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 completor(url, specific):
use_mcm = True
up = componentInfo(mcm=use_mcm, soft=['mcm'])
if not up.check(): return
use_mcm = up.status['mcm']
if use_mcm:
mcm = McMClient(dev=False)
CI = campaignInfo()
SI = siteInfo()
UC = unifiedConfiguration()
wfs = []
wfs.extend( session.query(Workflow).filter(Workflow.status == 'away').all() )
wfs.extend( session.query(Workflow).filter(Workflow.status.startswith('assistance')).all() )
## just take it in random order so that not always the same is seen
random.shuffle( wfs )
## by workflow a list of fraction / timestamps
completions = json.loads( open('%s/completions.json'%monitor_dir).read())
good_fractions = {}
timeout = {}
for c in CI.campaigns:
if 'force-complete' in CI.campaigns[c]:
good_fractions[c] = CI.campaigns[c]['force-complete']
if 'force-timeout' in CI.campaigns[c]:
timeout[c] = CI.campaigns[c]['force-timeout']
long_lasting = {}
overrides = getForceCompletes()
if use_mcm:
## add all workflow that mcm wants to get force completed
mcm_force = mcm.get('/restapi/requests/forcecomplete')
## assuming this will be a list of actual prepids
overrides['mcm'] = mcm_force
print "can force complete on"
print json.dumps( good_fractions ,indent=2)
print json.dumps( overrides, indent=2)
max_force = UC.get("max_force_complete")
#wfs_no_location_in_GQ = set()
#block_locations = defaultdict(lambda : defaultdict(list))
#wfs_no_location_in_GQ = defaultdict(list)
set_force_complete = set()
for wfo in wfs:
if specific and not specific in wfo.name: continue
print "looking at",wfo.name
## get all of the same
wfi = workflowInfo(url, wfo.name)
pids = wfi.getPrepIDs()
skip=False
if not any([c in wfo.name for c in good_fractions]): skip=True
for user,spec in overrides.items():
if wfi.request['RequestStatus']!='force-complete':
if any(s in wfo.name for s in spec) or (wfo.name in spec) or any(pid in spec for pid in pids) or any(s in pids for s in spec):
sendEmail('force-complete requested','%s is asking for %s to be force complete'%(user,wfo.name))
wfi = workflowInfo(url, wfo.name)
forceComplete(url , wfi )
skip=True
wfi.notifyRequestor("The workflow %s was force completed by request of %s"%(wfo.name,user), do_batch=False)
wfi.sendLog('completor','%s is asking for %s to be force complete'%(user,wfo.name))
if user == 'mcm' and use_mcm:
for pid in wfi.getPrepIDs():
mcm.delete('/restapi/requests/forcecomplete/%s'%pid)
#sendEmail('completor test','not force completing automatically, you have to go back to it')
#skip=False
break
if wfo.status.startswith('assistance'): skip = True
if skip:
continue
priority = wfi.request['RequestPriority']
if not 'Campaign' in wfi.request: continue
if not wfi.request['RequestStatus'] in ['acquired','running-open','running-closed']: continue
c = wfi.request['Campaign']
if not c in good_fractions: continue
good_fraction = good_fractions[c]
ignore_fraction = 2.
lumi_expected = None
event_expected = None
if not 'TotalInputEvents' in wfi.request:
if 'RequestNumEvents' in wfi.request:
event_expected = wfi.request['RequestNumEvents']
else:
print "truncated, cannot do anything"
continue
else:
lumi_expected = wfi.request['TotalInputLumis']
event_expected = wfi.request['TotalInputEvents']
now = time.mktime(time.gmtime()) / (60*60*24.)
running_log = filter(lambda change : change["Status"] in ["running-open","running-closed"],wfi.request['RequestTransition'])
if not running_log:
print "\tHas no running log"
# cannot figure out when the thing started running
continue
then = running_log[-1]['UpdateTime'] / (60.*60.*24.)
delay = now - then ## in days
(w,d) = divmod(delay, 7 )
print "\t"*int(w)+"Running since",delay,"[days] priority=",priority
monitor_delay = 7
allowed_delay = 14
if c in timeout:
allowed_delay = timeout[c]
monitor_delay = min(monitor_delay, allowed_delay)
### just skip if too early
if delay <= monitor_delay: continue
long_lasting[wfo.name] = { "delay" : delay }
percent_completions = {}
for output in wfi.request['OutputDatasets']:
if "/DQM" in output: continue ## that does not count
if not output in completions: completions[output] = { 'injected' : None, 'checkpoints' : [], 'workflow' : wfo.name}
## get completion fraction
event_count,lumi_count = getDatasetEventsAndLumis(dataset=output)
lumi_completion=0.
event_completion=0.
