def cachor(spec=None):
if duplicateLock(silent=True):
print "currently running"
return
try:
all_checks = json.loads(open('cached_transfer_statuses.json').read())
except:
all_checks = {}
all_transfers = [transfer for transfer in session.query(Transfer).filter(Transfer.phedexid>0).all()]
random.shuffle( all_transfers )
existing = all_checks.keys() ## strng keys
new = (set([str(transfer.phedexid) for transfer in all_transfers]) - set(existing))
print len(new),"transfers not look out at all, will do those first"
if spec: new = [spec]
for transfer in all_transfers:
if new and str(transfer.phedexid)!=sorted(new)[0]: continue
print "running the check on",transfer.phedexid
new_check = checkTransferStatus(url, transfer.phedexid, nocollapse=True)
if new_check :
all_checks[str(transfer.phedexid)] = new_check
else:
print "withouth an update, we are in deep shit"
sendLog('cachor','Failed transfer status check on %s'% transfer.phedexid, level='critical')
#do only one
break
for pid in sorted(all_checks.keys()):
if not all_checks[pid]:
all_checks.pop(pid)
open('cached_transfer_statuses.json','w').write(json.dumps( all_checks , indent=2))
def run(self):
try:
self.close()
except Exception as e:
import traceback
sendLog('closor','failed on %s due to %s and %s'%( self.wfo.name, str(e), traceback.format_exc()), level='critical')
self.failed = True
def run(self):
try:
self.close()
except Exception as e:
import traceback
sendLog('closor','failed on %s due to %s and %s'%( self.wfo.name, str(e), traceback.format_exc()), level='critical')
self.failed = True
def cachor(spec=None):
if duplicateLock(silent=True):
print "currently running"
return
try:
all_checks = json.loads(open('%s/cached_transfer_statuses.json'%base_eos_dir).read())
except:
all_checks = {}
## pop all that are now in inactive
for phedexid in all_checks.keys():
transfers = session.query(TransferImp).filter(TransferImp.phedexid==int(phedexid)).filter(TransferImp.active==True).all()
if not transfers:
print phedexid,"does not look relevant to be in cache anymore. poping"
print all_checks.pop( phedexid )
all_transfers = set()
for imp in session.query(TransferImp).filter(TransferImp.active==True).all():
all_transfers.add( imp.phedexid )
all_transfers = list(all_transfers)
random.shuffle( all_transfers )
existing = map(int,all_checks.keys()) ## strng keys
new = (set(all_transfers) - set(existing))
print len(new),"transfers not look out at all, will do those first",new
if spec: new = [int(spec)]
new = list(new)
random.shuffle( new )
#for transfer in all_transfers:
for phedexid in all_transfers:
#print phedexid
if new and phedexid!=new[0]: continue
print "running the check on",phedexid
new_check = checkTransferStatus(url, phedexid, nocollapse=True)
if new_check :
print json.dumps( new_check ,indent=2)
all_checks[str(phedexid)] = new_check
else:
print "withouth an update, we are in some trouble."
sendLog('cachor','Failed transfer status check on %s'% phedexid, level='critical')
#do only one
break
for pid in sorted(all_checks.keys()):
if not all_checks[pid]:
print "Removing empty report for",pid
all_checks.pop(pid)
open('%s/cached_transfer_statuses.json'%base_eos_dir,'w').write(json.dumps( all_checks , indent=2))
def cachor(spec=None):
mlock = moduleLock(silent=True)
if mlock():
print "currently running"
return
TS = transferStatuses()
print sorted(TS.all()), "cached transfers"
## pop all that are now in inactive
for phedexid in TS.all():
transfers = session.query(TransferImp).filter(
TransferImp.phedexid == int(phedexid)).filter(
TransferImp.active == True).all()
if not transfers:
print phedexid, "does not look relevant to be in cache anymore. poping"
TS.pop(phedexid)
all_transfers = set()
for imp in session.query(TransferImp).filter(
TransferImp.active == True).all():
all_transfers.add(imp.phedexid)
all_transfers = list(all_transfers)
random.shuffle(all_transfers)
existing = map(int, TS.all())
new = (set(all_transfers) - set(existing))
print len(new), "transfers not look out at all, will do those first", new
if spec: new = [int(spec)]
new = list(new)
random.shuffle(new)
#for transfer in all_transfers:
for phedexid in all_transfers:
#print phedexid
if new and phedexid != new[0]: continue
print "running the check on", phedexid
new_check = checkTransferStatus(url, phedexid, nocollapse=True)
if new_check:
print json.dumps(new_check, indent=2)
TS.add(phedexid, new_check)
else:
print "withouth an update, we are in some trouble."
sendLog('cachor',
'Failed transfer status check on %s' % phedexid,
level='critical')
#do only one
break
def cachor(spec=None):
if duplicateLock(silent=True):
print "currently running"
return
try:
all_checks = json.loads(open('cached_transfer_statuses.json').read())
except:
all_checks = {}
## pop all that are now in inactive
for phedexid in all_checks.keys():
transfers = session.query(TransferImp).filter(TransferImp.phedexid==int(phedexid)).filter(TransferImp.active==True).all()
if not transfers:
print phedexid,"does not look relevant to be in cache anymore. poping"
print all_checks.pop( phedexid )
#all_transfers = [transfer for transfer in session.query(Transfer).filter(Transfer.phedexid>0).all()]
all_transfers = list(set([imp.phedexid for imp in session.query(TransferImp).filter(TransferImp.active==True).all()]))
random.shuffle( all_transfers )
existing = map(int,all_checks.keys()) ## strng keys
new = (set(all_transfers) - set(existing))
print len(new),"transfers not look out at all, will do those first",new
if spec: new = [int(spec)]
#for transfer in all_transfers:
for phedexid in all_transfers:
#print phedexid
if new and phedexid!=sorted(new)[0]: continue
print "running the check on",phedexid
new_check = checkTransferStatus(url, phedexid, nocollapse=True)
if new_check :
print json.dumps( new_check ,indent=2)
all_checks[str(phedexid)] = new_check
else:
print "withouth an update, we are in some trouble."
sendLog('cachor','Failed transfer status check on %s'% phedexid, level='critical')
#do only one
break
for pid in sorted(all_checks.keys()):
if not all_checks[pid]:
print "Removing empty report for",pid
all_checks.pop(pid)
open('cached_transfer_statuses.json','w').write(json.dumps( all_checks , indent=2))
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
pileup_location = None
for secondary in wqe['PileupData']:
if pileup_location == None:
pileup_location = wqe['PileupData'][secondary]
else:
pileup_location = list(set(pileup_location) & set(wqe['PileupData'][secondary]))
#print pileup_location,"secondary"
## wqe site whitelist in terms of SE
#swl = [si.CE_to_SE(s) for s in wl]
swl = si.CE_to_SEs(wl)
print "WQE SE whitelist",sorted(swl)
if not swl:
sendLog('GQ',"There is no site at which the workflow %s can run Was provided with %s"%(wf['RequestName'], ','.join(wqe['SiteWhitelist'])), level='critical')
wfi.sendLog('GQ',"There is not site at which the workflow can run. Was provided with %s"%( ','.join(wqe['SiteWhitelist'])))
continue
not_processable = set()
wqe_ce = None
for b in wqe['Inputs']:
## what the global queue thinks about the block location
wqe_se = [si.CE_to_SE(s) for s in wqe['Inputs'][b]]
if not '#' in b:
print "acdc doc input",b
## this would be an ACDC document input
#retrieve it and check data file location ?
