory and cores
(lheinput, primary, parent, secondary,
sites_allowed) = wfh.getSiteWhiteList()
## check if by configuration we gave it a GO
no_go = False
if not wfh.go(log=True) and not options.go:
n_stalled += 1
no_go = True
allowed_secondary = set()
for campaign in wfh.getCampaigns():
if campaign in CI.campaigns and 'secondaries' in CI.campaigns[
campaign]:
allowed_secondary.update(CI.campaigns[campaign]['secondaries'])
if (secondary and allowed_secondary) and (
set(secondary) & allowed_secondary != set(secondary)):
wfh.sendLog(
'assignor', '%s is not an allowed secondary' %
(', '.join(set(secondary) - allowed_secondary)))
#sendEmail('secondary not allowed','%s is not an allowed secondary'%( ', '.join(set(secondary)-allowed_secondary)))
sendLog('assignor',
'%s is not an allowed secondary' %
(', '.join(set(secondary) - allowed_secondary)),
level='critical')
if not options.go:
n_stalled += 1
no_go = True
if no_go:
continue
## check on current status for by-passed assignment
if wfh.request['RequestStatus'] != 'assignment-approved':
if not options.test:
wfh.sendLog('assignor',
"setting %s away and skipping" % wfo.name)
## the module picking up from away will do what is necessary of it
wfo.wm_status = wfh.request['RequestStatus']
wfo.status = 'away'
session.commit()
continue
else:
print wfo.name, wfh.request['RequestStatus']
## retrieve from the schema, dbs and reqMgr what should be the next version
version = wfh.getNextVersion()
if not version:
if options and options.ProcessingVersion:
version = options.ProcessingVersion
else:
wfh.sendLog('assignor', "cannot decide on version number")
n_stalled += 1
wfo.status = 'trouble'
session.commit()
continue
original_sites_allowed = copy.deepcopy(sites_allowed)
wfh.sendLog('assignor', "Site white list %s" % sorted(sites_allowed))
override_sec_location = CI.get(wfh.request['Campaign'],
'SecondaryLocation', [])
blocks = []
if 'BlockWhitelist' in wfh.request:
blocks = wfh.request['BlockWhitelist']
if 'RunWhitelist' in wfh.request and wfh.request['RunWhitelist']:
## augment with run white list
for dataset in primary:
blocks = list(
set(blocks + getDatasetBlocks(
dataset, runs=wfh.request['RunWhitelist'])))
wfh.sendLog('assignor', "Allowed %s" % sorted(sites_allowed))
secondary_locations = None
primary_aaa = options.primary_aaa
if 'Campaign' in wfh.request and wfh.request[
'Campaign'] in CI.campaigns and 'primary_AAA' in CI.campaigns[
wfh.request['Campaign']]:
primary_aaa = primary_aaa or CI.campaigns[
wfh.request['Campaign']]['primary_AAA']
secondary_aaa = options.secondary_aaa
if 'Campaign' in wfh.request and wfh.request[
'Campaign'] in CI.campaigns and 'secondary_AAA' in CI.campaigns[
wfh.request['Campaign']]:
secondary_aaa = secondary_aaa or CI.campaigns[
wfh.request['Campaign']]['secondary_AAA']
for sec in list(secondary):
if override_sec_location:
print "We don't care where the secondary is"
print "Cannot pass for now"
sendEmail("tempting to pass sec location check",
"but we cannot yet IMO")
#pass
if secondary_aaa:
#just continue without checking
continue
presence = getDatasetPresence(url, sec)
print sec
print json.dumps(presence, indent=2)
one_secondary_locations = [
site for (site, (there, frac)) in presence.items()
if frac > 98.
]
#one_secondary_locations = [site for (site,(there,frac)) in presence.items() if there]
if secondary_locations == None:
secondary_locations = one_secondary_locations
else:
secondary_locations = list(
set(secondary_locations) & set(one_secondary_locations))
## reduce the site white list to site with secondary only
#sites_allowed = [site for site in sites_allowed if any([osite.startswith(site) for osite in one_secondary_locations])]
sites_allowed = [
site for site in sites_allowed
if SI.CE_to_SE(site) in one_secondary_locations
]
wfh.sendLog(
'assignor', "From secondary requirement, now Allowed%s" %
sorted(sites_allowed))
initial_sites_allowed = copy.deepcopy(
sites_allowed
) ## keep track of this, after secondary input location restriction : that's how you want to operate it
sites_all_data = copy.deepcopy(sites_allowed)
sites_with_data = copy.deepcopy(sites_allowed)
sites_with_any_data = copy.deepcopy(sites_allowed)
primary_locations = None
available_fractions = {}
set_lfn = '/store/mc' ## by default
endpoints = set()
for prim in list(primary):
if prim in dataset_endpoints:
print "endpoints from stagor", dataset_endpoints[prim]
endpoints.update(dataset_endpoints[prim])
set_lfn = getLFNbase(prim)
presence = getDatasetPresence(url, prim, only_blocks=blocks)
if talk:
print prim
print json.dumps(presence, indent=2)
available_fractions[prim] = getDatasetBlocksFraction(
url,
prim,
sites=[SI.CE_to_SE(site) for site in sites_allowed],
only_blocks=blocks)
#sites_all_data = [site for site in sites_with_data if any([osite.startswith(site) for osite in [psite for (psite,(there,frac)) in presence.items() if there]])]
#sites_with_data = [site for site in sites_with_data if any([osite.startswith(site) for osite in [psite for (psite,frac) in presence.items() if frac[1]>90.]])]
sites_all_data = [
site for site in sites_with_data if SI.CE_to_SE(site) in [
psite for (psite, (there, frac)) in presence.items()
if there
]
]
sites_with_data = [
site for site in sites_with_data if SI.CE_to_SE(site) in
[psite for (psite, frac) in presence.items() if frac[1] > 90.]
]
sites_with_any_data = [
site for site in sites_with_any_data
if SI.CE_to_SE(site) in presence.keys()
]
wfh.sendLog(
'assignor', "Holding the data but not allowed %s" % sorted(
list(
set([
se_site for se_site in presence.keys()
if not SI.SE_to_CE(se_site) in sites_allowed
]))))
if primary_locations == None:
primary_locations = presence.keys()
else:
primary_locations = list(
set(primary_locations) & set(presence.keys()))
sites_with_data = list(set(sites_with_data))
sites_with_any_data = list(set(sites_with_any_data))
opportunistic_sites = []
down_time = False
## opportunistic running where any piece of data is available
if secondary_locations or primary_locations:
## intersection of both any pieces of the primary and good IO
#opportunistic_sites = [SI.SE_to_CE(site) for site in list((set(secondary_locations) & set(primary_locations) & set(SI.sites_with_goodIO)) - set(sites_allowed))]
if secondary_locations and primary_locations:
opportunistic_sites = [
SI.SE_to_CE(site) for site in
list((set(secondary_locations) & set(primary_locations)) -
set([SI.CE_to_SE(site) for site in sites_allowed]))
]
elif primary_locations:
opportunistic_sites = [
SI.SE_to_CE(site) for site in list(
set(primary_locations) -
set([SI.CE_to_SE(site) for site in sites_allowed]))
]
else:
opportunistic_sites = []
wfh.sendLog(
'assignor', "We could be running in addition at %s" %
sorted(opportunistic_sites))
if any(
[osite in SI.sites_not_ready
for osite in opportunistic_sites]):
wfh.sendLog(
'assignor', "One of the usable site is in downtime %s" % ([
osite in SI.sites_not_ready
for osite in opportunistic_sites
]))
down_time = True
## should this be send back to considered ?
## should be 2 but for the time-being let's lower it to get things going
copies_wanted, cpuh = wfh.getNCopies()
wfh.sendLog('assignor', "we need %s CPUh" % cpuh)
if cpuh > max_cpuh_block and not options.go:
#sendEmail('large workflow','that wf %s has a large number of CPUh %s, not assigning, please check the logs'%(wfo.name, cpuh))#,destination=['*****@*****.**'])
sendLog(
'assignor',
'%s requires a large numbr of CPUh %s , not assigning, please check with requester'
% (wfo.name, cpuh),
level='critical')
wfh.sendLog(
'assignor',
"Requiring a large number of CPUh %s, not assigning" % cpuh)
continue
if 'Campaign' in wfh.request and wfh.request[
'Campaign'] in CI.campaigns and 'maxcopies' in CI.campaigns[
wfh.request['Campaign']]:
copies_needed_from_campaign = CI.campaigns[
wfh.request['Campaign']]['maxcopies']
copies_wanted = min(copies_needed_from_campaign, copies_wanted)
if not options.early:
less_copies_than_requested = UC.get("less_copies_than_requested")
copies_wanted = max(
1, copies_wanted -
less_copies_than_requested) # take one out for the efficiency
else:
## find out whether there is a site in the whitelist, that is lacking jobs and reduce to 1 copy needed to get things going
pass
wfh.sendLog('assignor',
"needed availability fraction %s" % copies_wanted)
## should also check on number of sources, if large enough, we should be able to overflow most, efficiently
if available_fractions and not all([
available >= copies_wanted
for available in available_fractions.values()
]):
not_even_once = not all([
available >= 1. for available in available_fractions.values()
])
wfh.sendLog(
'assignor',
"The input dataset is not available %s times, only %s" %
(copies_wanted, available_fractions.values()))
if down_time and not options.go and not options.early:
wfo.status = 'considered'
session.commit()
wfh.sendLog(
'assignor',
"sending back to considered because of site downtime, instead of waiting"
)
#sendEmail( "cannot be assigned due to downtime","%s is not sufficiently available, due to down time of a site in the whitelist. check the assignor logs. sending back to considered."% wfo.name)
sendLog(
'assignor',
'%s is not sufficiently available, due to down time of a site in the whitelist. sending back to considered.'