if lumi_expected:
lumi_completion = lumi_count / float( lumi_expected )
if event_expected:
event_completion = event_count / float( event_expected )
#take the less optimistic
percent_completions[output] = min( lumi_completion, event_completion )
completions[output]['checkpoints'].append( (now, event_completion ) )
if all([percent_completions[out] >= good_fraction for out in percent_completions]):
wfi.sendLog('completor', "all is above %s \n%s"%( good_fraction,
json.dumps( percent_completions, indent=2 )
))
else:
long_lasting[wfo.name].update({
'completion': sum(percent_completions.values()) / len(percent_completions),
'completions' : percent_completions
})
## do something about the agents this workflow is in
long_lasting[wfo.name]['agents'] = wfi.getAgents()
wfi.sendLog('completor', "%s not over bound %s\n%s"%(percent_completions.values(), good_fraction,
json.dumps( long_lasting[wfo.name]['agents'], indent=2) ))
continue
if all([percent_completions[out] >= ignore_fraction for out in percent_completions]):
print "all is done, just wait a bit"
continue
for output in percent_completions:
completions[output]['injected'] = then
#further check on delays
cpuh = wfi.getComputingTime(unit='d')
ran_at = wfi.request['SiteWhitelist']
wfi.sendLog('completor',"Required %s, time spend %s"%( cpuh, delay))
##### WILL FORCE COMPLETE BELOW
# only really force complete after n days
if delay <= allowed_delay: continue
## find ACDCs that might be running
if max_force>0:
forceComplete(url, wfi )
set_force_complete.add( wfo.name )
print "going for force-complete of",wfo.name
wfi.sendLog('completor','going for force completing')
wfi.notifyRequestor("The workflow %s was force completed for running too long"% wfo.name)
max_force -=1
else:
wfi.sendLog('completor',"too many completion this round, cannot force complete")
## do it once only for testing
#break
if set_force_complete:
sendLog('completor','The followings were set force-complete \n%s'%('\n'.join(set_force_complete)))
#sendEmail('set force-complete', 'The followings were set force-complete \n%s'%('\n'.join(set_force_complete)))
open('%s/completions.json'%monitor_dir,'w').write( json.dumps( completions , indent=2))
text="These have been running for long"
open('%s/longlasting.json'%monitor_dir,'w').write( json.dumps( long_lasting, indent=2 ))
for wf,info in sorted(long_lasting.items(), key=lambda tp:tp[1]['delay'], reverse=True):
delay = info['delay']
text += "\n %s : %s days"% (wf, delay)
if 'completion' in info:
text += " %d%%"%( info['completion']*100 )
#if wfs_no_location_in_GQ:
# sendEmail('workflow with no location in GQ',"there won't be able to run anytime soon\n%s"%( '\n'.join(wfs_no_location_in_GQ)))
#sendEmail("long lasting workflow",text)
## you can check the log
print text
def completor(url, specific):
CI = campaignInfo()
wfs = []
wfs.extend(session.query(Workflow).filter(Workflow.status == "away").all())
##wfs.extend( session.query(Workflow).filter(Workflow.status.startswith('assistance')).all() )
## just take it in random order
random.shuffle(wfs)
## by workflow a list of fraction / timestamps
completions = json.loads(open("completions.json").read())
good_fractions = {}
for c in CI.campaigns:
if "force-complete" in CI.campaigns[c]:
good_fractions[c] = CI.campaigns[c]["force-complete"]
print "can force complete on"
print json.dumps(good_fractions, indent=2)
for wfo in wfs:
if specific and not specific in wfo.name:
continue
if not any([c in wfo.name for c in good_fractions]):
continue
print "looking at", wfo.name
## get all of the same
wfi = workflowInfo(url, wfo.name)
if not "Campaign" in wfi.request:
continue
c = wfi.request["Campaign"]
if not c in good_fractions:
continue
good_fraction = good_fractions[c]
ignore_fraction = 2.0
lumi_expected = None
event_expected = None
if not "TotalInputEvents" in wfi.request:
if "RequestNumEvents" in wfi.request:
event_expected = wfi.request["RequestNumEvents"]
else:
continue
else:
lumi_expected = wfi.request["TotalInputLumis"]
event_expected = wfi.request["TotalInputEvents"]
now = time.mktime(time.gmtime()) / (60 * 60 * 24.0)
running_log = filter(
lambda change: change["Status"] in ["running-open", "running-closed"], wfi.request["RequestTransition"]
)
if not running_log:
print "\tHas no running log"
# cannot figure out when the thing started running
continue
then = running_log[-1]["UpdateTime"] / (60.0 * 60.0 * 24.0)
delay = now - then ## in days
(w, d) = divmod(delay, 7)
print "\t" * int(w) + "Running since", delay, "[days]"
if delay <= 4:
continue
if delay >= 7:
sendEmail("long lasting workflow", "%s has been running for %s days" % (wfo.name, delay))
percent_completions = {}
for output in wfi.request["OutputDatasets"]:
if not output in completions:
completions[output] = {"injected": None, "checkpoints": [], "workflow": wfo.name}
## get completion fraction
event_count, lumi_count = getDatasetEventsAndLumis(dataset=output)
lumi_completion = 0.0
event_completion = 0.0
if lumi_expected:
lumi_completion = lumi_count / float(lumi_expected)
if event_expected:
event_completion = event_count / float(event_expected)
# take the less optimistic
percent_completions[output] = min(lumi_completion, event_completion)