original = workflowInfo(url, wfi.request['OriginalRequestName'])
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):
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 equalizor(url , specific = None, options=None):
up = componentInfo(mcm=False, soft=['mcm'])
if not up.check(): return
if not specific:
workflows = getWorkflows(url, status='running-closed', details=True)
workflows.extend(getWorkflows(url, status='running-open', details=True))
## start from scratch
modifications = defaultdict(dict)
## define regionality site => fallback allowed. feed on an ssb metric ??
mapping = defaultdict(list)
reversed_mapping = defaultdict(list)
regions = defaultdict(list)
SI = siteInfo()
CI = campaignInfo()
UC = unifiedConfiguration()
for site in SI.sites_ready:
region = site.split('_')[1]
if not region in ['US','DE','IT']: continue
regions[region] = [region]
def site_in_depletion(s):
return True
if s in SI.sites_pressure:
(m, r, pressure) = SI.sites_pressure[s]
if float(m) < float(r):
print s,m,r,"lacking pressure"
return True
else:
print s,m,r,"pressure"
pass
return False
for site in SI.sites_ready:
region = site.split('_')[1]
## fallback to the region, to site with on-going low pressure
mapping[site] = [fb for fb in SI.sites_ready if any([('_%s_'%(reg) in fb and fb!=site and site_in_depletion(fb))for reg in regions[region]]) ]
use_T0 = ('T0_CH_CERN' in UC.get("site_for_overflow"))
if options.t0: use_T0 = True
#if options.augment : use_T0 = True
use_HLT = ('T2_CH_CERN_HLT' in UC.get("site_for_overflow"))
if options.hlt: use_HLT = True
#if options.augment : use_HLT=True
if use_HLT:
mapping['T2_CH_CERN'].append('T2_CH_CERN_HLT')
if use_T0:
mapping['T2_CH_CERN'].append('T0_CH_CERN')
#mapping['T1_FR_CCIN2P3'].append('T0_CH_CERN')
#mapping['T2_IT_Legnaro'].append('T1_IT_CNAF')
for reg in ['IT','DE','UK']:
mapping['T2_CH_CERN'].extend([fb for fb in SI.sites_ready if '_%s_'%reg in fb])
## make them appear as OK to use
force_sites = []
## overflow CERN to underutilized T1s
upcoming = json.loads( open('%s/GQ.json'%monitor_dir).read())
for possible in SI.sites_T1s:
if not possible in upcoming:
mapping['T2_CH_CERN'].append(possible)
## remove add-hoc sites from overflow mapping
prevent_sites = ['T2_US_Purdue']
for prevent in prevent_sites:
if prevent in mapping: mapping.pop( prevent )
for src in mapping:
for prevent in prevent_sites:
if prevent in mapping[src]:
mapping[src].remove( prevent )
## create the reverse mapping for the condor module
for site,fallbacks in mapping.items():
for fb in fallbacks:
reversed_mapping[fb].append(site)
## this is the fallback mapping
print "Direct mapping : site => overflow"
print json.dumps( mapping, indent=2)
print "Reverse mapping : dest <= from origin"
print json.dumps( reversed_mapping, indent=2)
altered_tasks = set()
def running_idle( wfi , task_name):
gmon = wfi.getGlideMon()
#print gmon
if not gmon: return (0,0)
if not task_name in gmon: return (0,0)
return (gmon[task_name]['Running'], gmon[task_name]['Idle'])
def needs_action( wfi, task, min_idled = 100, pressure = 0.2):
task_name = task.pathName.split('/')[-1]
running, idled = running_idle( wfi, task_name)
go = True
if not idled and not running :
go = False
if idled < 100:
go = False
if (not running and idled) or (running and (idled / float(running) > pressure)):
go = True
else:
go = False
return go, task_name, running, idled
def getPerf( task ):
task = task.split('/')[1]+'/'+task.split('/')[-1]
try:
u = 'http://cms-gwmsmon.cern.ch/prodview/json/history/memoryusage720/%s'%task
print u
perf_data = json.loads(os.popen('curl -s --retry 5 %s'%u).read())
except Exception as e:
print str(e)
return (None,None)
buckets = perf_data['aggregations']["2"]['buckets']
s_m = sum( bucket['key']*bucket['doc_count'] for bucket in buckets)
w_m = sum( bucket['doc_count'] for bucket in buckets)
m_m = max( bucket['key'] for bucket in buckets) if buckets else None
b_m = None
if w_m > 100:
b_m = m_m
try:
perf_data = json.loads(os.popen('curl -s --retry 5 http://cms-gwmsmon.cern.ch/prodview/json/history/runtime720/%s'%task).read())
except Exception as e:
print str(e)
return (b_m,None)
buckets = perf_data['aggregations']["2"]['buckets']
s_t = sum( bucket['key']*bucket['doc_count'] for bucket in buckets)
w_t = sum( bucket['doc_count'] for bucket in buckets)
m_t = max( bucket['key'] for bucket in buckets) if buckets else None
b_t = None
if w_t > 100:
b_t = m_t
return (b_m,b_t)
def getcampaign( task ):
taskname = task.pathName.split('/')[-1]
if hasattr( task, 'prepID'):
return task.prepID.split('-')[1]
elif taskname.count('-')>=1:
return taskname.split('-')[1]
else:
return None
def close( interface ):
open('%s/equalizor.json.new'%monitor_dir,'w').write( json.dumps( interface, indent=2))
os.system('mv %s/equalizor.json.new %s/equalizor.json'%(monitor_dir,monitor_dir))
os.system('cp %s/equalizor.json %s/logs/equalizor/equalizor.%s.json'%(monitor_dir,monitor_dir,time.mktime(time.gmtime())))
interface = {
'reversed_mapping' : reversed_mapping,
'modifications' : {}
}
if options.augment or options.remove:
interface['modifications'] = json.loads( open('%s/equalizor.json'%monitor_dir).read())['modifications']
if options.remove:
if specific in interface['modifications']:
print "poping",specific
interface['modifications'].pop(specific)
close( interface )
return
PU_locations = {}
PU_overflow = {}
LHE_overflow = {}
tune_performance = []
pending_HLT = 0
max_HLT = 60000
pending_T0 = 0
max_T0 = 60000
try:
gmon = json.loads(os.popen('curl -s http://cms-gwmsmon.cern.ch/prodview/json/T2_CH_CERN_HLT').read())
pending_HLT += gmon["Running"]
pending_HLT += gmon["MatchingIdle"]
except:
pass
stay_within_site_whitelist = False
specific_task=None
if specific and ":" in specific:
specific,specific_task = specific.split(':')
if specific:
wfs = session.query(Workflow).filter(Workflow.name.contains(specific)).all()
else:
wfs = session.query(Workflow).filter(Workflow.status == 'away').all()
performance = {}
no_routing = [
]
random.shuffle( wfs )
for wfo in wfs:
if wfo.name in no_routing and not options.augment:
continue
if specific and not specific in wfo.name:
continue
if specific:
wfi = workflowInfo(url, wfo.name)
else:
cached = filter(lambda d : d['RequestName']==wfo.name, workflows)
if not cached : continue
wfi = workflowInfo(url, wfo.name, request = cached[0])
## only running should get re-routed
if not wfi.request['RequestStatus'] in ['running-open','running-closed'] and not specific: continue
tasks_and_campaigns = []
for task in wfi.getWorkTasks():
tasks_and_campaigns.append( (task, getcampaign(task) ) )
_,_,_,sec = wfi.getIO()
## check needs override
needs_overide = False
if not needs_overide and options.augment: needs_overide=True
def overide_from_agent( wfi, needs_overide):
bad_agents = []#'http://cmssrv219.fnal.gov:5984']
if not bad_agents: return needs_overide
if needs_overide: return True
agents = wfi.getAgents()
wqss = ['Running','Acquired']
if any([agent in agents.get(wqs,{}).keys() for wqs,agent in itertools.product( wqss, bad_agents)]):
print "overriding the need for bad agent"
needs_overide = True
return needs_overide
## now parse this for action
for i_task,(task,campaign) in enumerate(tasks_and_campaigns):
if options.augment:
print task.pathName
print campaign
tune = CI.get(campaign,'tune',options.tune)
if tune and not campaign in tune_performance:
tune_performance.append( campaign )
overflow = CI.get(campaign,'overflow',{})
if overflow:
if "PU" in overflow and not campaign in PU_overflow:
PU_overflow[campaign] = copy.deepcopy(overflow['PU'])
print "adding",campaign,"to PU overflow rules"
if "LHE" in overflow and not campaign in LHE_overflow:
print "adding",campaign,"to light input overflow rules"
site_list = overflow['LHE']['site_list']
LHE_overflow[campaign] = copy.deepcopy( getattr(SI,site_list) )