% (wfo.name),
level='delay')
continue
#pass
print json.dumps(available_fractions)
if (options.go and not_even_once) or not options.go:
known = []
try:
known = json.loads(open('cannot_assign.json').read())
except:
pass
if not wfo.name in known and not options.limit and not options.go and not options.early and not options.partial:
wfh.sendLog(
'assignor',
"cannot be assigned, %s is not sufficiently available.\n %s"
% (wfo.name, json.dumps(available_fractions)))
sendEmail(
"cannot be assigned",
"%s is not sufficiently available.\n %s" %
(wfo.name, json.dumps(available_fractions)))
known.append(wfo.name)
open('cannot_assign.json',
'w').write(json.dumps(known, indent=2))
n_stalled += 1
if options.early:
if wfo.status == 'considered':
wfh.sendLog('assignor', "setting considered-tried")
wfo.status = 'considered-tried'
session.commit()
else:
print "tried but status is", wfo.status
if options.partial:
print "Will move on with partial locations"
else:
continue
## default back to white list to original white list with any data
print "Allowed", sorted(sites_allowed)
if primary_aaa:
sites_allowed = initial_sites_allowed
options.TrustSitelists = True
wfh.sendLog(
'assignor', "Selected to read primary through xrootd %s" %
sorted(sites_allowed))
else:
sites_allowed = sites_with_any_data
wfh.sendLog('assignor',
"Selected for any data %s" % sorted(sites_allowed))
if secondary_aaa:
options.TrustPUSitelists = True
wfh.sendLog(
'assignor', "Reading secondary through xrootd from %s" %
sorted(sites_allowed))
### check on endpoints for on-going transfers
if endpoints and options.partial:
sites_allowed = list(
set(sites_allowed + [SI.SE_to_CE(s) for s in endpoints]))
print "with added endpoints", sorted(sites_allowed)
if not len(sites_allowed):
wfh.sendLog('assignor', "cannot be assign with no matched sites")
sendLog('assignor',
'%s has no whitelist' % wfo.name,
level='critical')
n_stalled += 1
continue
t1_only = [ce for ce in sites_allowed if ce.startswith('T1')]
if t1_only:
# try to pick from T1 only first
sites_out = [SI.pick_dSE([SI.CE_to_SE(ce) for ce in t1_only])]
else:
# then pick any otherwise
sites_out = [
SI.pick_dSE([SI.CE_to_SE(ce) for ce in sites_allowed])
]
wfh.sendLog('assignor', "Placing the output on %s" % sites_out)
parameters = {
'SiteWhitelist': sites_allowed,
'NonCustodialSites': sites_out,
'AutoApproveSubscriptionSites': list(set(sites_out)),
'AcquisitionEra': wfh.acquisitionEra(),
'ProcessingString': wfh.processingString(),
'MergedLFNBase': set_lfn,
'ProcessingVersion': version,
}
## plain assignment here
team = 'production'
if os.getenv('UNIFIED_TEAM'): team = os.getenv('UNIFIED_TEAM')
if options and options.team:
team = options.team
if False and 'T2_CH_CERN' in parameters['SiteWhitelist']:
## add some check on
### the amount pending to HLT
### the size of the request
### the priority of the request (maybe not if we decide to overflow during runs)
parameters['SiteWhitelist'] = ['T2_CH_CERN_HLT']
team = 'hlt'
## reduce the splitting by factor of 4, regardless of type of splitting
sendEmail("sending work to HLT",
"%s was assigned to HLT" % wfo.name)
##parse options entered in command line if any
if options:
for key in reqMgrClient.assignWorkflow.keys:
v = getattr(options, key)
if v != None:
if type(v) == str and ',' in v:
parameters[key] = filter(None, v.split(','))
else:
parameters[key] = v
if lheinput:
## throttle reading LHE article
wfh.sendLog('assignor',
'Setting the number of events per job to 500k max')
parameters['EventsPerJob'] = 500000
## pick up campaign specific assignment parameters
parameters.update(CI.parameters(wfh.request['Campaign']))
if not options.test:
parameters['execute'] = True
split_check = wfh.checkWorkflowSplitting()
if split_check != True:
parameters.update(split_check)
if 'EventBased' in split_check.values():
wfh.sendLog('assignor', "Falling back to event splitting.")
#sendEmail("Fallback to EventBased","the workflow %s is too heavy to be processed as it is. Fallback to EventBased splitting"%wfo.name)
sendLog(
'assignor',
'the workflow %s is too heavy to be processed as it is. Fallback to EventBased splitting'
% wfo.name,
level='critical')
elif 'EventsPerJob' in split_check.values():
wfh.sendLog('assignor',
"Modifying the number of job per event")
#sendEmail("Modifying the job per events","the workflow %s is too heavy in number of jobs explosion"%wfo.name)
sendLog(
'assignor',
"the workflow %s is too heavy in number of jobs explosion"
% wfo.name,
level='critical')
# Handle run-dependent MC
pstring = wfh.processingString()
if 'PU_RD' in pstring:
numEvents = wfh.getRequestNumEvents()
eventsPerLumi = [getDatasetEventsPerLumi(prim) for prim in primary]
eventsPerLumi = sum(eventsPerLumi) / float(len(eventsPerLumi))
reqJobs = 500
if 'PU_RD2' in pstring:
reqJobs = 2000
eventsPerJob = int(numEvents / (reqJobs * 1.4))
lumisPerJob = int(eventsPerJob / eventsPerLumi)
if lumisPerJob == 0:
#sendEmail("issue with event splitting for run-dependent MC","%s needs to be split by event with %s per job"%(wfo.name, eventsPerJob))
sendLog('assignor',
"%s needs to be split by event with %s per job" %
(wfo.name, eventsPerJob),
level='critical')
wfh.sendLog(
'assignor',
"%s needs to be split by event with %s per job" %
(wfo.name, eventsPerJob))
parameters['EventsPerJob'] = eventsPerJob
else:
spl = wfh.getSplittings()[0]
eventsPerJobEstimated = spl[
'events_per_job'] if 'events_per_job' in spl else None
eventsPerJobEstimated = spl[
'avg_events_per_job'] if 'avg_events_per_job' in spl else None
if eventsPerJobEstimated and eventsPerJobEstimated > eventsPerJob:
#sendEmail("setting lumi splitting for run-dependent MC","%s was assigned with %s lumis/job"%( wfo.name, lumisPerJob))
sendLog('assignor',
"%s was assigned with %s lumis/job" %
(wfo.name, lumisPerJob),
level='critical')
wfh.sendLog(
'assignor', "%s was assigned with %s lumis/job" %
(wfo.name, lumisPerJob))
parameters['LumisPerJob'] = lumisPerJob
else:
#sendEmail("leaving splitting untouched for PU_RD*","please check on "+wfo.name)
sendLog(
'assignor',
"leaving splitting untouched for %s, please check on %s"
% (pstring, wfo.name),
level='critical')
wfh.sendLog(
'assignor',
"leaving splitting untouched for PU_RD*, please check."
)
result = reqMgrClient.assignWorkflow(url, wfo.name, team, parameters)
# set status
if not options.test:
if result:
wfo.status = 'away'
session.commit()
n_assigned += 1
wfh.sendLog(
'assignor', "Properly assigned\n%s" %
(json.dumps(parameters, indent=2)))
try:
## refetch information and lock output
new_wfi = workflowInfo(url, wfo.name)
(_, prim, _, sec) = new_wfi.getIO()
for secure in list(prim) + list(
sec) + new_wfi.request['OutputDatasets']:
## lock all outputs flat
NLI.lock(secure)
#for site in [SI.CE_to_SE(site) for site in sites_allowed]:
# for output in new_wfi.request['OutputDatasets']:
# LI.lock( output, site, 'dataset in production')
# for primary in prim:
# LI.lock( primary, site, 'dataset used in input')
# for secondary in sec:
# LI.lock( secondary, site, 'required for mixing' )
except Exception as e:
print "fail in locking output"
print str(e)
sendEmail("failed locking of output", str(e))
else:
print "ERROR could not assign", wfo.name
else:
pass
print "Assignment summary:"
sendLog('assignor', "Assigned %d Stalled %s" % (n_assigned, n_stalled))
def assignor(url, specific=None, talk=True, options=None):
if userLock('assignor'): return
CI = campaignInfo()
SI = siteInfo()
wfos = []
if specific:
wfos = session.query(Workflow).filter(Workflow.name == specific).all()
if not wfos:
if specific:
wfos = session.query(Workflow).filter(
Workflow.status == 'considered').all()
wfos.extend(
session.query(Workflow).filter(
Workflow.status == 'staging').all())
wfos.extend(
session.query(Workflow).filter(Workflow.status == 'staged').all())
for wfo in wfos:
if specific:
if not any(map(lambda sp: sp in wfo.name, specific.split(','))):
continue
#if not specific in wfo.name: continue
print wfo.name, "to be assigned"
wfh = workflowInfo(url, wfo.name)
## check if by configuration we gave it a GO
if not CI.go(wfh.request['Campaign']) and not options.go:
print "No go for", wfh.request['Campaign']
continue
## check on current status for by-passed assignment
if wfh.request['RequestStatus'] != 'assignment-approved':
print wfo.name, wfh.request['RequestStatus'], "skipping"
if not options.test:
continue
## retrieve from the schema, dbs and reqMgr what should be the next version
version = wfh.getNextVersion()
if not version:
if options and options.ProcessingVersion:
version = options.ProcessingVersion
else:
print "cannot decide on version number"
continue
(lheinput, primary, parent, secondary) = wfh.getIO()
sites_allowed = getSiteWhiteList(
(lheinput, primary, parent, secondary))
print "Allowed", sites_allowed
sites_out = [SI.pick_dSE([SI.CE_to_SE(ce) for ce in sites_allowed])]
sites_custodial = []
if len(sites_custodial) == 0:
print "No custodial, it's fine, it's covered in close-out"
if len(sites_custodial) > 1:
print "more than one custodial for", wfo.name
sys.exit(36)
secondary_locations = None
for sec in list(secondary):
presence = getDatasetPresence(url, sec)
print sec
print json.dumps(presence, indent=2)
#one_secondary_locations = [site for (site,(there,frac)) in presence.items() if frac>90.]
one_secondary_locations = [
site for (site, (there, frac)) in presence.items() if there
]
if secondary_locations == None:
secondary_locations = one_secondary_locations
else:
secondary_locations = list(
set(secondary_locations) & set(one_secondary_locations))
## reduce the site white list to site with secondary only
sites_allowed = [
site for site in sites_allowed if any([
osite.startswith(site) for osite in one_secondary_locations
])
]
sites_all_data = copy.deepcopy(sites_allowed)
sites_with_data = copy.deepcopy(sites_allowed)
sites_with_any_data = copy.deepcopy(sites_allowed)
primary_locations = None
available_fractions = {}
for prim in list(primary):
presence = getDatasetPresence(url, prim)
if talk:
print prim
print json.dumps(presence, indent=2)
available_fractions[prim] = getDatasetBlocksFraction(
url, prim, sites=[SI.CE_to_SE(site) for site in sites_allowed])
sites_all_data = [
site for site in sites_with_data if any([
osite.startswith(site) for osite in [
psite for (psite, (there, frac)) in presence.items()
if there
]
])
]
sites_with_data = [
site for site in sites_with_data if any([
osite.startswith(site) for osite in [
psite for (psite, frac) in presence.items()
if frac[1] > 90.