completions[output]["checkpoints"].append((now, event_completion))
if all([percent_completions[out] >= good_fraction for out in percent_completions]):
print "all is above", good_fraction, "for", wfo.name
print json.dumps(percent_completions, indent=2)
else:
print "\t", percent_completions.values(), "not over bound", good_fraction
# print json.dumps( percent_completions, indent=2 )
continue
if all([percent_completions[out] >= ignore_fraction for out in percent_completions]):
print "all is done, just wait a bit"
continue
for output in percent_completions:
completions[output]["injected"] = then
# further check on delays
cpuh = wfi.getComputingTime(unit="d")
ran_at = wfi.request["SiteWhitelist"]
print "Required:", cpuh,
print "Time spend:", delay
## find ACDCs that might be running
familly = getWorkflowById(url, wfi.request["PrepID"], details=True)
for member in familly:
### if member['RequestName'] == wl['RequestName']: continue ## set himself out
if member["RequestDate"] < wfi.request["RequestDate"]:
continue
if member["RequestStatus"] in ["None", None]:
continue
## then set force complete all members
if member["RequestStatus"] in ["running-opened", "running-closed"]:
print "setting", member["RequestName"], "force-complete"
print "NOT REALLY FORCING"
sendEmail(
"force completing",
"TAGGING %s is worth force completing\n%s" % (member["RequestName"], percent_completions),
)
##reqMgrClient.setWorkflowForceComplete(url, member['RequestName'])
## do it once only for testing
# break
open("completions.json", "w").write(json.dumps(completions, indent=2))
def closor(url, specific=None):
if not componentInfo().check(): return
CI = campaignInfo()
LI = lockInfo()
## manually closed-out workflows should get to close with checkor
for wfo in session.query(Workflow).filter(Workflow.status=='close').all():
if specific and not specific in wfo.name: continue
## what is the expected #lumis
wl = getWorkLoad(url, wfo.name)
wfo.wm_status = wl['RequestStatus']
if wl['RequestStatus'] in ['announced','normal-archived']:
## manually announced ??
wfo.status = 'done'
wfo.wm_status = wl['RequestStatus']
print wfo.name,"is announced already",wfo.wm_status
session.commit()
expected_lumis = 1
if not 'TotalInputLumis' in wl:
print wfo.name,"has not been assigned yet, or the database is corrupted"
else:
expected_lumis = wl['TotalInputLumis']
## what are the outputs
outputs = wl['OutputDatasets']
## check whether the number of lumis is as expected for each
all_OK = []
#print outputs
if len(outputs):
print wfo.name,wl['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()
print "\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)))
## only that status can let me go into announced
if wl['RequestStatus'] in ['closed-out']:
print wfo.name,"to be announced"
results=[]#'dummy']
if not results:
for (io,out) in enumerate(outputs):
if all_OK[io]:
results.append(setDatasetStatus(out, 'VALID'))
tier = out.split('/')[-1]
to_DDM = (wl['RequestType'] == 'ReDigi' and not ('DQM' in tier))
campaign = None
try:
campaign = out.split('/')[2].split('-')[0]
except:
if 'Campaign' in wl and wl['Campaign']:
campaign = wl['Campaign']
if campaign and campaign in CI.campaigns and 'toDDM' in CI.campaigns[campaign] and tier in CI.campaigns[campaign]['toDDM']:
to_DDM = True
## inject to DDM when necessary
passed_to_DDM=True
if to_DDM:
#print "Sending",out," to DDM"
status = subprocess.call(['python','assignDatasetToSite.py','--nCopies=2','--dataset='+out,'--exec'])
if status!=0:
print "Failed DDM, retrying a second time"
status = subprocess.call(['python','assignDatasetToSite.py','--nCopies=2','--dataset='+out,'--exec'])
if status!=0:
results.append("Failed DDM for %s"% out)
sendEmail("failed DDM injection","could not add "+out+" to DDM pool. check closor logs.")
passed_to_DDM=False
if passed_to_DDM:
## make a lock release
LI.release_everywhere( out, reason = 'global unlock after passing to DDM')
pass
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
results.append(reqMgrClient.announceWorkflowCascade(url, wfo.name))
#print results
if all(map(lambda result : result in ['None',None,True],results)):
wfo.status = 'done'
session.commit()
print wfo.name,"is announced"
else:
print "ERROR with ",wfo.name,"to be announced",json.dumps( results )
else:
print wfo.name,"not good for announcing:",wl['RequestStatus']
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 completor(url, specific):
use_mcm = True
up = componentInfo(mcm=use_mcm, soft=['mcm'])
if not up.check(): return
use_mcm = up.status['mcm']
if use_mcm:
mcm = McMClient(dev=False)
CI = campaignInfo()
SI = siteInfo()
UC = unifiedConfiguration()
wfs = []
wfs.extend( session.query(Workflow).filter(Workflow.status == 'away').all() )
wfs.extend( session.query(Workflow).filter(Workflow.status.startswith('assistance')).all() )
## just take it in random order so that not always the same is seen
random.shuffle( wfs )
max_per_round = UC.get('max_per_round').get('completor',None)
if max_per_round and not specific: wfs = wfs[:max_per_round]
## by workflow a list of fraction / timestamps
completions = json.loads( open('%s/completions.json'%monitor_dir).read())