### get the task performance, for further massaging.
if campaign in tune_performance or options.tune:
print "performance",task.taskType,task.pathName
if task.taskType in ['Processing','Production']:
set_memory,set_time = getPerf( task.pathName )
#print "Performance %s GB %s min"%( set_memory,set_time)
wfi.sendLog('equalizor','Performance tuning to %s GB %s min'%( set_memory,set_time))
## get values from gmwsmon
# massage the values : 95% percentile
performance[task.pathName] = {}
if set_memory:
performance[task.pathName]['memory']=set_memory
if set_time and False:
performance[task.pathName]['time'] = set_time
### rule to avoid the issue of taskchain secondary jobs being stuck at sites processing the initial step
if campaign in LHE_overflow:
if task.taskType in ['Processing']:
needs, task_name, running, idled = needs_action(wfi, task)
needs_overide = overide_from_agent( wfi, needs_overide)
extend_to = list(set(copy.deepcopy( LHE_overflow[campaign] )))
if stay_within_site_whitelist:
extend_to = list(set(extend_to) & set(wfi.request['SiteWhitelist'])) ## restrict to stupid-site-whitelist
extend_to = list(set(extend_to) & set(SI.sites_ready + force_sites))
if extend_to and needs or needs_overide:
modifications[wfo.name][task.pathName] = { "ReplaceSiteWhitelist" : extend_to ,"Running" : running, "Pending" : idled, "Priority" : wfi.request['RequestPriority']}
wfi.sendLog('equalizor','%s of %s is running %d and pending %d, taking action : ReplaceSiteWhitelist \n %s'%( task_name,
wfo.name,
running,
idled ,
json.dumps( sorted(modifications[wfo.name][task.pathName]['ReplaceSiteWhitelist']))))
altered_tasks.add( task.pathName )
else:
wfi.sendLog('equalizor','%s of %s is running %d and pending %d'%( task_name, wfo.name, running, idled))
### overflow the 76 digi-reco to the site holding the pileup
if campaign in PU_overflow:
force = PU_overflow[campaign]['force'] if 'force' in PU_overflow[campaign] else False
secondary_locations = set(SI.sites_ready + force_sites)
for s in sec:
if not s in PU_locations:
presence = getDatasetPresence( url, s)
#one_secondary_locations = [site for (site,(there,frac)) in presence.items() if there]
one_secondary_locations = [site for (site,(there,frac)) in presence.items() if frac>98.]
PU_locations[s] = one_secondary_locations
print "secondary is at",sorted(PU_locations[s])
secondary_locations = set([SI.SE_to_CE(site) for site in PU_locations[s]]) & secondary_locations
## we should add all sites that hold the secondary input if any
### given that we have the secondary location available, it is not necessary to use the add-hoc list
##secondary_locations = list(set(PU_overflow[campaign]['sites']) & set( SI.sites_ready ))
if any([task.pathName.endswith(finish) for finish in ['_0','StepOneProc','Production']]) :
needs, task_name, running, idled = needs_action(wfi, task)
## removing the ones in the site whitelist already since they encode the primary input location
if stay_within_site_whitelist:
original_site_in_use = set(wfi.request['SiteWhitelist'])
else:
original_site_in_use = set(secondary_locations)
## remove the sites that have already running jobs
gmon = wfi.getGlideMon()
if gmon and task_name in gmon and 'Sites' in gmon[task_name]:
site_in_use = set(gmon[task_name]['Sites'])
print "removing",sorted(site_in_use)
## that determines where you want to run in addition
augment_by = list((set(secondary_locations)- site_in_use) & original_site_in_use)
else:
print "no existing running site"
augment_by = list(original_site_in_use)
needs_overide = overide_from_agent( wfi, needs_overide)
if augment_by and (needs or needs_overide or force) and PU_overflow[campaign]['pending'] < PU_overflow[campaign]['max']:
PU_overflow[campaign]['pending'] += idled
print "raising overflow to",PU_overflow[campaign]['pending'],"for",PU_overflow[campaign]['max']
## the step with an input ought to be the digi part : make this one go anywhere
modifications[wfo.name][task.pathName] = { "AddWhitelist" : augment_by , "Running" : running, "Pending" : idled, "Priority" : wfi.request['RequestPriority']}
altered_tasks.add( task.pathName )
wfi.sendLog('equalizor','%s of %s is running %d and pending %d, taking action : AddWhitelist \n %s'%( task_name, wfo.name,
running, idled,
json.dumps( sorted(augment_by), indent=2 )))
else:
print task_name,"of",wfo.name,"running",running,"and pending",idled
### overflow the skims back to multi-core
if campaign in ['Run2015D','Run2015C_25ns'] and task.taskType =='Skim':
original_swl = wfi.request['SiteWhitelist']
needs, task_name, running, idled = needs_action(wfi, task)
if (needs or needs_overide):
modifications[wfo.name][task.pathName] = { 'AddWhitelist' : original_swl,
"Running" : running, "Pending" : idled, "Priority" : wfi.request['RequestPriority']}
altered_tasks.add( task.pathName )
wfi.sendLog('equalizor','%s of %s is running %d and pending %d, taking action : AddWhitelist \n %s'%( task_name, wfo.name,
running, idled,
json.dumps( sorted(original_swl), indent=2 )))
if options.augment:
print sorted(wfi.request['SiteWhitelist']),i_task,use_HLT
### add the HLT at partner of CERN
if 'T2_CH_CERN' in wfi.request['SiteWhitelist'] and i_task in [0,1] and use_HLT:
needs, task_name, running, idled = needs_action(wfi, task)
if options.augment: needs=True
needs = True
##needs = random.random()<0.40 remove the random, just add up to a limit
if (needs or needs_overide) and pending_HLT < max_HLT:
pending_HLT += idled
if task.pathName in modifications[wfo.name] and 'AddWhitelist' in modifications[wfo.name][task.pathName]:
modifications[wfo.name][task.pathName]["AddWhitelist"].append( "T2_CH_CERN_HLT" )
print "\t",wfo.name,"adding addHLT up to",pending_HLT,"for",max_HLT
print task.pathName
## this Replace does not work at all for HLT
#elif task.pathName in modifications[wfo.name] and 'ReplaceSiteWhitelist' in modifications[wfo.name][task.pathName]:
#modifications[wfo.name][task.pathName]["ReplaceSiteWhitelist"].append( "T2_CH_CERN_HLT" )
#print "\t",wfo.name,"adding replace HLT up to",pending_HLT,"for",max_HLT
else:
modifications[wfo.name][task.pathName] = { "AddWhitelist" : ["T2_CH_CERN_HLT"],
"Priority" : wfi.request['RequestPriority'],
"Running" : running,
"Pending" : idled}
wfi.sendLog('equalizor','adding the HLT in whitelist of %s to %d for %d'%( task.pathName, pending_HLT, max_HLT))
if i_task==0 and not sec and use_T0:
needs, task_name, running, idled = needs_action(wfi, task)
if options.augment: needs=True
#needs = True
good_type = wfi.request['RequestType'] in ['MonteCarlo','MonteCarloFromGEN']
read_lhe = ((not 'LheInputFiles' in wfi.request) or bool(wfi.request['LheInputFiles']))