]
])
]
sites_with_any_data = [
site for site in sites_with_any_data
if any([osite.startswith(site) for osite in presence.keys()])
]
if primary_locations == None:
primary_locations = presence.keys()
else:
primary_locations = list(
set(primary_locations) & set(presence.keys()))
sites_with_data = list(set(sites_with_data))
sites_with_any_data = list(set(sites_with_any_data))
opportunistic_sites = []
## opportunistic running where any piece of data is available
if secondary_locations and primary_locations:
## intersection of both any pieces of the primary and good IO
#opportunistic_sites = [SI.SE_to_CE(site) for site in list((set(secondary_locations) & set(primary_locations) & set(SI.sites_with_goodIO)) - set(sites_allowed))]
opportunistic_sites = [
SI.SE_to_CE(site) for site in list((set(secondary_locations)
& set(primary_locations)) -
set(sites_allowed))
]
print "We could be running at", opportunistic_sites, "in addition"
if available_fractions and not all(
[available >= 1. for available in available_fractions.values()]):
print "The input dataset is not located in full at any site"
print json.dumps(available_fractions)
if not options.test and not options.go: continue ## skip skip skip
copies_wanted = 2.
if available_fractions and not all([
available >= copies_wanted
for available in available_fractions.values()
]):
print "The input dataset is not available", copies_wanted, "times, only", available_fractions.values(
)
if not options.go:
continue
## default back to white list to original white list with any data
print "Allowed", sites_allowed
sites_allowed = sites_with_any_data
print "Selected for any data", sites_allowed
if options.restrict:
print "Allowed", sites_allowed
sites_allowed = sites_with_any_data
print "Selected", sites_allowed
else:
if set(sites_with_data) != set(sites_allowed):
## the data is not everywhere we wanted to run at : enable aaa
print "Sites with 90% data not matching site white list (block choping!)"
print "Resorting to AAA reading for", list(
set(sites_allowed) - set(sites_with_data)), "?"
print "Whitelist site with any data", list(
set(sites_allowed) - set(sites_with_any_data))
#options.useSiteListAsLocation = True
#print "Not commissioned yet"
#continue
#print "We could be running at",opportunistic_sites,"in addition"
##sites_allowed = list(set(sites_allowed+ opportunistic_sites))
if not len(sites_allowed):
print wfo.name, "cannot be assign with no matched sites"
continue
parameters = {
'SiteWhitelist': sites_allowed,
'CustodialSites': sites_custodial,
'NonCustodialSites': sites_out,
'AutoApproveSubscriptionSites': list(set(sites_out)),
'AcquisitionEra': wfh.acquisitionEra(),
'ProcessingString': wfh.processingString(),
'MergedLFNBase': '/store/mc', ## to be figured out ! from Hi shit
'ProcessingVersion': version,
}
##parse options entered in command line if any
if options:
for key in reqMgrClient.assignWorkflow.keys:
v = getattr(options, key)
if v != None:
if ',' in v: parameters[key] = filter(None, v.split(','))
else: parameters[key] = v
## pick up campaign specific assignment parameters
parameters.update(CI.parameters(wfh.request['Campaign']))
if not options.test:
parameters['execute'] = True
if not wfh.checkWorkflowSplitting():
## needs to go to event based ? fail for now
print "Falling back to event splitting ?"
#parameters['SplittingAlgorithm'] = 'EventBased'
continue
## plain assignment here
team = 'production'
if options and options.team:
team = options.team
result = reqMgrClient.assignWorkflow(url, wfo.name, team, parameters)
# set status
if not options.test:
if result:
wfo.status = 'away'
session.commit()
else:
print "ERROR could not assign", wfo.name
else:
pass
def assignor(url ,specific = None, talk=True, options=None):
CI = campaignInfo()
SI = siteInfo()
wfos=[]
if specific:
wfos = session.query(Workflow).filter(Workflow.name==specific).all()
if not wfos:
if specific:
wfos = session.query(Workflow).filter(Workflow.status=='considered').all()
wfos.extend( session.query(Workflow).filter(Workflow.status=='staging').all())
wfos.extend(session.query(Workflow).filter(Workflow.status=='staged').all())
for wfo in wfos:
if specific:
if not any(map(lambda sp: sp in wfo.name,specific.split(','))): continue
#if not specific in wfo.name: continue
print wfo.name,"to be assigned"
wfh = workflowInfo( url, wfo.name)
#wl = getWorkLoad(url, wfo.name )
if not CI.go( wfh.request['Campaign'] ):
print "No go for",wfh.request['Campaign']
continue
injection_time = time.mktime(time.strptime('.'.join(map(str,wfh.request['RequestDate'])),"%Y.%m.%d.%H.%M.%S")) / (60.*60.)
now = time.mktime(time.gmtime()) / (60.*60.)
if float(now - injection_time) < 4.:
print "It is too soon to inject: %3.2fH remaining"%(now - injection_time)
if not options.test:
continue
#grace_period = 4 #days
#if float(now - injection_time) > grace_period*24.:
# print "it has been",grace_period,"need to do something"
# options.restrict = True
#else:
# print now,injection_time,now - injection_time
#print wl
if wfh.request['RequestStatus'] !='assignment-approved':
print wfo.name,wfh.request['RequestStatus'],"skipping"
if not options.test:
continue
version=wfh.getNextVersion()
(lheinput,primary,parent,secondary) = wfh.getIO()
sites_allowed = getSiteWhiteList( (lheinput,primary,parent,secondary) )
sites_custodial = list(set(itertools.chain.from_iterable([findCustodialLocation(url, prim) for prim in primary])))
sites_out = [SI.pick_dSE([SI.CE_to_SE(ce) for ce in sites_allowed])]
if len(sites_custodial)==0:
sites_custodial = [SI.pick_SE()]
print "picked",sites_custodial," as custodial for",wfo.name
if len(sites_custodial)>1:
print "more than one custodial for",wfo.name
sys.exit(36)
sites_with_data = copy.deepcopy( sites_allowed )
for prim in list(primary)+list(secondary):
presence = getDatasetPresence( url, prim )
if talk:
print prim,presence
sites_with_data = [site for site in sites_with_data if any([osite.startswith(site) for osite in [psite for (psite,frac) in presence.items() if frac[1]>90.]])]
sites_with_data = list(set(sites_with_data))
if options.restrict:
if talk:
print sites_allowed
sites_allowed = sites_with_data
else:
if set(sites_with_data) != set(sites_allowed):
## the data is not everywhere we wanted to run at : enable aaa
print "Resorting to AAA reading for",list(set(sites_allowed) - set(sites_with_data))
#options.useSiteListAsLocation = True
print "Not commissioned yet"
continue
if not len(sites_allowed):
print wfo.name,"cannot be assign with no matched sites"
continue
parameters={
'SiteWhitelist' : sites_allowed,
'CustodialSites' : sites_custodial,
'NonCustodialSites' : sites_out,
'AutoApproveSubscriptionSites' : list(set(sites_out+sites_custodial)),
'AcquisitionEra' : wfh.acquisitionEra(),
'ProcessingString' : wfh.processingString(),
'MergedLFNBase' : '/store/mc', ## to be figured out ! from Hi shit
'ProcessingVersion' : version,
}
##parse options entered in command line if any
if options:
for key in reqMgrClient.assignWorkflow.keys:
v=getattr(options,key)
if v!=None:
if ',' in v: parameters[key] = filter(None,v.split(','))
else: parameters[key] = v
## take care of a few exceptions
if (wfh.request['Memory']*1000) > 3000000:
parameters['MaxRSS'] = 4000000
## pick up campaign specific assignment parameters
parameters.update( CI.parameters(wfh.request['Campaign']) )
if not options.test:
parameters['execute'] = True
if not wfh.checkWorkflowSplitting():
## needs to go to event based ? fail for now
print "Falling back to event splitting ?"
parameters['SplittingAlgorithm'] = 'EventBased'
## plain assignment here
result = reqMgrClient.assignWorkflow(url, wfo.name, 'production', parameters)
# set status
if not options.test:
if result:
wfo.status = 'away'
session.commit()
else:
print "ERROR could not assign",wfo.name
else:
pass
def assignor(url ,specific = None, talk=True, options=None):
if userLock(): return
if duplicateLock(): return
if not componentInfo().check(): return
UC = unifiedConfiguration()
CI = campaignInfo()
#SI = siteInfo()
SI = global_SI()
#NLI = newLockInfo()
#if not NLI.free() and not options.go: return
LI = lockInfo()
if not LI.free() and not options.go: return
n_assigned = 0
n_stalled = 0
wfos=[]
fetch_from = []
if specific or options.early:
fetch_from.extend(['considered','staging'])
if specific:
fetch_from.extend(['considered-tried'])
fetch_from.extend(['staged'])
if options.from_status:
fetch_from = options.from_status.split(',')
print "Overriding to read from",fetch_from
for status in fetch_from:
wfos.extend(session.query(Workflow).filter(Workflow.status==status).all())