good_fractions = {}
timeout = {}
for c in CI.campaigns:
if 'force-complete' in CI.campaigns[c]:
good_fractions[c] = CI.campaigns[c]['force-complete']
if 'force-timeout' in CI.campaigns[c]:
timeout[c] = CI.campaigns[c]['force-timeout']
long_lasting = {}
overrides = getForceCompletes()
if use_mcm:
## add all workflow that mcm wants to get force completed
mcm_force = mcm.get('/restapi/requests/forcecomplete')
## assuming this will be a list of actual prepids
overrides['mcm'] = mcm_force
print "can force complete on"
print json.dumps( good_fractions ,indent=2)
print json.dumps( overrides, indent=2)
max_force = UC.get("max_force_complete")
#wfs_no_location_in_GQ = set()
#block_locations = defaultdict(lambda : defaultdict(list))
#wfs_no_location_in_GQ = defaultdict(list)
set_force_complete = set()
for wfo in wfs:
if specific and not specific in wfo.name: continue
print "looking at",wfo.name
## get all of the same
wfi = workflowInfo(url, wfo.name)
pids = wfi.getPrepIDs()
skip=False
if not any([c in wfo.name for c in good_fractions]): skip=True
for user,spec in overrides.items():
if wfi.request['RequestStatus']!='force-complete':
if any(s in wfo.name for s in spec) or (wfo.name in spec) or any(pid in spec for pid in pids) or any(s in pids for s in spec):
sendEmail('force-complete requested','%s is asking for %s to be force complete'%(user,wfo.name))
wfi = workflowInfo(url, wfo.name)
forceComplete(url , wfi )
skip=True
wfi.notifyRequestor("The workflow %s was force completed by request of %s"%(wfo.name,user), do_batch=False)
wfi.sendLog('completor','%s is asking for %s to be force complete'%(user,wfo.name))
break
if wfo.status.startswith('assistance'): skip = True
if skip:
continue
priority = wfi.request['RequestPriority']
if not 'Campaign' in wfi.request: continue
if not wfi.request['RequestStatus'] in ['acquired','running-open','running-closed']: continue
c = wfi.request['Campaign']
if not c in good_fractions: continue
good_fraction = good_fractions[c]
ignore_fraction = 2.
lumi_expected = None
event_expected = None
if not 'TotalInputEvents' in wfi.request:
if 'RequestNumEvents' in wfi.request:
event_expected = wfi.request['RequestNumEvents']
else:
print "truncated, cannot do anything"
continue
else:
lumi_expected = wfi.request['TotalInputLumis']
event_expected = wfi.request['TotalInputEvents']
now = time.mktime(time.gmtime()) / (60*60*24.)
running_log = filter(lambda change : change["Status"] in ["running-open","running-closed"],wfi.request['RequestTransition'])
if not running_log:
print "\tHas no running log"
# cannot figure out when the thing started running
continue
then = running_log[-1]['UpdateTime'] / (60.*60.*24.)
delay = now - then ## in days
(w,d) = divmod(delay, 7 )
print "\t"*int(w)+"Running since",delay,"[days] priority=",priority
monitor_delay = 7
allowed_delay = 14
if c in timeout:
allowed_delay = timeout[c]
monitor_delay = min(monitor_delay, allowed_delay)
### just skip if too early
if delay <= monitor_delay: continue
long_lasting[wfo.name] = { "delay" : delay }
percent_completions = {}
for output in wfi.request['OutputDatasets']:
if "/DQM" in output: continue ## that does not count
if not output in completions: completions[output] = { 'injected' : None, 'checkpoints' : [], 'workflow' : wfo.name}
## get completion fraction
event_count,lumi_count = getDatasetEventsAndLumis(dataset=output)
lumi_completion=0.
event_completion=0.
if lumi_expected:
lumi_completion = lumi_count / float( lumi_expected )
if event_expected:
event_completion = event_count / float( event_expected )
#take the less optimistic
percent_completions[output] = min( lumi_completion, event_completion )
completions[output]['checkpoints'].append( (now, event_completion ) )
if all([percent_completions[out] >= good_fraction for out in percent_completions]):
wfi.sendLog('completor', "all is above %s \n%s"%( good_fraction,
json.dumps( percent_completions, indent=2 )
))
else:
long_lasting[wfo.name].update({
'completion': sum(percent_completions.values()) / len(percent_completions),
'completions' : percent_completions
})
## do something about the agents this workflow is in
long_lasting[wfo.name]['agents'] = wfi.getAgents()
wfi.sendLog('completor', "%s not over bound %s\n%s"%(percent_completions.values(), good_fraction,
json.dumps( long_lasting[wfo.name]['agents'], indent=2) ))
continue
if all([percent_completions[out] >= ignore_fraction for out in percent_completions]):
print "all is done, just wait a bit"
continue
for output in percent_completions:
completions[output]['injected'] = then
#further check on delays
cpuh = wfi.getComputingTime(unit='d')
ran_at = wfi.request['SiteWhitelist']
wfi.sendLog('completor',"Required %s, time spend %s"%( cpuh, delay))
##### WILL FORCE COMPLETE BELOW
# only really force complete after n days
if delay <= allowed_delay: continue
## find ACDCs that might be running
if max_force>0:
forceComplete(url, wfi )
set_force_complete.add( wfo.name )
print "going for force-complete of",wfo.name
wfi.sendLog('completor','going for force completing')
wfi.notifyRequestor("The workflow %s was force completed for running too long"% wfo.name)