good_type &= not read_lhe
if not good_type and not options.augment: needs = False
##needs = random.random()<0.40 remove the random, just add up to a limit
if (needs or needs_overide):
pending_T0 += idled
if task.pathName in modifications[wfo.name] and 'AddWhitelist' in modifications[wfo.name][task.pathName]:
if not "T0_CH_CERN" in modifications[wfo.name][task.pathName]["AddWhitelist"]:
modifications[wfo.name][task.pathName]["AddWhitelist"].append( "T0_CH_CERN" )
wfi,sendLog('equalizor','adding the T0 for %s to %d for %d'%( task.pathName, pending_T0, max_T0))
elif task.pathName in modifications[wfo.name] and 'ReplaceSiteWhitelist' in modifications[wfo.name][task.pathName]:
if not "T0_CH_CERN" in modifications[wfo.name][task.pathName]["ReplaceSiteWhitelist"]:
modifications[wfo.name][task.pathName]["ReplaceSiteWhitelist"].append( "T0_CH_CERN" )
wfi,sendLog('equalizor','adding the T0 to replacement for %s to %d for %d'%( task.pathName, pending_T0, max_T0))
else:
modifications[wfo.name][task.pathName] = { "AddWhitelist" : ["T0_CH_CERN"],
"Priority" : wfi.request['RequestPriority'],
"Running" : running,
"Pending" : idled}
wfi,sendLog('equalizor','adding the T0 for %s to %d for %d'%( task.pathName, pending_T0, max_T0))
interface['modifications'].update( modifications )
### manage the number of core and job resizing
interface['cores']={'T2_CH_CERN_HLT': {'min':4,'max':16}, 'default': {'min':1, 'max':4}}
interface['resizes'] = ['RunIISpring16DR80']
### manage the modification of the memory and target time
interface['time'] = defaultdict(list)
interface['memory'] = defaultdict(list)
max_N_mem = 10
max_N_time = 10
## discretize the memory to 10 at most values
mems = set([o['memory'] for t,o in performance.items() if 'memory' in o])
times = set([o['time'] for t,o in performance.items() if 'time' in o])
if len(mems)>max_N_mem:
mem_step = int((max(mems) - min(mems))/ float(max_N_mem))
for t in performance:
if not 'memory' in performance[t]: continue
(m,r) = divmod(performance[t]['memory'], mem_step)
performance[t]['memory'] = (m+1)*mem_step
if len(times)>max_N_time:
time_step = int((max(times) - min(times))/float(max_N_time))
for t in performance:
if not 'time' in performance[t]: continue
(m,r) = divmod(performance[t]['time'], time_step)
performance[t]['time'] = (m+1)*time_step
for t,o in performance.items():
if 'time' in o:
interface['time'][str(o['time'])] .append( t )
if 'memory' in o:
interface['memory'][str(o['memory'])].append( t )
## close and save
close( interface )
def batchor(url):
UC = unifiedConfiguration()
SI = global_SI()
CI = campaignInfo()
BI = batchInfo()
## get all workflows in assignment-approved with SubRequestType = relval
all_wfs = []
for user in UC.get("user_relval"):
all_wfs.extend(
getWorkflows(url,
'assignment-approved',
details=True,
user=user,
rtype='TaskChain'))
wfs = filter(
lambda r: r['SubRequestType'] == 'RelVal'
if 'SubRequestType' in r else False, all_wfs)
## need a special treatment for those
hi_wfs = filter(
lambda r: r['SubRequestType'] == 'HIRelVal'
if 'SubRequestType' in r else False, all_wfs)
by_campaign = defaultdict(set)
by_hi_campaign = defaultdict(set)
for wf in wfs:
print "Relval:", wf['RequestName'], wf['Campaign']
by_campaign[wf['Campaign']].add(wf['PrepID'])
for wf in hi_wfs:
print "HI Relval:", wf['RequestName'], wf['Campaign']
by_hi_campaign[wf['Campaign']].add(wf['PrepID'])
default_setup = {
"go": True,
"parameters": {
"SiteWhitelist": ["T1_US_FNAL"],
"MergedLFNBase": "/store/relval",
"Team": "relval",
"NonCustodialGroup": "RelVal"
},
"custodial_override": "notape",
"phedex_group": "RelVal",
"lumisize": -1,
"fractionpass": 0.0,
"maxcopies": 1
}
default_hi_setup = copy.deepcopy(default_setup)
add_on = {}
relval_routing = UC.get('relval_routing')
def pick_one_site(p):
## modify the parameters on the spot to have only one site
if "parameters" in p and "SiteWhitelist" in p["parameters"] and len(
p["parameters"]["SiteWhitelist"]) > 1:
choose_from = list(
set(p["parameters"]["SiteWhitelist"]) & set(SI.sites_ready))
picked = random.choice(choose_from)
print "picked", picked, "from", choose_from
p["parameters"]["SiteWhitelist"] = [picked]
batches = BI.all()
for campaign in by_campaign:
if campaign in batches: continue
## get a bunch of information
setup = copy.deepcopy(default_setup)
for key in relval_routing:
if key in campaign:
## augment with the routing information
augment_with = relval_routing[key]
print "Modifying the batch configuration because of keyword", key
print "with", augment_with
setup = deep_update(setup, augment_with)
pick_one_site(setup)
add_on[campaign] = setup
sendLog('batchor',
'Adding the relval campaigns %s with parameters \n%s' %
(campaign, json.dumps(setup, indent=2)),
level='critical')
BI.update(campaign, by_campaign[campaign])
for campaign in by_hi_campaign:
if campaign in batches: continue
## get a bunch of information
setup = copy.deepcopy(default_hi_setup)
possible_sites = set(["T1_DE_KIT", "T1_FR_CCIN2P3"])
hi_site = random.choice(list(possible_sites))
setup["parameters"]["SiteWhitelist"] = [hi_site]
pick_one_site(setup)
add_on[campaign] = setup
sendLog('batchor',
'Adding the HI relval campaigns %s with parameters \n%s' %
(campaign, json.dumps(setup, indent=2)),
level='critical')
BI.update(campaign, by_hi_campaign[campaign])
## only new campaigns in announcement
for new_campaign in list(
set(add_on.keys()) - set(CI.all(c_type='relval'))):
## this is new, and can be announced as such
print new_campaign, "is new stuff"
subject = "Request of RelVal samples batch %s" % new_campaign
text = """Dear all,
A new batch of relval workflows was requested.
Batch ID:
%s
Details of the workflows:
https://dmytro.web.cern.ch/dmytro/cmsprodmon/requests.php?campaign=%s
This is an automated message""" % (
new_campaign,
new_campaign,
)
print subject
print text
to = ['*****@*****.**']
sendEmail(subject, text, destination=to)
sendLog('batchor', text, level='critical')
## go through all existing campaigns and remove the ones not in use anymore ?
for old_campaign in CI.all(c_type='relval'):
all_in_batch = getWorkflowByCampaign(url, old_campaign, details=True)
if not all_in_batch: continue
is_batch_done = all(
map(
lambda s: not s in [
'completed', 'force-complete', 'running-open',
'running-closed', 'acquired', 'assigned',
'assignment-approved'
], [wf['RequestStatus'] for wf in all_in_batch]))
## check all statuses
if is_batch_done:
#print "batch",old_campaign,"can be closed or removed if necessary"
#campaigns[old_campaign]['go'] = False ## disable
CI.pop(old_campaign) ## or just drop it all together ?