## in case of partial, go for fetching a list from json ?
#if options.partial and not specific:
# pass
dataset_endpoints = json.loads(open('%s/dataset_endpoints.json'%monitor_dir).read())
aaa_mapping = json.loads(open('%s/equalizor.json'%monitor_pub_dir).read())['mapping']
all_stuck = set()
all_stuck.update( json.loads( open('%s/stuck_transfers.json'%monitor_dir).read() ))
all_stuck.update( getAllStuckDataset())
max_per_round = UC.get('max_per_round').get('assignor',None)
max_cpuh_block = UC.get('max_cpuh_block')
random.shuffle( wfos )
for wfo in wfos:
if options.limit and (n_stalled+n_assigned)>options.limit:
break
if max_per_round and (n_stalled+n_assigned)>max_per_round:
break
if specific:
if not any(map(lambda sp: sp in wfo.name,specific.split(','))): continue
#if not specific in wfo.name: continue
print "\n\n"
wfh = workflowInfo( url, wfo.name)
if options.priority and int(wfh.request['RequestPriority']) < options.priority:
continue
options_text=""
if options.early: options_text+=", early option is ON"
if options.partial:
options_text+=", partial option is ON"
options_text+=", good fraction is %.2f"%options.good_enough
wfh.sendLog('assignor',"%s to be assigned%s"%(wfo.name, options_text))
## the site whitelist takes into account siteInfo, campaignInfo, memory and cores
(lheinput,primary,parent,secondary, sites_allowed) = wfh.getSiteWhiteList()
output_tiers = list(set([o.split('/')[-1] for o in wfh.request['OutputDatasets']]))
is_stuck = (all_stuck & primary)
if is_stuck:
wfh.sendLog('assignor',"%s are stuck input"%(','.join( is_stuck)))
## check if by configuration we gave it a GO
no_go = False
if not wfh.go(log=True) and not options.go:
no_go = True
allowed_secondary = {}
assign_parameters = {}
check_secondary = False
for campaign in wfh.getCampaigns():
if campaign in CI.campaigns:
assign_parameters.update( CI.campaigns[campaign] )
if campaign in CI.campaigns and 'secondaries' in CI.campaigns[campaign]:
if CI.campaigns[campaign]['secondaries']:
allowed_secondary.update( CI.campaigns[campaign]['secondaries'] )
check_secondary = True
if campaign in CI.campaigns and 'banned_tier' in CI.campaigns[campaign]:
banned_tier = list(set(CI.campaigns[campaign]['banned_tier']) & set(output_tiers))
if banned_tier:
no_go=True
wfh.sendLog('assignor','These data tiers %s are not allowed'%(','.join( banned_tier)))
sendLog('assignor','These data tiers %s are not allowed'%(','.join( banned_tier)), level='critical')
if secondary and check_secondary:
if (set(secondary)&set(allowed_secondary.keys())!=set(secondary)):
wfh.sendLog('assignor','%s is not an allowed secondary'%(', '.join(set(secondary)-set(allowed_secondary.keys()))))
sendLog('assignor','%s is not an allowed secondary'%(', '.join(set(secondary)-set(allowed_secondary.keys()))), level='critical')
if not options.go:
no_go = True
## then get whether there is something more to be done by secondary
for sec in secondary:
if sec in allowed_secondary:# and 'parameters' in allowed_secondary[sec]:
assign_parameters.update( allowed_secondary[sec] )
if no_go:
n_stalled+=1
continue
## check on current status for by-passed assignment
if wfh.request['RequestStatus'] !='assignment-approved':
if not options.test:
wfh.sendLog('assignor',"setting %s away and skipping"%wfo.name)
## the module picking up from away will do what is necessary of it
wfo.wm_status = wfh.request['RequestStatus']
wfo.status = 'away'
session.commit()
continue
else:
print wfo.name,wfh.request['RequestStatus']
## retrieve from the schema, dbs and reqMgr what should be the next version
version=wfh.getNextVersion()
if not version:
if options and options.ProcessingVersion:
version = options.ProcessingVersion
else:
wfh.sendLog('assignor',"cannot decide on version number")
n_stalled+=1
wfo.status = 'trouble'
session.commit()
continue
original_sites_allowed = copy.deepcopy( sites_allowed )
wfh.sendLog('assignor',"Site white list %s"%sorted(sites_allowed))
override_sec_location = CI.get(wfh.request['Campaign'], 'SecondaryLocation', [])
blocks = []
if 'BlockWhitelist' in wfh.request:
blocks = wfh.request['BlockWhitelist']
if 'RunWhitelist' in wfh.request and wfh.request['RunWhitelist']:
## augment with run white list
for dataset in primary:
blocks = list(set( blocks + getDatasetBlocks( dataset, runs=wfh.request['RunWhitelist'] ) ))
wfh.sendLog('assignor',"Allowed %s"%sorted(sites_allowed))
secondary_locations=None
primary_aaa = options.primary_aaa
secondary_aaa = options.secondary_aaa
do_partial = False #options.good_enough if options.partial else 0
if 'Campaign' in wfh.request and wfh.request['Campaign'] in CI.campaigns:
assign_parameters.update( CI.campaigns[wfh.request['Campaign']] )
if 'primary_AAA' in assign_parameters:
primary_aaa = primary_aaa or assign_parameters['primary_AAA']
if 'secondary_AAA' in assign_parameters:
secondary_aaa = secondary_aaa or assign_parameters['secondary_AAA']
if 'partial_copy' in assign_parameters:
## can this only work if there is a stuck input ? maybe not
## this is a number. 0 means no
print "Could do partial disk copy assignment"
if is_stuck or options.partial:
do_partial = assign_parameters['partial_copy']
wfh.sendLog('assignor',"Overiding partial copy assignment to %.2f fraction"% do_partial)
#sendEmail('stuck input to assignment','%s is stuck for assigning %s and going fractional'%(','.join( is_stuck), wfo.name))
do_partial = options.good_enough if options.partial else do_partial
for sec in list(secondary):
if override_sec_location:
print "We don't care where the secondary is"
print "Cannot pass for now"
#sendEmail("tempting to pass sec location check","but we cannot yet IMO")
#pass
if secondary_aaa:
#just continue without checking
continue
presence = getDatasetPresence( url, sec )
print sec
print json.dumps(presence, indent=2)
one_secondary_locations = [site for (site,(there,frac)) in presence.items() if frac>98.]
#one_secondary_locations = [site for (site,(there,frac)) in presence.items() if there]
if secondary_locations==None:
secondary_locations = one_secondary_locations
else:
secondary_locations = list(set(secondary_locations) & set(one_secondary_locations))
## reduce the site white list to site with secondary only
#sites_allowed = [site for site in sites_allowed if any([osite.startswith(site) for osite in one_secondary_locations])]
sites_allowed = [site for site in sites_allowed if SI.CE_to_SE(site) in one_secondary_locations]
wfh.sendLog('assignor',"From secondary requirement, now Allowed%s"%sorted(sites_allowed))
initial_sites_allowed = copy.deepcopy( sites_allowed ) ## keep track of this, after secondary input location restriction : that's how you want to operate it
sites_all_data = copy.deepcopy( sites_allowed )
sites_with_data = copy.deepcopy( sites_allowed )
sites_with_any_data = copy.deepcopy( sites_allowed )
primary_locations = None
available_fractions = {}
set_lfn = '/store/mc' ## by default
endpoints = set()
for prim in list(primary):
if prim in dataset_endpoints:
print "endpoints from stagor",dataset_endpoints[prim]
endpoints.update( dataset_endpoints[prim] )
set_lfn = getLFNbase( prim )
presence = getDatasetPresence( url, prim , only_blocks=blocks)
if talk:
print prim
print json.dumps(presence, indent=2)
available_fractions[prim] = getDatasetBlocksFraction(url, prim, sites = [SI.CE_to_SE(site) for site in sites_allowed] , only_blocks = blocks)
#sites_all_data = [site for site in sites_with_data if any([osite.startswith(site) for osite in [psite for (psite,(there,frac)) in presence.items() if there]])]
#sites_with_data = [site for site in sites_with_data if any([osite.startswith(site) for osite in [psite for (psite,frac) in presence.items() if frac[1]>90.]])]
sites_all_data = [site for site in sites_with_data if SI.CE_to_SE(site) in [psite for (psite,(there,frac)) in presence.items() if there]]
sites_with_data = [site for site in sites_with_data if SI.CE_to_SE(site) in [psite for (psite,frac) in presence.items() if frac[1]>90.]]
sites_with_any_data = [site for site in sites_with_any_data if SI.CE_to_SE(site) in presence.keys()]
wfh.sendLog('assignor',"Holding the data but not allowed %s"%sorted(list(set([se_site for se_site in presence.keys() if not SI.SE_to_CE(se_site) in sites_allowed]))))
if primary_locations==None:
primary_locations = presence.keys()
else:
primary_locations = list(set(primary_locations) & set(presence.keys() ))
sites_with_data = list(set(sites_with_data))
sites_with_any_data = list(set(sites_with_any_data))
opportunistic_sites=[]
down_time = False
## opportunistic running where any piece of data is available
if secondary_locations or primary_locations:
## intersection of both any pieces of the primary and good IO
#opportunistic_sites = [SI.SE_to_CE(site) for site in list((set(secondary_locations) & set(primary_locations) & set(SI.sites_with_goodIO)) - set(sites_allowed))]
if secondary_locations and primary_locations:
opportunistic_sites = [SI.SE_to_CE(site) for site in list((set(secondary_locations) & set(primary_locations)) - set([SI.CE_to_SE(site) for site in sites_allowed]))]
elif primary_locations:
opportunistic_sites = [SI.SE_to_CE(site) for site in list(set(primary_locations) - set([SI.CE_to_SE(site) for site in sites_allowed]))]
else:
opportunistic_sites = []
wfh.sendLog('assignor',"We could be running in addition at %s"% sorted(opportunistic_sites))
if any([osite in SI.sites_not_ready for osite in opportunistic_sites]):
wfh.sendLog('assignor',"One of the usable site is in downtime %s"%([osite for osite in opportunistic_sites if osite in SI.sites_not_ready]))