max_force -=1
else:
wfi.sendLog('completor',"too many completion this round, cannot force complete")
## do it once only for testing
#break
if set_force_complete:
sendLog('completor','The followings were set force-complete \n%s'%('\n'.join(set_force_complete)))
#sendEmail('set force-complete', 'The followings were set force-complete \n%s'%('\n'.join(set_force_complete)))
open('%s/completions.json'%monitor_dir,'w').write( json.dumps( completions , indent=2))
text="These have been running for long"
open('%s/longlasting.json'%monitor_dir,'w').write( json.dumps( long_lasting, indent=2 ))
for wf,info in sorted(long_lasting.items(), key=lambda tp:tp[1]['delay'], reverse=True):
delay = info['delay']
text += "\n %s : %s days"% (wf, delay)
if 'completion' in info:
text += " %d%%"%( info['completion']*100 )
#if wfs_no_location_in_GQ:
# sendEmail('workflow with no location in GQ',"there won't be able to run anytime soon\n%s"%( '\n'.join(wfs_no_location_in_GQ)))
#sendEmail("long lasting workflow",text)
## you can check the log
print text
def completor(url, specific):
use_mcm = True
up = componentInfo(mcm=use_mcm, soft=['mcm'])
if not up.check(): return
use_mcm = up.status['mcm']
if use_mcm:
mcm = McMClient(dev=False)
CI = campaignInfo()
SI = siteInfo()
UC = unifiedConfiguration()
wfs = []
wfs.extend( session.query(Workflow).filter(Workflow.status == 'away').all() )
wfs.extend( session.query(Workflow).filter(Workflow.status.startswith('assistance')).all() )
## just take it in random order so that not always the same is seen
random.shuffle( wfs )
max_per_round = UC.get('max_per_round').get('completor',None)
if max_per_round and not specific: wfs = wfs[:max_per_round]
all_stuck = set()
## take into account what stagor was saying
all_stuck.update( json.loads( open('%s/stuck_transfers.json'%monitor_dir).read() ))
## take into account the block that needed to be repositioned recently
all_stuck.update( [b.split('#')[0] for b in json.loads( open('%s/missing_blocks.json'%monitor_dir).read()) ] )
## take into account all stuck block and dataset from transfer team
all_stuck.update( getAllStuckDataset())
## by workflow a list of fraction / timestamps
completions = json.loads( open('%s/completions.json'%monitor_dir).read())
good_fractions = {}
truncate_fractions = {}
timeout = {}
for c in CI.campaigns:
if 'force-complete' in CI.campaigns[c]:
good_fractions[c] = CI.campaigns[c]['force-complete']
if 'truncate-complete' in CI.campaigns[c]:
truncate_fractions[c] = CI.campaigns[c]['truncate-complete']
if 'force-timeout' in CI.campaigns[c]:
timeout[c] = CI.campaigns[c]['force-timeout']
long_lasting = {}
overrides = getForceCompletes()
if use_mcm:
## add all workflow that mcm wants to get force completed
mcm_force = mcm.get('/restapi/requests/forcecomplete')
## assuming this will be a list of actual prepids
overrides['mcm'] = mcm_force
print "can force complete on"
print json.dumps( good_fractions ,indent=2)
print "can truncate complete on"
print json.dumps( truncate_fractions ,indent=2)
print "can overide on"
print json.dumps( overrides, indent=2)
max_force = UC.get("max_force_complete")
#wfs_no_location_in_GQ = set()
#block_locations = defaultdict(lambda : defaultdict(list))
#wfs_no_location_in_GQ = defaultdict(list)
set_force_complete = set()
for wfo in wfs:
if specific and not specific in wfo.name: continue
print "looking at",wfo.name
## get all of the same
wfi = workflowInfo(url, wfo.name)
pids = wfi.getPrepIDs()
skip=False
if not any([c in wfo.name for c in good_fractions]) and not any([c in wfo.name for c in truncate_fractions]): skip=True
for user,spec in overrides.items():
if wfi.request['RequestStatus']!='force-complete':
if any(s in wfo.name for s in spec) or (wfo.name in spec) or any(pid in spec for pid in pids) or any(s in pids for s in spec):
wfi = workflowInfo(url, wfo.name)
forceComplete(url , wfi )
skip=True
wfi.notifyRequestor("The workflow %s was force completed by request of %s"%(wfo.name,user), do_batch=False)
wfi.sendLog('completor','%s is asking for %s to be force complete'%(user,wfo.name))
break
if wfo.status.startswith('assistance'): skip = True
if skip:
continue
priority = wfi.request['RequestPriority']
if not 'Campaign' in wfi.request: continue
if not wfi.request['RequestStatus'] in ['acquired','running-open','running-closed']: continue
if wfi.request['RequestStatus'] in ['running-open','running-closed'] and (random.random() < 0.01):
try:
parse_one(url, wfo.name )
except Exception as e:
print "failed error parsing"
print str(e)
c = wfi.request['Campaign']
#if not c in good_fractions: continue
good_fraction = good_fractions.get(c,1000.)
truncate_fraction = truncate_fractions.get(c,1000.)
print "force at",good_fraction,"truncate at",truncate_fraction
ignore_fraction = 2.
lumi_expected = None
event_expected = None
if not 'TotalInputEvents' in wfi.request:
if 'RequestNumEvents' in wfi.request:
event_expected = wfi.request['RequestNumEvents']
else:
print "truncated, cannot do anything"
continue
else:
lumi_expected = wfi.request['TotalInputLumis']
event_expected = wfi.request['TotalInputEvents']
now = time.mktime(time.gmtime()) / (60*60*24.)