BI.pop(old_campaign)
print "batch", old_campaign, " configuration was removed"
## merge all anyways
CI.update(add_on, c_type='relval')
def batchor( url ):
UC = unifiedConfiguration()
## get all workflows in assignment-approved with SubRequestType = relval
all_wfs = []
for user in UC.get("user_relval"):
all_wfs = getWorkflows(url, 'assignment-approved', details=True, user=user, rtype='TaskChain')
wfs = filter( lambda r :r['SubRequestType'] == 'RelVal' if 'SubRequestType' in r else False, all_wfs)
## need a special treatment for those
hi_wfs = filter( lambda r :r['SubRequestType'] == 'HIRelVal' if 'SubRequestType' in r else False, all_wfs)
by_campaign = defaultdict(set)
by_hi_campaign = defaultdict(set)
for wf in wfs:
print "Relval:",wf['RequestName'], wf['Campaign']
by_campaign[wf['Campaign']].add( wf['RequestName'] )
for wf in hi_wfs:
print "HI Relval:",wf['RequestName'], wf['Campaign']
by_hi_campaign[wf['Campaign']].add( wf['RequestName'] )
default_setup = {
"go" :True,
"parameters" : {
"SiteWhitelist": [ "T1_US_FNAL" ],
"MergedLFNBase": "/store/relval",
"Team" : "relval",
"NonCustodialGroup" : "RelVal"
},
"custodial" : "T1_US_FNAL_MSS",
"phedex_group" : "RelVal",
"lumisize" : -1,
"fractionpass" : 0.0,
"maxcopies" : 1
}
default_hi_setup = copy.deepcopy( default_setup )
add_on = {}
batches = json.loads( open('batches.json').read() )
for campaign in by_campaign:
## get a bunch of information
setup = copy.deepcopy( default_setup )
add_on[campaign] = setup
sendLog('batchor','Adding the relval campaigns %s with parameters \n%s'%( campaign, json.dumps( setup, indent=2)),level='critical')
if not campaign in batches: batches[campaign] = []
batches[campaign] = list(set(list(copy.deepcopy( by_campaign[campaign] )) + batches[campaign] ))
for campaign in by_hi_campaign:
## get a bunch of information
setup = copy.deepcopy( default_hi_setup )
hi_site = random.choice(["T1_DE_KIT","T1_FR_CCIN2P3"])
setup["parameters"]["SiteWhitelist"]=[ hi_site ]
add_on[campaign] = setup
sendLog('batchor','Adding the HI relval campaigns %s with parameters \n%s'%( campaign, json.dumps( setup, indent=2)),level='critical')
if not campaign in batches: batches[campaign] = []
batches[campaign] = list(set(list(copy.deepcopy( by_hi_campaign[campaign] )) + batches[campaign] ))
open('batches.json','w').write( json.dumps( batches , indent=2 ) )
## open the campaign configuration
campaigns = json.loads( open('campaigns.relval.json').read() )
## protect for overwriting ??
for new_campaign in list(set(add_on.keys())-set(campaigns.keys())):
## this is new, and can be announced as such
print new_campaign,"is new stuff"
workflows = by_campaign[new_campaign]
requester = list(set([wf.split('_')[0] for wf in workflows]))
subject = "Request of RelVal samples batch %s"% new_campaign
text="""Dear all,
A new batch of relval workflows was requested.
Batch ID:
%s
Requestor:
%s
Details of the workflows:
https://dmytro.web.cern.ch/dmytro/cmsprodmon/requests.php?campaign=%s
This is an automated message"""%( new_campaign,
', '.join(requester),
new_campaign,
#'\n'.join( sorted(workflows) )
)
print subject
print text
to = ['*****@*****.**']
sendEmail(subject, text, destination=to)
sendLog('batchor',text, level='critical')
## merge all anyways
campaigns.update( add_on )
## write it out for posterity
open('campaigns.json.updated','w').write(json.dumps( campaigns , indent=2))
## read back
rread = json.loads(open('campaigns.json.updated').read())
os.system('mv campaigns.json.updated campaigns.relval.json')
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 spawn_harvesting(url, wfi , in_full):
#SI = siteInfo()
SI = global_SI()
all_OK = {}
requests = []
outputs = wfi.request['OutputDatasets']
if ('EnableHarvesting' in wfi.request and wfi.request['EnableHarvesting']) or ('DQMConfigCacheID' in wfi.request and wfi.request['DQMConfigCacheID']):
if not 'MergedLFNBase' in wfi.request:
print "f****d up"
sendEmail('screwed up wl cache','%s wl cache is bad'%(wfi.request['RequestName']))
all_OK['fake'] = False
return all_OK,requests
wfi = workflowInfo(url, wfi.request['RequestName'])
dqms = [out for out in outputs if '/DQM' in out]
if not all([in_full[dqm_input] for dqm_input in dqms]):
wfi.sendLog('closor',"will not be able to assign the harvesting: holding up")
for dqm_input in dqms:
all_OK[dqm_input] = False
## raise the subscription to high priority
sites = set(wfi.request['NonCustodialSites'])
for site in sites:
res = updateSubscription(url, site, dqm_input, priority='high')
print "increased priority",res
return all_OK,requests
for dqm_input in dqms:
## handle it properly
harvesting_schema = {
'Requestor': os.getenv('USER'),
'RequestType' : 'DQMHarvest',
'Group' : 'DATAOPS'
}
copy_over = [
'AcquisitionEra',
'ProcessingString',
'DQMUploadUrl',
'CMSSWVersion',
'CouchDBName',
'CouchWorkloadDBName',
'CouchURL',
'DbsUrl',
'inputMode',
'DQMConfigCacheID',
'OpenRunningTimeout',
'ScramArch',
'CMSSWVersion',
'Campaign',
'Memory', #dummy
'SizePerEvent', #dummy
'GlobalTag', #dummy
]
for item in copy_over:
if item in wfi.request:
harvesting_schema[item] = copy.deepcopy(wfi.request[item])
else:
print item,"is not in initial schema"
harvesting_schema['InputDataset'] = dqm_input
harvesting_schema['TimePerEvent'] = 1
harvesting_schema['PrepID'] = 'Harvest-'+wfi.request['PrepID']
if len(wfi.request['RequestString'])>60:
wfi.request['RequestString']= wfi.request['RequestString'][:60]
print "truncating request string",wfi.request['RequestString']
harvesting_schema['RequestString'] = 'HARVEST-'+wfi.request['RequestString']
harvesting_schema['DQMHarvestUnit'] = 'byRun'
harvesting_schema['ConfigCacheUrl'] = harvesting_schema['CouchURL'] ## uhm, how stupid is that ?