down_time = True
## should this be send back to considered ?
## should be 2 but for the time-being let's lower it to get things going
copies_wanted,cpuh = wfh.getNCopies()
wfh.sendLog('assignor',"we need %s CPUh"%cpuh)
if cpuh>max_cpuh_block and not options.go:
#sendEmail('large workflow','that wf %s has a large number of CPUh %s, not assigning, please check the logs'%(wfo.name, cpuh))#,destination=['*****@*****.**'])
sendLog('assignor','%s requires a large numbr of CPUh %s , not assigning, please check with requester'%( wfo.name, cpuh), level='critical')
wfh.sendLog('assignor',"Requiring a large number of CPUh %s, not assigning"%cpuh)
continue
if 'Campaign' in wfh.request and wfh.request['Campaign'] in CI.campaigns and 'maxcopies' in CI.campaigns[wfh.request['Campaign']]:
copies_needed_from_campaign = CI.campaigns[wfh.request['Campaign']]['maxcopies']
copies_wanted = min(copies_needed_from_campaign, copies_wanted)
if not options.early:
less_copies_than_requested = UC.get("less_copies_than_requested")
copies_wanted = max(1,copies_wanted-less_copies_than_requested) # take one out for the efficiency
else:
## find out whether there is a site in the whitelist, that is lacking jobs and reduce to 1 copy needed to get things going
pass
wfh.sendLog('assignor',"needed availability fraction %s"% copies_wanted)
## should also check on number of sources, if large enough, we should be able to overflow most, efficiently
## default back to white list to original white list with any data
wfh.sendLog('assignor',"Allowed sites :%s"% sorted(sites_allowed))
if primary_aaa:
## remove the sites not reachable localy if not in having the data
if not sites_all_data:
wfh.sendLog('assignor',"Overiding the primary on AAA setting to Off")
primary_aaa=False
else:
aaa_grid = set(sites_all_data)
for site in list(aaa_grid):
aaa_grid.update( aaa_mapping.get(site,[]) )
sites_allowed = list(set(initial_sites_allowed) & aaa_grid)
wfh.sendLog('assignor',"Selected to read primary through xrootd %s"%sorted(sites_allowed))
if not primary_aaa:
sites_allowed = sites_with_any_data
wfh.sendLog('assignor',"Selected for any data %s"%sorted(sites_allowed))
### check on endpoints for on-going transfers
if do_partial:
if endpoints:
end_sites = [SI.SE_to_CE(s) for s in endpoints]
sites_allowed = list(set(sites_allowed + end_sites))
if down_time and not any(osite in SI.sites_not_ready for osite in end_sites):
print "Flip the status of downtime, since our destinations are good"
down_time = False
print "with added endpoints",sorted(end_sites)
else:
print "Cannot do partial assignment without knowin the endpoints"
n_stalled+=1
continue
#if not len(sites_allowed):
# if not options.early:
# wfh.sendLog('assignor',"cannot be assign with no matched sites")
# sendLog('assignor','%s has no whitelist'% wfo.name, level='critical')
# n_stalled+=1
# continue
low_pressure = SI.sites_low_pressure(0.4)
## if any of the site allowed is low pressure : reduce to 1 copy so that it gets started
allowed_and_low = sorted(set(low_pressure) & set(sites_allowed))
if allowed_and_low:
wfh.sendLog('assignor',"The workflow can run at %s under low pressure currently"%( ','.join( allowed_and_low )))
copies_wanted = max(1., copies_wanted-1.)
if available_fractions and not all([available>=copies_wanted for available in available_fractions.values()]):
not_even_once = not all([available>=1. for available in available_fractions.values()])
above_good = all([available >= do_partial for available in available_fractions.values()])
wfh.sendLog('assignor',"The input dataset is not available %s times, only %s"%( copies_wanted, available_fractions.values()))
if down_time and not options.go and not options.early:
wfo.status = 'considered'
session.commit()
wfh.sendLog('assignor',"sending back to considered because of site downtime, instead of waiting")
#sendEmail( "cannot be assigned due to downtime","%s is not sufficiently available, due to down time of a site in the whitelist. check the assignor logs. sending back to considered."% wfo.name)
sendLog('assignor','%s is not sufficiently available, due to down time of a site in the whitelist. sending back to considered.'%( wfo.name ), level='delay')
n_stalled+=1
continue
#pass
print json.dumps(available_fractions)
if (options.go and not_even_once) or not options.go:
known = []
try:
known = json.loads(open('cannot_assign.json').read())
except:
pass
if not wfo.name in known and not options.limit and not options.go and not options.early and not (do_partial and above_good):
wfh.sendLog('assignor',"cannot be assigned, %s is not sufficiently available.\n %s"%(wfo.name,json.dumps(available_fractions)))
#sendEmail( "cannot be assigned","%s is not sufficiently available.\n %s"%(wfo.name,json.dumps(available_fractions)))
known.append( wfo.name )
open('cannot_assign.json','w').write(json.dumps( known, indent=2))
if options.early:
if wfo.status == 'considered':
wfh.sendLog('assignor',"setting considered-tried")
wfo.status = 'considered-tried'
session.commit()
else:
print "tried but status is",wfo.status
if do_partial and above_good:
print "Will move on with partial locations"
else:
n_stalled+=1
continue
if not len(sites_allowed):
if not options.early:
wfh.sendLog('assignor',"cannot be assign with no matched sites")
sendLog('assignor','%s has no whitelist'% wfo.name, level='critical')
n_stalled+=1
continue
t1_only = [ce for ce in sites_allowed if ce.startswith('T1')]
if t1_only:
# try to pick from T1 only first
sites_out = [SI.pick_dSE([SI.CE_to_SE(ce) for ce in t1_only])]
else:
# then pick any otherwise
sites_out = [SI.pick_dSE([SI.CE_to_SE(ce) for ce in sites_allowed])]
wfh.sendLog('assignor',"Placing the output on %s"%sites_out)
parameters={
'SiteWhitelist' : sites_allowed,
'NonCustodialSites' : sites_out,
'AutoApproveSubscriptionSites' : list(set(sites_out)),
'AcquisitionEra' : wfh.acquisitionEra(),
'ProcessingString' : wfh.processingString(),
'MergedLFNBase' : set_lfn,
'ProcessingVersion' : version,
}
if primary_aaa:
parameters['TrustSitelists'] = True
wfh.sendLog('assignor',"Reading primary through xrootd at %s"%sorted(sites_allowed))
if secondary_aaa:
parameters['TrustPUSitelists'] = True
wfh.sendLog('assignor',"Reading secondary through xrootd at %s"%sorted(sites_allowed))
if 'parameters' in assign_parameters:
parameters.update( assign_parameters['parameters'] )
## plain assignment here
team='production'
if os.getenv('UNIFIED_TEAM'): team = os.getenv('UNIFIED_TEAM')
if options and options.team:
team = options.team
if False and 'T2_CH_CERN' in parameters['SiteWhitelist']:
## add some check on
### the amount pending to HLT
### the size of the request
### the priority of the request (maybe not if we decide to overflow during runs)
parameters['SiteWhitelist'] = ['T2_CH_CERN_HLT']
team = 'hlt'
## reduce the splitting by factor of 4, regardless of type of splitting
sendEmail("sending work to HLT","%s was assigned to HLT"%wfo.name)
##parse options entered in command line if any
if options:
for key in reqMgrClient.assignWorkflow.keys:
v=getattr(options,key)
if v!=None:
if type(v)==str and ',' in v:
parameters[key] = filter(None,v.split(','))
else:
parameters[key] = v
if lheinput:
## throttle reading LHE article
wfh.sendLog('assignor', 'Setting the number of events per job to 500k max')
parameters['EventsPerJob'] = 500000
## pick up campaign specific assignment parameters
#parameters.update( CI.parameters(wfh.request['Campaign']) )
parameters.update( assign_parameters.get('parameters',{}) )
if not options.test:
parameters['execute'] = True
split_check = wfh.checkWorkflowSplitting()
if split_check!=True:
parameters.update( split_check )
if 'NoGo' in split_check.values():
wfh.sendLog('assignor', "Failing splitting check")
sendLog('assignor','the workflow %s is failing the splitting check. Verify in the logs'% wfo.name, level='critical')
n_stalled+=1
continue
if 'EventBased' in split_check.values():
wfh.sendLog('assignor', "Falling back to event splitting.")
#sendEmail("Fallback to EventBased","the workflow %s is too heavy to be processed as it is. Fallback to EventBased splitting"%wfo.name)
sendLog('assignor','the workflow %s is too heavy to be processed as it is. Fallback to EventBased splitting ?'%wfo.name, level='critical')
## we have a problem here, that EventBased should never be used as a backup
if not options.go:
n_stalled+=1
continue
continue ## skip all together
elif 'EventsPerJob' in split_check.values():
wfh.sendLog('assignor', "Modifying the number of events per job")
#sendEmail("Modifying the job per events","the workflow %s is too heavy in number of jobs explosion"%wfo.name)
sendLog('assignor',"the workflow %s is too heavy in number of jobs explosion"%wfo.name, level='critical')
elif 'EventsPerLumi' in split_check.values():
wfh.sendLog('assignor', "Modifying the number of events per lumi to be able to process this")
# Handle run-dependent MC
pstring = wfh.processingString()
if 'PU_RD' in pstring:
numEvents = wfh.getRequestNumEvents()
eventsPerLumi = [getDatasetEventsPerLumi(prim) for prim in primary]
eventsPerLumi = sum(eventsPerLumi)/float(len(eventsPerLumi))
reqJobs = 500
if 'PU_RD2' in pstring:
reqJobs = 2000
eventsPerJob = int(numEvents/(reqJobs*1.4))
lumisPerJob = int(eventsPerJob/eventsPerLumi)
if lumisPerJob==0:
#sendEmail("issue with event splitting for run-dependent MC","%s needs to be split by event with %s per job"%(wfo.name, eventsPerJob))
sendLog('assignor', "%s needs to be split by event with %s per job"%(wfo.name, eventsPerJob), level='critical')
wfh.sendLog('assignor', "%s needs to be split by event with %s per job"%(wfo.name, eventsPerJob))
parameters['EventsPerJob'] = eventsPerJob
else:
spl = wfh.getSplittings()[0]
eventsPerJobEstimated = spl['events_per_job'] if 'events_per_job' in spl else None
eventsPerJobEstimated = spl['avg_events_per_job'] if 'avg_events_per_job' in spl else None
if eventsPerJobEstimated and eventsPerJobEstimated > eventsPerJob:
#sendEmail("setting lumi splitting for run-dependent MC","%s was assigned with %s lumis/job"%( wfo.name, lumisPerJob))
sendLog('assignor',"%s was assigned with %s lumis/job"%( wfo.name, lumisPerJob), level='critical')
wfh.sendLog('assignor',"%s was assigned with %s lumis/job"%( wfo.name, lumisPerJob))
parameters['LumisPerJob'] = lumisPerJob
else:
#sendEmail("leaving splitting untouched for PU_RD*","please check on "+wfo.name)
sendLog('assignor',"leaving splitting untouched for %s, please check on %s"%( pstring, wfo.name), level='critical')
wfh.sendLog('assignor',"leaving splitting untouched for PU_RD*, please check.")