running_log = filter(lambda change : change["Status"] in ["running-open","running-closed"],wfi.request['RequestTransition'])
if not running_log:
print "\tHas no running log"
# cannot figure out when the thing started running
continue
then = running_log[-1]['UpdateTime'] / (60.*60.*24.)
delay = now - then ## in days
## this is supposed to be the very initial request date, inherited from clones
try:
injected_on =time.mktime(time.strptime("-".join(map(lambda n : "%02d"%n, wfi.request['RequestDate'])), "%Y-%m-%d-%H-%M-%S")) / (60.*60.*24.)
injection_delay = now - injected_on
except:
injection_delay = None
(w,d) = divmod(delay, 7 )
print "\t"*int(w)+"Running since",delay,"[days] priority=",priority
if injection_delay!=None and injection_delay > 50 and False:
tail_cutting_priority = 200000
## bump up to 200k
print "Injected since",injection_delay,"[days] priority=",priority
if wfi.request['RequestPriority'] != tail_cutting_priority:
wfi.sendLog('completor','bumping priority to %d for being injected since'% tail_cutting_priority)
reqMgrClient.changePriorityWorkflow(url, wfo.name, tail_cutting_priority)
_,prim,_,_ = wfi.getIO()
is_stuck = all_stuck & prim
if is_stuck: wfi.sendLog('completor','%s is stuck'%','.join(is_stuck))
monitor_delay = 7
allowed_delay = 14
if c in timeout:
allowed_delay = timeout[c]
monitor_delay = min(monitor_delay, allowed_delay)
### just skip if too early
if delay <= monitor_delay: continue
long_lasting[wfo.name] = { "delay" : delay,
"injection_delay" : injection_delay }
percent_completions = {}
for output in wfi.request['OutputDatasets']:
if "/DQM" in output: continue ## that does not count
if not output in completions: completions[output] = { 'injected' : None, 'checkpoints' : [], 'workflow' : wfo.name}
## get completion fraction
event_count,lumi_count = getDatasetEventsAndLumis(dataset=output)
lumi_completion=0.
event_completion=0.
if lumi_expected:
lumi_completion = lumi_count / float( lumi_expected )
if event_expected:
event_completion = event_count / float( event_expected )
#take the less optimistic
## taking the max or not : to force-complete on over-pro
percent_completions[output] = min( lumi_completion, event_completion )
completions[output]['checkpoints'].append( (now, event_completion ) )
if all([percent_completions[out] >= good_fraction for out in percent_completions]):
wfi.sendLog('completor', "all is above %s \n%s"%( good_fraction,
json.dumps( percent_completions, indent=2 )
))
elif is_stuck and (all([percent_completions[out] >= truncate_fraction for out in percent_completions])):
wfi.sendLog('completor', "all is above %s truncation level, and the input is stuck\n%s"%( truncate_fraction,
json.dumps( percent_completions, indent=2 ) ) )
sendEmail('workflow for truncation','check on %s'%wfo.name)
##continue ## because you do not want to get this one right now
else:
long_lasting[wfo.name].update({
'completion': sum(percent_completions.values()) / len(percent_completions),
'completions' : percent_completions
})
## do something about the agents this workflow is in
long_lasting[wfo.name]['agents'] = wfi.getAgents()
wfi.sendLog('completor', "%s not over bound %.3f (complete) %.3f truncate \n%s"%(percent_completions.values(),
good_fraction, truncate_fraction,
json.dumps( long_lasting[wfo.name]['agents'], indent=2) ))
continue
if all([percent_completions[out] >= ignore_fraction for out in percent_completions]):
print "all is done, just wait a bit"
continue
for output in percent_completions:
completions[output]['injected'] = then
#further check on delays
cpuh = wfi.getComputingTime(unit='d')
ran_at = wfi.request['SiteWhitelist']
wfi.sendLog('completor',"Required %s, time spend %s"%( cpuh, delay))
##### WILL FORCE COMPLETE BELOW
# only really force complete after n days
if delay <= allowed_delay: continue
## find ACDCs that might be running
if max_force>0:
forceComplete(url, wfi )
set_force_complete.add( wfo.name )
print "going for force-complete of",wfo.name
wfi.sendLog('completor','going for force completing')
wfi.notifyRequestor("The workflow %s was force completed for running too long"% wfo.name)
max_force -=1
else:
wfi.sendLog('completor',"too many completion this round, cannot force complete")
## do it once only for testing
#break
if delay >= 40:
print wfo.name,"has been running for",dealy,"days"
## bumping up the priority?
if set_force_complete:
sendLog('completor','The followings were set force-complete \n%s'%('\n'.join(set_force_complete)))
open('%s/completions.json'%monitor_dir,'w').write( json.dumps( completions , indent=2))
text="These have been running for long"
open('%s/longlasting.json'%monitor_dir,'w').write( json.dumps( long_lasting, indent=2 ))
for wf,info in sorted(long_lasting.items(), key=lambda tp:tp[1]['delay'], reverse=True):
delay = info['delay']
text += "\n %s : %s days"% (wf, delay)
if 'completion' in info:
text += " %d%%"%( info['completion']*100 )
print text
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
#!/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 closor(url, specific=None):
CI = campaignInfo()
## manually closed-out workflows should get to close with checkor
for wfo in session.query(Workflow).filter(Workflow.status=='close').all():
if specific and not specific in wfo.name: continue
## what is the expected #lumis
wl = getWorkLoad(url, wfo.name)
wfo.wm_status = wl['RequestStatus']
if wl['RequestStatus'] in ['announced','normal-archived']:
## manually announced ??
wfo.status = 'done'
wfo.wm_status = wl['RequestStatus']
print wfo.name,"is done already",wfo.wm_status
session.commit()
if not 'TotalInputLumis' in wl:
print wfo.name,"has not been assigned yet"
continue
expected_lumis = wl['TotalInputLumis']
## what are the outputs
outputs = wl['OutputDatasets']
## check whether the number of lumis is as expected for each
all_OK = []
#print outputs
if len(outputs):
print wfo.name,wl['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()
print "\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)))
## only that status can let me go into announced
if wl['RequestStatus'] in ['closed-out']:
print wfo.name,"to be announced"
results=[]#'dummy']
if not results:
results.append(reqMgrClient.announceWorkflowCascade(url, wfo.name))
for (io,out) in enumerate(outputs):
if all_OK[io]:
results.append(setDatasetStatusDBS3.setStatusDBS3('https://cmsweb.cern.ch/dbs/prod/global/DBSWriter', out, 'VALID' ,''))
tier = out.split('/')[-1]
to_DDM = (wl['RequestType'] == 'ReDigi' and not ('DQM' in tier))
campaign = None
try:
campaign = out.split('/')[2].split('-')[0]
except:
if 'Campaign' in wl and wl['Campaign']:
campaign = wl['Campaign']
if campaign and campaign in CI.campaigns and 'toDDM' in CI.campaigns[campaign] and tier in CI.campaigns[campaign]['toDDM']:
to_DDM = True
## inject to DDM everything from ReDigi
if to_DDM:
print "Sending",out," to DDM"
subprocess.call(['python','assignDatasetToSite.py','--dataset='+out,'--exec'])
else:
print wfo.name,"no stats for announcing",out
results.append(None)
#print results
if all(map(lambda result : result in ['None',None],results)):
wfo.status = 'done'
session.commit()
print wfo.name,"is announced"
else:
print "ERROR with ",wfo.name,"to be announced",json.dumps( results )
else:
print wfo.name,"not good for announcing:",wl['RequestStatus']
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 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 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 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 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):
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 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):
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_parentage_to_announce = UC.get('check_parentage_to_announce')
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:
all_OK[out] = True
#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)
# sites_for_DQMHarvest = UC.get("sites_for_DQMHarvest")
# (OK, requests) = spawn_harvesting(url, wfi, sites_for_DQMHarvest)
# print "Harvesting workflow has been created and assigned to: "
# print sites_for_DQMHarvest
# 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 in UC.get("tiers_to_rucio_relval"):
wfi.sendLog(
'closor',
"Data Tier: %s is blacklisted, so skipping dataset placement for: %s"
% (tier, out))
continue
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))))
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_rucio_nonrelval"):
wfi.sendLog(
'closor',
"Data Tier: %s is blacklisted, so skipping dataset placement for: %s"
% (tier, out))
continue
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
else:
print wfo.name, "no stats for announcing", out
results.append('No Stats')
# adding check for PrentageResolved flag from ReqMgr:
if wfi.request[
'RequestType'] == 'StepChain' and check_parentage_to_announce:
if wfi.request['ParentageResolved']:
results.append(True)
else:
wfi.sendLog(
'closor',
"Delayed announcement of %s due to unresolved Parentage dependencies"
% wfi.request['RequestName'])
results.append('No ParentageResolved')
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 completor(url, specific):
up = componentInfo(mcm=False, soft=['mcm'])
if not up.check(): return
CI = campaignInfo()
SI = siteInfo()
wfs = []
wfs.extend(session.query(Workflow).filter(Workflow.status == 'away').all())
wfs.extend(
session.query(Workflow).filter(
Workflow.status.startswith('assistance')).all())
## just take it in random order so that not always the same is seen
random.shuffle(wfs)
## by workflow a list of fraction / timestamps
completions = json.loads(open('%s/completions.json' % monitor_dir).read())
good_fractions = {}
for c in CI.campaigns:
if 'force-complete' in CI.campaigns[c]:
good_fractions[c] = CI.campaigns[c]['force-complete']
long_lasting = {}
overrides = {}
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
continue
try:
overrides[rider] = json.loads(open(rider_file).read())
except:
print "cannot get force complete list from", rider
sendEmail("malformated force complet file",
"%s is not json readable" % rider_file,
destination=[email])
print "can force complete on"
print json.dumps(good_fractions, indent=2)
print json.dumps(overrides, indent=2)
max_force = 5
wfs_no_location_in_GQ = set()
for wfo in wfs:
if specific and not specific in wfo.name: continue
print "looking at", wfo.name
## get all of the same
wfi = workflowInfo(url, wfo.name)
skip = False
if not any([c in wfo.name for c in good_fractions]): skip = True
for user, spec in overrides.items():
#print spec
if wfo.name in spec and wfi.request[
'RequestStatus'] != 'force-complete':
#skip=False ## do not do it automatically yet
sendEmail(
'force-complete requested',
'%s is asking for %s to be force complete' %
(user, wfo.name))
wfi = workflowInfo(url, wfo.name)
forceComplete(url, wfi)
skip = True
wfi.notifyRequestor(
"The workflow %s was force completed by request of %s" %
(wfo.name, user),
do_batch=False)
wfi.sendLog(
'completor', '%s is asking for %s to be force complete' %
(user, wfo.name))
break
if wfo.status.startswith('assistance'): skip = True
if skip:
continue
priority = wfi.request['RequestPriority']
if not 'Campaign' in wfi.request: continue
if not wfi.request['RequestStatus'] in [
'acquired', 'running-open', 'running-closed'
]:
continue
c = wfi.request['Campaign']
if not c in good_fractions: continue
good_fraction = good_fractions[c]
ignore_fraction = 2.