harvesting_schema['RequestPriority'] = wfi.request['RequestPriority']*10
harvest_request = reqMgrClient.submitWorkflow(url, harvesting_schema)
if not harvest_request:
print "Error in making harvesting for",wfi.request['RequestName']
print "schema"
print json.dumps( harvesting_schema, indent = 2)
harvest_request = reqMgrClient.submitWorkflow(url, harvesting_schema)
if not harvest_request:
print "Error twice in harvesting for",wfi.request['RequestName']
print "schema"
print json.dumps( harvesting_schema, indent = 2)
if harvest_request:
requests.append( harvest_request )
## should we protect for setting approved ? no, it's notified below, assignment will fail, likely
data = reqMgrClient.setWorkflowApproved(url, harvest_request)
print "created",harvest_request,"for harvesting of",dqm_input
wfi.sendLog('closor',"created %s for harvesting of %s"%( harvest_request, dqm_input))
## assign it directly
team = wfi.request['Teams'][0]
parameters={
'SiteWhitelist' : [SI.SE_to_CE(se) for se in wfi.request['NonCustodialSites']],
'AcquisitionEra' : wfi.acquisitionEra(),
'ProcessingString' : wfi.processingString(),
'MergedLFNBase' : wfi.request['MergedLFNBase'],
'ProcessingVersion' : wfi.request['ProcessingVersion'],
'execute' : True
}
if in_full[dqm_input]:
print "using full copy at",in_full[dqm_input]
parameters['SiteWhitelist'] = [SI.SE_to_CE(se) for se in in_full[dqm_input]]
else:
print "cannot do anything if not having a full copy somewhere"
all_OK[dqm_input]=False
continue
result = reqMgrClient.assignWorkflow(url, harvest_request, team, parameters)
if not result:
#sendEmail('harvesting request created','%s was created at announcement of %s in %s, failed to assign'%(harvest_request, dqm_input, wfi.request['RequestName']), destination=[wfi.request['Requestor']+'@cern.ch'])
wfi.sendLog('closor','%s was created at announcement of %s in %s, failed to assign'%(harvest_request, dqm_input, wfi.request['RequestName']))
sendLog('closor','%s was created at announcement of %s in %s, failed to assign'%(harvest_request, dqm_input, wfi.request['RequestName']), level='critical')
else:
#sendEmail('harvesting request assigned','%s was created at announcement of %s in %s, and assigned'%(harvest_request, dqm_input, wfi.request['RequestName']), destination=[wfi.request['Requestor']+'@cern.ch'])
wfi.sendLog('closor','%s was created at announcement of %s in %s, and assigned'%(harvest_request, dqm_input, wfi.request['RequestName']))
else:
#print "could not make the harvesting for",wfo.name,"not announcing"
wfi.sendLog('closor',"could not make the harvesting request")
sendLog('closor',"could not make the harvesting request for %s"% wfi.request['RequestName'], level='critical')
all_OK[dqm_input]=False
return (all_OK, requests)
def completor(url, specific):
mlock = moduleLock(silent=True)
if mlock(): return
use_mcm = True
up = componentInfo(soft=['mcm', 'wtc', 'jira'])
if not up.check(): return
use_mcm = up.status['mcm']
if use_mcm:
mcm = McMClient(dev=False)
safe_mode = False
CI = campaignInfo()
SI = siteInfo()
UC = unifiedConfiguration()
JC = JIRAClient() if up.status.get('jira', False) else None
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
for itry in range(5):
try:
all_stuck.update(
json.loads(eosRead('%s/stuck_transfers.json' %
monitor_pub_dir)))
break
except:
time.sleep(2)
for itry in range(5):
try:
## take into account the block that needed to be repositioned recently
all_stuck.update([
b.split('#')[0] for b in json.loads(
eosRead('%s/missing_blocks.json' % monitor_dir))
])
break
except:
time.sleep(2)
## take into account all stuck block and dataset from transfer team
all_stuck.update(getAllStuckDataset())
good_fractions = {}
overdoing_fractions = {}
truncate_fractions = {}
timeout = {}
campaign_injection_delay = {}
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']
if 'injection-delay' in CI.campaigns[c]:
campaign_injection_delay[c] = CI.campaigns[c]['injection-delay']
if 'overdoing-complete' in CI.campaigns[c]:
overdoing_fractions[c] = CI.campaigns[c]['overdoing-complete']
long_lasting = {}
WI = wtcInfo()
overrides = WI.getForce()
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")
max_priority = UC.get("max_tail_priority")
injection_delay_threshold = UC.get("injection_delay_threshold")
injection_delay_priority = UC.get("injection_delay_priority")
delay_priority_increase = UC.get("delay_priority_increase")
default_fraction_overdoing = UC.get('default_fraction_overdoing')
set_force_complete = set()
# priority and time above which to fire a JIRA
jira_priority_and_delays = {
110000: 21,
90000: 28,
# 80000 : 60,
#0 : 90
}
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
campaigns = wfi.getCampaigns()
#if not any([c in good_fractions.keys() for c in campaigns]): skip=True
#if not any([c in truncate_fractions.keys() for c in campaigns]): skip=True
for user, spec in overrides.items():
if not spec: continue
spec = filter(None, spec)
if not wfi.request['RequestStatus'] in [
'force-complete', 'completed'
]:
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
## until we can map the output to task ...
output_per_task = wfi.getOutputPerTask(
) ## can use that one, and follow mapping
good_fraction_per_out = {}
good_fraction_nodelay_per_out = {}
truncate_fraction_per_out = {}
#allowed_delay_per_out = {}
for task, outs in output_per_task.items():
task_campaign = wfi.getCampaignPerTask(task)
for out in outs:
good_fraction_per_out[out] = good_fractions.get(
task_campaign, 1000.)
good_fraction_nodelay_per_out[out] = overdoing_fractions.get(
task_campaign, default_fraction_overdoing)
truncate_fraction_per_out[out] = truncate_fractions.get(
task_campaign, 1000.)
#allowed_delay_per_out[out] = timeout.get(task_campaign, 14)
#print "force at", json.dumps( good_fraction_per_out, indent=2)
#print "truncate at",json.dumps( truncate_fraction_per_out, indent=2)
now = time.mktime(time.gmtime()) / (60 * 60 * 24.)
priority_log = filter(lambda change: change['Priority'] == priority,
wfi.request.get('PriorityTransition', []))
if not priority_log:
print "\tHas no priority log"
priority_delay = 0
else:
then = max([change['UpdateTime']
for change in priority_log]) / (60. * 60. * 24.)
priority_delay = now - then ## in days
print "priority was set to", priority, priority_delay, "[days] ago"
running_log = filter(
lambda change: change["Status"
] in ["running-open", "running-closed"],
wfi.request['RequestTransition'])
if not running_log:
print "\tHas no running log"
delay = 0
else:
then = max([change['UpdateTime']
for change in running_log]) / (60. * 60. * 24.)
delay = now - then ## in days
#further check on delays
cpuh = wfi.getComputingTime(unit='d')
wfi.sendLog(
'completor',
"Requires %.1f [CPUd], running since %.1f [day], last priority change to %d done %.1f [day] ago"
% (cpuh, delay, priority, priority_delay))
if priority_delay != 0 and priority_delay < delay:
## regardless when it started running, set the delay to when priority was changed last
delay = priority_delay
## this is supposed to be the very initial request date, inherited from clones
injection_delay = None
original = wfi
if 'OriginalRequestName' in original.request:
## go up the clone chain
original = workflowInfo(url,
original.request['OriginalRequestName'])
injected_log = filter(
lambda change: change["Status"] in ["assignment-approved"],
original.request['RequestTransition'])
if injected_log:
injected_on = injected_log[-1]['UpdateTime'] / (60. * 60. * 24.)
injection_delay = now - injected_on
delay_for_priority_increase = injection_delay
#delay_for_priority_increase = delay
(w, d) = divmod(delay, 7)
print "\t" * int(
w) + "Running since", delay, "[days] priority=", priority
pop_a_jira = False
ping_on_jira = 7 * (24 * 60 * 60) # 7 days
for jp, jd in jira_priority_and_delays.items():
if priority >= jp and delay >= jd: pop_a_jira = True
if pop_a_jira and JC:
j, reopened, just_created = JC.create_or_last(
prepid=wfi.request['PrepID'],
priority=wfi.request['RequestPriority'],
label='Late',
reopen=True)
last_time = JC.last_time(j)
since_last_ping = time.mktime(time.gmtime()) - last_time
if since_last_ping > ping_on_jira or just_created:
j_comment = "Running since %.1f [days] at priority %d" % (
delay, priority)
JC.comment(j.key, j_comment)
if delay_for_priority_increase != None and delay_for_priority_increase > injection_delay_threshold and priority >= injection_delay_priority:
quantized = 5000 ## quantize priority
tail_cutting_priority = wfi.request['InitialPriority'] + int(
(delay_priority_increase *
(delay_for_priority_increase - injection_delay_threshold) / 7)
/ quantized) * quantized
tail_cutting_priority += 101 ## to signal it is from this mechanism
tail_cutting_priority = min(
400000, tail_cutting_priority) ## never go above 400k priority
tail_cutting_priority = max(
tail_cutting_priority,
priority) ## never go below the current value
if priority < tail_cutting_priority:
if max_priority:
sendLog(
'completor',
"%s Injected since %s [days] priority=%s, increasing to %s"
% (wfo.name, delay_for_priority_increase, priority,
tail_cutting_priority),
level='critical')
wfi.sendLog(
'completor',
'bumping priority to %d for being injected since %s' %
(tail_cutting_priority, delay_for_priority_increase))
reqMgrClient.changePriorityWorkflow(
url, wfo.name, tail_cutting_priority)
max_priority -= 1
else:
sendLog(
'completor',
"%s Injected since %s [days] priority=%s, would like to increase to %s"
% (wfo.name, delay_for_priority_increase, priority,
tail_cutting_priority),
level='critical')
wfi.sendLog(
'completor',
'would like to bump priority to %d for being injected since %s'
% (tail_cutting_priority, delay_for_priority_increase))
print "Could be changing the priority to higher value, but too many already were done"
_, 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 = max([timeout.get(c, 14) for c in campaigns])
monitor_delay = min(monitor_delay, allowed_delay)