result = reqMgrClient.assignWorkflow(url, wfo.name, team, parameters)
# set status
if not options.test:
if result:
wfo.status = 'away'
session.commit()
n_assigned+=1
wfh.sendLog('assignor',"Properly assigned\n%s"%(json.dumps( parameters, indent=2)))
try:
## refetch information and lock output
new_wfi = workflowInfo( url, wfo.name)
(_,prim,_,sec) = new_wfi.getIO()
for secure in list(prim)+list(sec)+new_wfi.request['OutputDatasets']:
## lock all outputs flat
#NLI.lock( secure )
LI.lock( secure, reason = 'assigning')
#for site in [SI.CE_to_SE(site) for site in sites_allowed]:
# for output in new_wfi.request['OutputDatasets']:
# LI.lock( output, site, 'dataset in production')
# for primary in prim:
# LI.lock( primary, site, 'dataset used in input')
# for secondary in sec:
# LI.lock( secondary, site, 'required for mixing' )
except Exception as e:
print "fail in locking output"
print str(e)
sendEmail("failed locking of output",str(e))
else:
wfh.sendLog('assignor',"Failed to assign. Please check the logs")
print "ERROR could not assign",wfo.name
else:
pass
print "Assignment summary:"
sendLog('assignor',"Assigned %d Stalled %s"%(n_assigned, n_stalled))
def transferor(url ,specific = None, talk=True, options=None):
if userLock('transferor'): return
if options and options.test:
execute = False
else:
execute = True
SI = siteInfo()
CI = campaignInfo()
mcm = McMClient(dev=False)
dss = DSS()
print "counting all being handled..."
being_handled = len(session.query(Workflow).filter(Workflow.status == 'away').all())
being_handled += len(session.query(Workflow).filter(Workflow.status.startswith('stag')).all())
being_handled += len(session.query(Workflow).filter(Workflow.status.startswith('assistance')).all())
max_to_handle = options.maxworkflows
allowed_to_handle = max(0,max_to_handle - being_handled)
wf_buffer = 5
if allowed_to_handle<=wf_buffer: ## buffer for having several wf per transfer
print "Not allowed to run more than",max_to_handle,"at a time. Currently",being_handled,"and",wf_buffer,"buffer"
else:
print being_handled,"already being handled",max_to_handle,"max allowed,",allowed_to_handle,"remaining","and",wf_buffer,"buffer"
print "... done"
all_transfers=defaultdict(list)
workflow_dependencies = defaultdict(set) ## list of wf.id per input dataset
wfs_and_wfh=[]
print "getting all wf to consider ..."
cache = getWorkflows(url, 'assignment-approved', details=True)
for wfo in session.query(Workflow).filter(Workflow.status=='considered').all():
if specific and not specific in wfo.name: continue
cache_r =filter(lambda d:d['RequestName']==wfo.name, cache)
if len(cache_r):
wfs_and_wfh.append( (wfo, workflowInfo( url, wfo.name, spec=False, request = cache_r[0]) ) )
else:
wfs_and_wfh.append( (wfo, workflowInfo( url, wfo.name, spec=False) ) )
print "... done"
input_sizes = {}
## list the size of those in transfer already
in_transfer_priority=0
min_transfer_priority=100000000
print "getting all wf in staging ..."
for wfo in session.query(Workflow).filter(Workflow.status=='staging').all():
wfh = workflowInfo( url, wfo.name, spec=False)
(_,primary,_,_) = wfh.getIO()
for prim in primary:
input_sizes[prim] = dss.get( prim )
in_transfer_priority = max(in_transfer_priority, int(wfh.request['RequestPriority']))
min_transfer_priority = min(min_transfer_priority, int(wfh.request['RequestPriority']))
print "... done"
print "Max priority in transfer already",in_transfer_priority
print "Min priority in transfer already",min_transfer_priority
in_transfer_already = sum(input_sizes.values())
#sort by priority higher first
wfs_and_wfh.sort(cmp = lambda i,j : cmp(int(i[1].request['RequestPriority']),int(j[1].request['RequestPriority']) ), reverse=True)
## list the size of all inputs
print "getting all input sizes ..."
for (wfo,wfh) in wfs_and_wfh:
(_,primary,_,_) = wfh.getIO()
for prim in primary:
input_sizes[prim] = dss.get( prim )
print "... done"
grand_total = sum(input_sizes.values())
to_transfer = grand_total - in_transfer_already
grand_transfer_limit = options.maxtransfer
transfer_limit = grand_transfer_limit - in_transfer_already
print "%15.4f GB already being transfered"%in_transfer_already
print "%15.4f GB is the current requested transfer load"%to_transfer
print "%15.4f GB is the global transfer limit"%grand_transfer_limit
print "%15.4f GB is the available limit"%transfer_limit
# the max priority value per dataset.
max_priority = defaultdict(int)
needs_transfer=0 ## so that we can count'em
passing_along = 0
transfer_sizes={}
went_over_budget=False
for (wfo,wfh) in wfs_and_wfh:
print wfh.request['RequestPriority']
print wfo.name,"to be transfered"
#wfh = workflowInfo( url, wfo.name)
(_,primary,_,_) = wfh.getIO()
this_load=sum([input_sizes[prim] for prim in primary])
if ( this_load and (sum(transfer_sizes.values())+this_load > transfer_limit or went_over_budget ) ):
if went_over_budget:
print "Transfer has gone over bubget."
else:
print "Transfer will go over bubget."
print "%15.4f GB this load"%this_load
print "%15.4f GB already this round"%sum(transfer_sizes.values())
print "%15.4f GB is the available limit"%transfer_limit
went_over_budget=True
if int(wfh.request['RequestPriority']) >= in_transfer_priority and min_transfer_priority!=in_transfer_priority:
print "Higher priority sample",wfh.request['RequestPriority'],">=",in_transfer_priority,"go-on over budget"
else:
if not options.go:
print min_transfer_priority,"minimum priority",wfh.request['RequestPriority'],"<",in_transfer_priority,"stop"
continue
## throtlle by campaign go
if not CI.go( wfh.request['Campaign'] ):
print "No go for",wfh.request['Campaign']
if not options.go: continue
## check if the batch is announced
announced=False
is_real=False
for b in mcm.getA('batches',query='contains=%s'% wfo.name):
is_real = True
if b['status']=='announced':
announced=True
break
if not announced:
print wfo.name,"does not look announced."# skipping?, rejecting?, reporting?"
if not is_real:
print wfo.name,"does not appear to be genuine."
## prevent any duplication. if the wf is not mentioned in any batch, regardless of status
continue
## check on a grace period
injection_time = time.mktime(time.strptime('.'.join(map(str,wfh.request['RequestDate'])),"%Y.%m.%d.%H.%M.%S")) / (60.*60.)
now = time.mktime(time.gmtime()) / (60.*60.)
if float(now - injection_time) < 4.:
if not options.go and not announced:
print "It is too soon to start transfer: %3.2fH remaining"%(now - injection_time)
continue
passing_along += 1
if passing_along >= allowed_to_handle:
if int(wfh.request['RequestPriority']) >= in_transfer_priority and min_transfer_priority!=in_transfer_priority:
print "Higher priority sample",wfh.request['RequestPriority'],">=",in_transfer_priority,"go-on over",max_to_handle
else:
print "Not allowed to pass more than",max_to_handle,"at a time. Currently",being_handled,"handled, and adding",passing_along
break
(lheinput,primary,parent,secondary) = wfh.getIO()
if options and options.tosites:
sites_allowed = options.tosites.split(',')
else:
sites_allowed = getSiteWhiteList( (lheinput,primary,parent,secondary) )
if 'SiteWhitelist' in CI.parameters(wfh.request['Campaign']):
sites_allowed = CI.parameters(wfh.request['Campaign'])['SiteWhitelist']
blocks = []
if 'BlockWhitelist' in wfh.request and wfh.request['BlockWhitelist']:
blocks = wfh.request['BlockWhitelist']
can_go = True
staging=False
if primary:
if talk:
print wfo.name,'reads',', '.join(primary),'in primary'
## chope the primary dataset
for prim in primary:
max_priority[prim] = max(max_priority[prim],int(wfh.request['RequestPriority']))
sites_really_allowed = [site for site in sites_allowed if not any([osite.startswith(site) for osite in SI.sites_veto_transfer])]
print "Sites allowed minus the vetoed transfer"
print sites_really_allowed
copies_needed = int(0.35*len(sites_really_allowed))+1 ## should just go for a fixed number based if the white list grows that big
print "Would make",copies_needed,"copies"
if options.maxcopy>0:
copies_needed = min(options.maxcopy,copies_needed)
## remove the sites that do not want transfers
print "need",copies_needed
workflow_dependencies[prim].add( wfo.id )
presence = getDatasetPresence( url, prim )
prim_location = [site for site,pres in presence.items() if pres[0]==True]
if len(prim_location) >= copies_needed:
print "The output is all fully in place at",len(prim_location),"sites"
continue
# reduce the number of copies required by existing full copies
copies_needed = max(0,copies_needed - len(prim_location))
print "now need",copies_needed
subscriptions = listSubscriptions( url , prim )
prim_destination = list(set([site for (site,(tid,decision)) in subscriptions.items() if decision and not any([site.endswith(veto) for veto in ['MSS','Export','Buffer']])]))
## need to reject from that list the ones with a full copy already: i.e the transfer corresponds to the copy in place
prim_destination = [site for site in prim_destination if not site in prim_location]
## add transfer dependencies
latching_on_transfers = list(set([ tid for (site,(tid,decision)) in subscriptions.items() if decision and site in prim_destination and not any([site.endswith(veto) for veto in ['MSS','Export','Buffer']])]))
print latching_on_transfers
for latching in latching_on_transfers:
tfo = session.query(Transfer).filter(Transfer.phedexid == latching).first()
if not tfo:
tfo = Transfer( phedexid = latching)
tfo.workflows_id = []
session.add(tfo)
if not wfo.id in tfo.workflows_id:
print "adding",wfo.id,"to",tfo.id,"with phedexid",latching
l = copy.deepcopy( tfo.workflows_id )
l.append( wfo.id )
tfo.workflows_id = l
if not options.test:
session.commit()
else:
session.flush() ## regardless of commit later on, we need to let the next wf feeding on this transfer to see it in query
can_go = False
transfer_sizes[prim] = input_sizes[prim]