lumi_expected = None
event_expected = None
if not 'TotalInputEvents' in wfi.request:
if 'RequestNumEvents' in wfi.request:
event_expected = wfi.request['RequestNumEvents']
else:
print "truncated, cannot do anything"
continue
else:
lumi_expected = wfi.request['TotalInputLumis']
event_expected = wfi.request['TotalInputEvents']
now = time.mktime(time.gmtime()) / (60 * 60 * 24.)
running_log = filter(
lambda change: change["Status"
] in ["running-open", "running-closed"],
wfi.request['RequestTransition'])
if not running_log:
print "\tHas no running log"
# cannot figure out when the thing started running
continue
then = running_log[-1]['UpdateTime'] / (60. * 60. * 24.)
delay = now - then ## in days
(w, d) = divmod(delay, 7)
print "\t" * int(
w) + "Running since", delay, "[days] priority=", priority
if delay <= 7: continue
if delay >= 7:
long_lasting[wfo.name] = {"delay": delay}
pass
percent_completions = {}
for output in wfi.request['OutputDatasets']:
if "/DQM" in output: continue ## that does not count
if not output in completions:
completions[output] = {
'injected': None,
'checkpoints': [],
'workflow': wfo.name
}
## get completion fraction
event_count, lumi_count = getDatasetEventsAndLumis(dataset=output)
lumi_completion = 0.
event_completion = 0.
if lumi_expected:
lumi_completion = lumi_count / float(lumi_expected)
if event_expected:
event_completion = event_count / float(event_expected)
#take the less optimistic
percent_completions[output] = min(lumi_completion,
event_completion)
completions[output]['checkpoints'].append((now, event_completion))
if all([
percent_completions[out] >= good_fraction
for out in percent_completions
]):
print "all is above", good_fraction, "for", wfo.name
print json.dumps(percent_completions, indent=2)
else:
print "\t", percent_completions.values(
), "not over bound", good_fraction
long_lasting[wfo.name].update({
'completion':
sum(percent_completions.values()) / len(percent_completions),
'completions':
percent_completions
})
#print json.dumps( percent_completions, indent=2 )
## do something about the agents this workflow is in
long_lasting[wfo.name]['agents'] = wfi.getAgents()
print json.dumps(long_lasting[wfo.name]['agents'], indent=2)
## pick up on possible issue with global queue data location
locs = wfi.getGQLocations()
for b, loc in locs.items():
if not loc:
print b, "has no location for GQ in", wfi.request[
'RequestName']
## this is severe !
wfs_no_location_in_GQ.add(wfo.name)
## check location and site white list
can_run = set([SI.SE_to_CE(se) for se in loc]) & set(
wfi.request['SiteWhitelist'])
if loc and not can_run:
print b, "is missing site to run within the whitelist"
wfs_no_location_in_GQ.add(wfo.name)
can_run = can_run & set(SI.sites_ready)
if loc and not can_run:
print b, "is missing available site to run"
wfs_no_location_in_GQ.add(wfo.name)
continue
if all([
percent_completions[out] >= ignore_fraction
for out in percent_completions
]):
print "all is done, just wait a bit"
continue
for output in percent_completions:
completions[output]['injected'] = then
#further check on delays
cpuh = wfi.getComputingTime(unit='d')
ran_at = wfi.request['SiteWhitelist']
print "Required:", cpuh,
print "Time spend:", delay
# only really force complete after n days
if delay <= 14: continue
print "going for force-complete of", wfo.name
## find ACDCs that might be running
if max_force > 0:
forceComplete(url, wfi)
max_force -= 1
else:
print "too many completion this round"
## do it once only for testing
#break
open('%s/completions.json' % monitor_dir,
'w').write(json.dumps(completions, indent=2))
text = "These have been running for long"
open('%s/longlasting.json' % monitor_dir,
'w').write(json.dumps(long_lasting, indent=2))
for wf, info in sorted(long_lasting.items(),
key=lambda tp: tp[1]['delay'],
reverse=True):
delay = info['delay']
text += "\n %s : %s days" % (wf, delay)
if 'completion' in info:
text += " %d%%" % (info['completion'] * 100)
if wfs_no_location_in_GQ:
sendEmail(
'workflow with no location in GQ',
"there won't be able to run anytime soon\n%s" %
('\n'.join(wfs_no_location_in_GQ)))
#sendEmail("long lasting workflow",text)
## you can check the log
print text