### just skip if too early, just for the sake of not computing the completion fraction just now.
# maybe this is fast enough that we can do it for all
if delay <= monitor_delay:
print "not enough time has passed yet"
continue
long_lasting[wfo.name] = {
"delay": delay,
"injection_delay": injection_delay
}
percent_completions = wfi.getCompletionFraction(caller='completor')
if not percent_completions:
sendLog('completor',
'%s has no output at all' % wfo.name,
level='critical')
continue
is_over_allowed_delay = (all([
percent_completions[out] >= good_fraction_per_out.get(out, 1000.)
for out in percent_completions
]) and delay >= allowed_delay)
is_over_truncation_delay = (is_stuck and (all([
percent_completions[out] >= truncate_fraction_per_out.get(
out, 1000.) for out in percent_completions
])) and delay >= allowed_delay)
is_over_completion = (all([
percent_completions[out] >= good_fraction_nodelay_per_out.get(
out, 1000.) for out in percent_completions
]))
if is_over_completion:
wfi.sendLog(
'completor', "all is over completed %s\n %s" %
(json.dumps(good_fraction_nodelay_per_out, indent=2),
json.dumps(percent_completions, indent=2)))
elif is_over_allowed_delay:
wfi.sendLog(
'completor', "all is above %s \n%s" %
(json.dumps(good_fraction_per_out, indent=2),
json.dumps(percent_completions, indent=2)))
elif is_over_truncation_delay:
wfi.sendLog(
'completor',
"all is above %s truncation level, and the input is stuck\n%s"
% (json.dumps(truncate_fraction_per_out, indent=2),
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 \ncomplete at %s \n truncate at %s \nRunning %s"
% (json.dumps(percent_completions, indent=2),
json.dumps(good_fraction_per_out, indent=2),
json.dumps(truncate_fraction_per_out, indent=2),
json.dumps(long_lasting[wfo.name]['agents'], indent=2)))
continue
#for output in percent_completions:
# completions[output]['injected'] = then
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
## find ACDCs that might be running
if max_force > 0:
print "going for force-complete of", wfo.name
if not safe_mode:
forceComplete(url, wfi)
set_force_complete.add(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:
sendEmail(
'completor',
'The workflow %s is ready for force complete, but completor is in safe mode'
% wfo.name)
else:
wfi.sendLog(
'completor',
"too many completion this round, cannot force complete")
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 ))
eosFile('%s/longlasting.json' % monitor_dir,
'w').write(json.dumps(long_lasting, indent=2)).close()
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
url = reqmgr_url
wfs = getWorkflows(url, 'assigned', details=True)
now = time.mktime( time.gmtime())
for wf in wfs:
assigned_log = filter(lambda change : change["Status"] in ["assigned"],wf['RequestTransition'])
if assigned_log:
then = assigned_log[-1]['UpdateTime']
since = (now-then)/float(1*24*60*60.)
if since>1.:
print "workflow",wf['RequestName'],"is assigned since",then," that is",since,"days"
sendLog('GQ','The workflow %s has been assigned for %.2f days'%(wf['RequestName'], since), level='critical')
may_have_one=set()
may_have_one.update([wfo.name for wfo in session.query(Workflow).filter(Workflow.status.startswith('away')).all()])
may_have_one.update([wfo.name for wfo in session.query(Workflow).filter(Workflow.status.startswith('assistance')).all()])
wfs = []
wfs.extend( getWorkflows(url, 'running-open', details=True))
wfs.extend( getWorkflows(url, 'running-closed', details=True))
wfs.extend( getWorkflows(url, 'completed', details=True))
may_have_one_too = set()
for wf in wfs:
if wf['RequestName'] in may_have_one:
#print wf['RequestName'],"and familly"
may_have_one_too.update( getWorkflowById(url, wf['PrepID']) )
def assignor(url, specific=None, talk=True, options=None):
if userLock(): return
if duplicateLock(): return
if not componentInfo().check(): return
UC = unifiedConfiguration()
CI = campaignInfo()
SI = siteInfo()
NLI = newLockInfo()
n_assigned = 0
n_stalled = 0
wfos = []
if specific or options.early:
wfos.extend(
session.query(Workflow).filter(
Workflow.status == 'considered').all())
wfos.extend(
session.query(Workflow).filter(Workflow.status == 'staging').all())
if specific:
wfos.extend(
session.query(Workflow).filter(
Workflow.status == 'considered-tried').all())
wfos.extend(
session.query(Workflow).filter(Workflow.status == 'staged').all())
dataset_endpoints = json.loads(
open('%s/dataset_endpoints.json' % monitor_dir).read())
max_per_round = UC.get('max_per_round').get('assignor', None)
max_cpuh_block = UC.get('max_cpuh_block')
random.shuffle(wfos)
for wfo in wfos:
if options.limit and (n_stalled + n_assigned) > options.limit:
break
if max_per_round and (n_stalled + n_assigned) > max_per_round:
break
if specific:
if not any(map(lambda sp: sp in wfo.name, specific.split(','))):
continue
#if not specific in wfo.name: continue
print "\n\n"
wfh = workflowInfo(url, wfo.name)
wfh.sendLog('assignor', "%s to be assigned" % wfo.name)
## the site whitelist takes into account siteInfo, campaignInfo, memory and cores
(lheinput, primary, parent, secondary,
sites_allowed) = wfh.getSiteWhiteList()
## check if by configuration we gave it a GO
no_go = False
if not wfh.go(log=True) and not options.go:
n_stalled += 1
no_go = True
allowed_secondary = set()
for campaign in wfh.getCampaigns():
if campaign in CI.campaigns and 'secondaries' in CI.campaigns[
campaign]:
allowed_secondary.update(CI.campaigns[campaign]['secondaries'])
if (secondary and allowed_secondary) and (
set(secondary) & allowed_secondary != set(secondary)):
wfh.sendLog(
'assignor', '%s is not an allowed secondary' %
(', '.join(set(secondary) - allowed_secondary)))
#sendEmail('secondary not allowed','%s is not an allowed secondary'%( ', '.join(set(secondary)-allowed_secondary)))
sendLog('assignor',
'%s is not an allowed secondary' %
(', '.join(set(secondary) - allowed_secondary)),
level='critical')
if not options.go:
n_stalled += 1
no_go = True
if no_go:
continue
## check on current status for by-passed assignment
if wfh.request['RequestStatus'] != 'assignment-approved':
if not options.test:
wfh.sendLog('assignor',
"setting %s away and skipping" % wfo.name)
## the module picking up from away will do what is necessary of it
wfo.wm_status = wfh.request['RequestStatus']
wfo.status = 'away'
session.commit()
continue
else:
print wfo.name, wfh.request['RequestStatus']
## retrieve from the schema, dbs and reqMgr what should be the next version
version = wfh.getNextVersion()
if not version:
if options and options.ProcessingVersion:
version = options.ProcessingVersion
else:
wfh.sendLog('assignor', "cannot decide on version number")
n_stalled += 1
wfo.status = 'trouble'
session.commit()
continue
original_sites_allowed = copy.deepcopy(sites_allowed)
wfh.sendLog('assignor', "Site white list %s" % sorted(sites_allowed))
override_sec_location = CI.get(wfh.request['Campaign'],
'SecondaryLocation', [])
blocks = []
if 'BlockWhitelist' in wfh.request:
blocks = wfh.request['BlockWhitelist']
if 'RunWhitelist' in wfh.request and wfh.request['RunWhitelist']:
## augment with run white list
for dataset in primary:
blocks = list(
set(blocks + getDatasetBlocks(
dataset, runs=wfh.request['RunWhitelist'])))
wfh.sendLog('assignor', "Allowed %s" % sorted(sites_allowed))
secondary_locations = None
primary_aaa = options.primary_aaa
if 'Campaign' in wfh.request and wfh.request[
'Campaign'] in CI.campaigns and 'primary_AAA' in CI.campaigns[
wfh.request['Campaign']]:
primary_aaa = primary_aaa or CI.campaigns[
wfh.request['Campaign']]['primary_AAA']
secondary_aaa = options.secondary_aaa
if 'Campaign' in wfh.request and wfh.request[
'Campaign'] in CI.campaigns and 'secondary_AAA' in CI.campaigns[
wfh.request['Campaign']]:
secondary_aaa = secondary_aaa or CI.campaigns[
wfh.request['Campaign']]['secondary_AAA']
for sec in list(secondary):
if override_sec_location:
print "We don't care where the secondary is"
print "Cannot pass for now"
sendEmail("tempting to pass sec location check",
"but we cannot yet IMO")
#pass
if secondary_aaa:
#just continue without checking
continue
presence = getDatasetPresence(url, sec)
print sec
print json.dumps(presence, indent=2)
one_secondary_locations = [
site for (site, (there, frac)) in presence.items()
if frac > 98.