staging = True
# reduce the number of copies required by the on-going full transfer : how do we bootstrap on waiting for them ??
copies_needed = max(0,copies_needed - len(prim_destination))
print "then need",copies_needed
if copies_needed == 0:
print "The output is either fully in place or getting in full somewhere with",latching_on_transfers
can_go = True
continue
prim_to_distribute = [site for site in sites_allowed if not any([osite.startswith(site) for osite in prim_location])]
prim_to_distribute = [site for site in prim_to_distribute if not any([osite.startswith(site) for osite in prim_destination])]
## take out the ones that cannot receive transfers
prim_to_distribute = [site for site in prim_to_distribute if not any([osite.startswith(site) for osite in SI.sites_veto_transfer])]
if len(prim_to_distribute)>0: ## maybe that a parameter we can play with to limit the
if not options or options.chop:
spreading = distributeToSites( getDatasetChops(prim), prim_to_distribute, n_copies = copies_needed, weights=SI.cpu_pledges)
else:
spreading = {}
for site in prim_to_distribute: spreading[site]=[prim]
can_go = False
transfer_sizes[prim] = input_sizes[prim]
for (site,items) in spreading.items():
all_transfers[site].extend( items )
if secondary:
if talk:
print wfo.name,'reads',', '.join(secondary),'in secondary'
for sec in secondary:
workflow_dependencies[sec].add( wfo.id )
presence = getDatasetPresence( url, sec )
sec_location = [site for site,pres in presence.items() if pres[1]>90.] ## more than 90% of the minbias at sites
subscriptions = listSubscriptions( url ,sec )
sec_destination = [site for site in subscriptions]
sec_to_distribute = [site for site in sites_allowed if not any([osite.startswith(site) for osite in sec_location])]
sec_to_distribute = [site for site in sec_to_distribute if not any([osite.startswith(site) for osite in sec_destination])]
sec_to_distribute = [site for site in sec_to_distribute if not any([osite.startswith(site) for osite in SI.sites_veto_transfer])]
if len( sec_to_distribute )>0:
for site in sec_to_distribute:
all_transfers[site].append( sec )
can_go = False
## is that possible to do something more
if can_go:
## no explicit transfer required this time
if staging:
## but using existing ones
print wfo.name,"latches on existing transfers, and nothing else"
wfo.status = 'staging'
else:
print wfo.name,"should just be assigned NOW to",sites_allowed
wfo.status = 'staged'
print "setting status to",wfo.status
session.commit()
continue
else:
## there is an explicit transfer required
if staging:
## and also using an existing one
print wfo.name,"latches on existing transfers"
if not options.test:
wfo.status = 'staging'
print "setting status to",wfo.status
session.commit()
print wfo.name,"needs a transfer"
needs_transfer+=1
#print json.dumps(all_transfers)
fake_id=-1
wf_id_in_prestaging=set()
for (site,items_to_transfer) in all_transfers.iteritems():
items_to_transfer = list(set(items_to_transfer))
## convert to storage element
site_se = SI.CE_to_SE(site)
## site that do not want input datasets
if site in SI.sites_veto_transfer:
print site,"does not want transfers"
continue
## throttle the transfer size to T2s ? we'd be screwed by a noPU sample properly configured.
## massage a bit the items
blocks = [it for it in items_to_transfer if '#' in it]
datasets = [it for it in items_to_transfer if not '#' in it]
if execute:
print "Making a replica to",site,"(CE)",site_se,"(SE) for"
else:
print "Would make a replica to",site,"(CE)",site_se,"(SE) for"
print "\t",len(blocks),"blocks"
## remove blocks if full dataset is send out
blocks = [block for block in blocks if not block.split('#')[0] in datasets]
print "\t",len(blocks),"needed blocks for",list(set([block.split('#')[0] for block in blocks]))
print "\t",len(datasets),"datasets"
print "\t",datasets
items_to_transfer = blocks + datasets
## operate the transfer
if options and options.stop:
## ask to move-on
answer = raw_input('Continue with that ?')
if not answer.lower() in ['y','yes','go']:
continue
if execute:
result = makeReplicaRequest(url, site_se, items_to_transfer, 'prestaging', priority='normal')
## make use of max_priority dataset:priority to set the subscriptions priority
"""
## does not function
once = True
for item in items_to_transfer:
bds = item.split('#')[0]
if max_priority[bds] >= 90000:
if once:
w=10
print "waiting",w,"s before raising priority"
time.sleep(w)
once=False
## raise it to high priority
print item,"subscription priority raised to high at",site_se
#print "This does not work yet properly it seems"
print updateSubscription(url, site_se, item, priority='high')
"""
else:
#result= {'phedex':{'request_created' : [{'id' : fake_id}]}}
result= {'phedex':{'request_created' : []}}
fake_id-=1
if not result:
print "ERROR Could not make a replica request for",site,items_to_transfer,"pre-staging"
continue
for phedexid in [o['id'] for o in result['phedex']['request_created']]:
new_transfer = session.query(Transfer).filter(Transfer.phedexid == phedexid).first()
print phedexid,"transfer created"
if not new_transfer:
new_transfer = Transfer( phedexid = phedexid)
session.add( new_transfer )
new_transfer.workflows_id = set()
for transfering in list(set(map(lambda it : it.split('#')[0], items_to_transfer))):
new_transfer.workflows_id.update( workflow_dependencies[transfering] )
new_transfer.workflows_id = list(new_transfer.workflows_id)
wf_id_in_prestaging.update(new_transfer.workflows_id)
session.commit()
## auto approve it
if execute:
approved = approveSubscription(url, phedexid, [site_se])
for wfid in wf_id_in_prestaging:
tr_wf = session.query(Workflow).get(wfid)
if tr_wf and tr_wf.status!='staging':
if execute:
tr_wf.status = 'staging'
if talk:
print "setting",tr_wf.name,"to staging"
session.commit()
def assignor(url ,specific = None, talk=True, options=None):
if userLock(): return
if duplicateLock(): return
#if notRunningBefore( 'stagor' ): return
if not componentInfo().check(): return
UC = unifiedConfiguration()
CI = campaignInfo()
SI = global_SI
#LI = lockInfo()
NLI = newLockInfo()
n_assigned = 0
n_stalled = 0
wfos=[]
if specific:
wfos = session.query(Workflow).filter(Workflow.name==specific).all()
if not wfos:
if specific:
wfos = session.query(Workflow).filter(Workflow.status=='considered').all()
wfos.extend( session.query(Workflow).filter(Workflow.status=='staging').all())
wfos.extend(session.query(Workflow).filter(Workflow.status=='staged').all())
for wfo in wfos:
if specific:
if not any(map(lambda sp: sp in wfo.name,specific.split(','))): continue
#if not specific in wfo.name: continue
print "\n\n",wfo.name,"\n\tto be assigned"
wfh = workflowInfo( url, wfo.name)
## check if by configuration we gave it a GO
if not CI.go( wfh.request['Campaign'] ) and not options.go:
print "No go for",wfh.request['Campaign']
n_stalled+=1
continue
## check on current status for by-passed assignment
if wfh.request['RequestStatus'] !='assignment-approved':
if not options.test:
print wfo.name,wfh.request['RequestStatus'],"setting away and skipping"
## the module picking up from away will do what is necessary of it
wfo.wm_status = wfh.request['RequestStatus']
wfo.status = 'away'
session.commit()
continue
else:
print wfo.name,wfh.request['RequestStatus']
## retrieve from the schema, dbs and reqMgr what should be the next version
version=wfh.getNextVersion()
if not version:
if options and options.ProcessingVersion:
version = options.ProcessingVersion
else:
print "cannot decide on version number"
n_stalled+=1
continue
#(lheinput,primary,parent,secondary) = wfh.getIO()
#sites_allowed = getSiteWhiteList( (lheinput,primary,parent,secondary) )
(lheinput,primary,parent,secondary, sites_allowed) = wfh.getSiteWhiteList()
print "Site white list",sorted(sites_allowed)
override_sec_location = CI.get(wfh.request['Campaign'], 'SecondaryLocation', [])
c_sites_allowed = CI.get(wfh.request['Campaign'], 'SiteWhitelist' , [])
if c_sites_allowed:
print "Would like to use the new whitelist, but will not until things went through a bit"
sendEmail("using a restricted site white list","for %s"%(c_sites_allowed))
sites_allowed = list(set(sites_allowed) & set(c_sites_allowed))
c_black_list = CI.get(wfh.request['Campaign'], 'SiteBlacklist', [])
if c_black_list:
print "Reducing the whitelist due to black list in campaign configuration"
print "Removing",c_black_list
sites_allowed = list(set(sites_allowed) - set(c_black_list))
blocks = []
if 'BlockWhitelist' in wfh.request:
blocks = wfh.request['BlockWhitelist']
ncores = wfh.request.get('Multicore',1)
memory_allowed = SI.sitesByMemory( wfh.request['Memory'] , maxCore=ncores)
if memory_allowed!=None:
print "sites allowing", wfh.request['Memory'],"MB and",ncores,"core are",memory_allowed
sites_allowed = list(set(sites_allowed) & set(memory_allowed))
print "Allowed",sorted(sites_allowed)
secondary_locations=None
for sec in list(secondary):
if override_sec_location:
print "We don't care where the secondary is"
print "Cannot pass for now"
sendEmail("tempting to pass sec location check","but we cannot yet IMO")
#pass
presence = getDatasetPresence( url, sec )
print sec
print json.dumps(presence, indent=2)
one_secondary_locations = [site for (site,(there,frac)) in presence.items() if frac>98.]
#one_secondary_locations = [site for (site,(there,frac)) in presence.items() if there]
if secondary_locations==None:
secondary_locations = one_secondary_locations
else:
secondary_locations = list(set(secondary_locations) & set(one_secondary_locations))
## reduce the site white list to site with secondary only
#sites_allowed = [site for site in sites_allowed if any([osite.startswith(site) for osite in one_secondary_locations])]
sites_allowed = [site for site in sites_allowed if SI.CE_to_SE(site) in one_secondary_locations]
print "From secondary requirement, now Allowed",sorted(sites_allowed)
initial_sites_allowed = copy.deepcopy( sites_allowed ) ## keep track of this, after secondary input location restriction : that's how you want to operate it
sites_all_data = copy.deepcopy( sites_allowed )
sites_with_data = copy.deepcopy( sites_allowed )
sites_with_any_data = copy.deepcopy( sites_allowed )
primary_locations = None
available_fractions = {}
set_lfn = '/store/mc' ## by default
for prim in list(primary):
set_lfn = getLFNbase( prim )
presence = getDatasetPresence( url, prim , only_blocks=blocks)
if talk:
print prim
print json.dumps(presence, indent=2)
available_fractions[prim] = getDatasetBlocksFraction(url, prim, sites = [SI.CE_to_SE(site) for site in sites_allowed] , only_blocks = blocks)
#sites_all_data = [site for site in sites_with_data if any([osite.startswith(site) for osite in [psite for (psite,(there,frac)) in presence.items() if there]])]
#sites_with_data = [site for site in sites_with_data if any([osite.startswith(site) for osite in [psite for (psite,frac) in presence.items() if frac[1]>90.]])]
sites_all_data = [site for site in sites_with_data if SI.CE_to_SE(site) in [psite for (psite,(there,frac)) in presence.items() if there]]
sites_with_data = [site for site in sites_with_data if SI.CE_to_SE(site) in [psite for (psite,frac) in presence.items() if frac[1]>90.]]
sites_with_any_data = [site for site in sites_with_any_data if SI.CE_to_SE(site) in presence.keys()]
print "Holding the data but not allowed",list(set([se_site for se_site in presence.keys() if not SI.SE_to_CE(se_site) in sites_allowed]))
if primary_locations==None:
primary_locations = presence.keys()
else:
primary_locations = list(set(primary_locations) & set(presence.keys() ))
sites_with_data = list(set(sites_with_data))
sites_with_any_data = list(set(sites_with_any_data))
opportunistic_sites=[]
down_time = False
## opportunistic running where any piece of data is available
if secondary_locations or primary_locations:
## intersection of both any pieces of the primary and good IO
#opportunistic_sites = [SI.SE_to_CE(site) for site in list((set(secondary_locations) & set(primary_locations) & set(SI.sites_with_goodIO)) - set(sites_allowed))]
if secondary_locations and primary_locations:
opportunistic_sites = [SI.SE_to_CE(site) for site in list((set(secondary_locations) & set(primary_locations)) - set([SI.CE_to_SE(site) for site in sites_allowed]))]
elif primary_locations:
opportunistic_sites = [SI.SE_to_CE(site) for site in list(set(primary_locations) - set([SI.CE_to_SE(site) for site in sites_allowed]))]
else:
opportunistic_sites = []
print "We could be running at",sorted(opportunistic_sites),"in addition"
if any([osite in SI.sites_not_ready for osite in opportunistic_sites]):
print "One of the destination site is in downtime"
down_time = True
## should this be send back to considered ?
"""
if available_fractions and not all([available>=1. for available in available_fractions.values()]):
print "The input dataset is not located in full over sites"
print json.dumps(available_fractions)
if not options.test and not options.go:
known = []
try:
known = json.loads(open('cannot_assign.json').read())
except:
pass
if not wfo.name in known:
sendEmail( "cannot be assigned","%s is not full over sites \n %s"%(wfo.name,json.dumps(available_fractions)))
known.append( wfo.name )
open('cannot_assign.json','w').write(json.dumps( known, indent=2))
n_stalled+=1
continue ## skip skip skip
"""
## should be 2 but for the time-being let's lower it to get things going
copies_wanted,cpuh = wfh.getNCopies()
less_copies_than_requested = UC.get("less_copies_than_requested")
copies_wanted = max(1,copies_wanted-less_copies_than_requested) # take one out for the efficiency
if available_fractions and not all([available>=copies_wanted for available in available_fractions.values()]):
not_even_once = not all([available>=1. for available in available_fractions.values()])
print "The input dataset is not available",copies_wanted,"times, only",available_fractions.values()
if down_time and not options.go:
wfo.status = 'considered'
session.commit()
print "sending back to considered because of site downtime, instead of waiting"
sendEmail( "cannot be assigned due to downtime","%s is not sufficiently available, due to down time of a site in the whitelist. check the assignor logs. sending back to considered."% wfo.name)
continue
#pass
print json.dumps(available_fractions)
if (options.go and not_even_once) or not options.go:
known = []
try:
known = json.loads(open('cannot_assign.json').read())
except:
pass
if not wfo.name in known:
sendEmail( "cannot be assigned","%s is not sufficiently available. Probably phedex information lagging behind. \n %s"%(wfo.name,json.dumps(available_fractions)))
known.append( wfo.name )
open('cannot_assign.json','w').write(json.dumps( known, indent=2))
n_stalled+=1
continue
## default back to white list to original white list with any data
print "Allowed",sites_allowed
if options.primary_aaa:
sites_allowed = initial_sites_allowed
options.useSiteListAsLocation = True
else:
sites_allowed = sites_with_any_data
print "Selected for any data",sites_allowed
if options.restrict:
print "Allowed",sites_allowed
sites_allowed = sites_with_any_data
print "Selected",sites_allowed
else:
if set(sites_with_data) != set(sites_allowed):
## the data is not everywhere we wanted to run at : enable aaa
print "Sites with 90% data not matching site white list (block choping!)"
print "Resorting to AAA reading for",list(set(sites_allowed) - set(sites_with_data)),"?"
print "Whitelist site with any data",list(set(sites_allowed) - set(sites_with_any_data))
#options.useSiteListAsLocation = True
#print "Not commissioned yet"
#continue
#print "We could be running at",opportunistic_sites,"in addition"
##sites_allowed = list(set(sites_allowed+ opportunistic_sites))
if not len(sites_allowed):
print wfo.name,"cannot be assign with no matched sites"
sendEmail( "cannot be assigned","%s has no whitelist"%(wfo.name))
n_stalled+=1
continue
t1_only = [ce for ce in sites_allowed if ce.startswith('T1')]
if t1_only:
# try to pick from T1 only first
sites_out = [SI.pick_dSE([SI.CE_to_SE(ce) for ce in t1_only])]
else:
# then pick any otherwise
sites_out = [SI.pick_dSE([SI.CE_to_SE(ce) for ce in sites_allowed])]
print "Placing the output on", sites_out
parameters={
'SiteWhitelist' : sites_allowed,
#'CustodialSites' : sites_custodial,
'NonCustodialSites' : sites_out,
'AutoApproveSubscriptionSites' : list(set(sites_out)),
'AcquisitionEra' : wfh.acquisitionEra(),
'ProcessingString' : wfh.processingString(),
'MergedLFNBase' : set_lfn,
'ProcessingVersion' : version,
}
## plain assignment here
team='production'
if options and options.team:
team = options.team
#if wfh.request['RequestPriority'] >= 100000 and (wfh.request['TimePerEvent']*int(wfh.getRequestNumEvents()))/(8*3600.) < 10000:
# team = 'highprio'
# sendEmail("sending work with highprio team","%s"% wfo.name, destination=['*****@*****.**'])
if "T2_US_UCSD" in sites_with_data and random.random() < -0.5 and wfh.request['Campaign']=='RunIISpring15DR74' and int(wfh.getRequestNumEvents()) < 600000 and not any([out.endswith('RAW') for out in wfh.request['OutputDatasets']]):
## consider SDSC
parameters['SiteWhitelist'] = ['T2_US_UCSD','T3_US_SDSC']
parameters['useSiteListAsLocation'] = True
team = 'allocation-based'
sendEmail("sending work to SDSC","%s was assigned to SDSC/UCSD"% wfo.name, destination=['*****@*****.**'])
if wfh.request['Campaign']=='RunIIWinter15GS' and random.random() < -1.0:
parameters['SiteWhitelist'] = ['T3_US_SDSC']
team = 'allocation-based'
sendEmail("sending work to SDSC","%s was assigned to SDSC"% wfo.name, destination=['*****@*****.**'])
##parse options entered in command line if any
if options:
for key in reqMgrClient.assignWorkflow.keys:
v=getattr(options,key)
if v!=None:
if type(v)==str and ',' in v:
parameters[key] = filter(None,v.split(','))
else:
parameters[key] = v
## pick up campaign specific assignment parameters
parameters.update( CI.parameters(wfh.request['Campaign']) )
if not options.test:
parameters['execute'] = True
split_check = wfh.checkWorkflowSplitting()
if split_check!=True:
parameters.update( split_check )
if 'EventBased' in split_check.values():
print "Falling back to event splitting."
sendEmail("Fallback to EventBased","the workflow %s is too heavy to be processed as it is. Fallback to EventBased splitting"%wfo.name)
elif 'EventsPerJob' in split_check.values():
print "Modifying the number of job per event"
sendEmail("Modifying the job per events","the workflow %s is too heavy in number of jobs explosion"%wfo.name)
# Handle run-dependent MC
pstring = wfh.processingString()
if 'PU_RD' in pstring:
numEvents = wfh.getRequestNumEvents()
eventsPerLumi = [getDatasetEventsPerLumi(prim) for prim in primary]
eventsPerLumi = sum(eventsPerLumi)/float(len(eventsPerLumi))
reqJobs = 500
if 'PU_RD2' in pstring:
reqJobs = 2000
eventsPerJob = int(numEvents/(reqJobs*1.4))
lumisPerJob = int(eventsPerJob/eventsPerLumi)
if lumisPerJob==0:
print "There is no go for assigning that request without event splitting"
sendEmail("issue with event splitting for run-dependent MC","%s needs to be split by event with %s per job"%(wfo.name, eventsPerJob))
print "need to go down to",eventsPerJob,"events per job"
parameters['EventsPerJob'] = eventsPerJob
else:
spl = wfh.getSplittings()[0]
eventsPerJobEstimated = spl['events_per_job'] if 'events_per_job' in spl else None
eventsPerJobEstimated = spl['avg_events_per_job'] if 'avg_events_per_job' in spl else None
if eventsPerJobEstimated and eventsPerJobEstimated > eventsPerJob:
print "need to go down to",lumisPerJob,"in assignment"
sendEmail("setting lumi splitting for run-dependent MC","%s was assigned with %s lumis/job"%( wfo.name, lumisPerJob))
parameters['LumisPerJob'] = lumisPerJob
else:
print "the regular splitting should work for",pstring
sendEmail("leaving splitting untouched for PU_RD*","please check on "+wfo.name)
result = reqMgrClient.assignWorkflow(url, wfo.name, team, parameters)
# set status
if not options.test:
if result:
wfo.status = 'away'
session.commit()
n_assigned+=1
try:
## refetch information and lock output
new_wfi = workflowInfo( url, wfo.name)
(_,prim,_,sec) = new_wfi.getIO()
for secure in list(prim)+list(sec)+new_wfi.request['OutputDatasets']:
## lock all outputs flat
NLI.lock( secure )
#for site in [SI.CE_to_SE(site) for site in sites_allowed]:
# for output in new_wfi.request['OutputDatasets']:
# LI.lock( output, site, 'dataset in production')
# for primary in prim:
# LI.lock( primary, site, 'dataset used in input')
# for secondary in sec:
# LI.lock( secondary, site, 'required for mixing' )
except Exception as e:
print "fail in locking output"
print str(e)
sendEmail("failed locking of output",str(e))
else:
print "ERROR could not assign",wfo.name
else:
pass
print "Assignment summary:"
print "Assigned",n_assigned
print "Stalled",n_stalled
#!/usr/bin/env python from utils import getDatasetPresence import json import sys url = "cmsweb.cern.ch" presence = getDatasetPresence(url, sys.argv[1]) print json.dumps(presence, indent=2)