]
#one_secondary_locations = [site for (site,(there,frac)) in presence.items() if there]
if secondary_locations == None:
secondary_locations = one_secondary_locations
else:
secondary_locations = list(
set(secondary_locations) & set(one_secondary_locations))
## reduce the site white list to site with secondary only
#sites_allowed = [site for site in sites_allowed if any([osite.startswith(site) for osite in one_secondary_locations])]
sites_allowed = [
site for site in sites_allowed
if SI.CE_to_SE(site) in one_secondary_locations
]
wfh.sendLog(
'assignor', "From secondary requirement, now Allowed%s" %
sorted(sites_allowed))
initial_sites_allowed = copy.deepcopy(
sites_allowed
) ## keep track of this, after secondary input location restriction : that's how you want to operate it
sites_all_data = copy.deepcopy(sites_allowed)
sites_with_data = copy.deepcopy(sites_allowed)
sites_with_any_data = copy.deepcopy(sites_allowed)
primary_locations = None
available_fractions = {}
set_lfn = '/store/mc' ## by default
endpoints = set()
for prim in list(primary):
if prim in dataset_endpoints:
print "endpoints from stagor", dataset_endpoints[prim]
endpoints.update(dataset_endpoints[prim])
set_lfn = getLFNbase(prim)
presence = getDatasetPresence(url, prim, only_blocks=blocks)
if talk:
print prim
print json.dumps(presence, indent=2)
available_fractions[prim] = getDatasetBlocksFraction(
url,
prim,
sites=[SI.CE_to_SE(site) for site in sites_allowed],
only_blocks=blocks)
#sites_all_data = [site for site in sites_with_data if any([osite.startswith(site) for osite in [psite for (psite,(there,frac)) in presence.items() if there]])]
#sites_with_data = [site for site in sites_with_data if any([osite.startswith(site) for osite in [psite for (psite,frac) in presence.items() if frac[1]>90.]])]
sites_all_data = [
site for site in sites_with_data if SI.CE_to_SE(site) in [
psite for (psite, (there, frac)) in presence.items()
if there
]
]
sites_with_data = [
site for site in sites_with_data if SI.CE_to_SE(site) in
[psite for (psite, frac) in presence.items() if frac[1] > 90.]
]
sites_with_any_data = [
site for site in sites_with_any_data
if SI.CE_to_SE(site) in presence.keys()
]
wfh.sendLog(
'assignor', "Holding the data but not allowed %s" % sorted(
list(
set([
se_site for se_site in presence.keys()
if not SI.SE_to_CE(se_site) in sites_allowed
]))))
if primary_locations == None:
primary_locations = presence.keys()
else:
primary_locations = list(
set(primary_locations) & set(presence.keys()))
sites_with_data = list(set(sites_with_data))
sites_with_any_data = list(set(sites_with_any_data))
opportunistic_sites = []
down_time = False
## opportunistic running where any piece of data is available
if secondary_locations or primary_locations:
## intersection of both any pieces of the primary and good IO
#opportunistic_sites = [SI.SE_to_CE(site) for site in list((set(secondary_locations) & set(primary_locations) & set(SI.sites_with_goodIO)) - set(sites_allowed))]
if secondary_locations and primary_locations:
opportunistic_sites = [
SI.SE_to_CE(site) for site in
list((set(secondary_locations) & set(primary_locations)) -
set([SI.CE_to_SE(site) for site in sites_allowed]))
]
elif primary_locations:
opportunistic_sites = [
SI.SE_to_CE(site) for site in list(
set(primary_locations) -
set([SI.CE_to_SE(site) for site in sites_allowed]))
]
else:
opportunistic_sites = []
wfh.sendLog(
'assignor', "We could be running in addition at %s" %
sorted(opportunistic_sites))
if any(
[osite in SI.sites_not_ready
for osite in opportunistic_sites]):
wfh.sendLog(
'assignor', "One of the usable site is in downtime %s" % ([
osite in SI.sites_not_ready
for osite in opportunistic_sites
]))
down_time = True
## should this be send back to considered ?
## should be 2 but for the time-being let's lower it to get things going
copies_wanted, cpuh = wfh.getNCopies()
wfh.sendLog('assignor', "we need %s CPUh" % cpuh)
if cpuh > max_cpuh_block and not options.go:
#sendEmail('large workflow','that wf %s has a large number of CPUh %s, not assigning, please check the logs'%(wfo.name, cpuh))#,destination=['*****@*****.**'])
sendLog(
'assignor',
'%s requires a large numbr of CPUh %s , not assigning, please check with requester'
% (wfo.name, cpuh),
level='critical')
wfh.sendLog(
'assignor',
"Requiring a large number of CPUh %s, not assigning" % cpuh)
continue
if 'Campaign' in wfh.request and wfh.request[
'Campaign'] in CI.campaigns and 'maxcopies' in CI.campaigns[
wfh.request['Campaign']]:
copies_needed_from_campaign = CI.campaigns[
wfh.request['Campaign']]['maxcopies']
copies_wanted = min(copies_needed_from_campaign, copies_wanted)
if not options.early:
less_copies_than_requested = UC.get("less_copies_than_requested")
copies_wanted = max(
1, copies_wanted -
less_copies_than_requested) # take one out for the efficiency
else:
## find out whether there is a site in the whitelist, that is lacking jobs and reduce to 1 copy needed to get things going
pass
wfh.sendLog('assignor',
"needed availability fraction %s" % copies_wanted)
## should also check on number of sources, if large enough, we should be able to overflow most, efficiently
if available_fractions and not all([
available >= copies_wanted
for available in available_fractions.values()
]):
not_even_once = not all([
available >= 1. for available in available_fractions.values()
])
wfh.sendLog(
'assignor',
"The input dataset is not available %s times, only %s" %
(copies_wanted, available_fractions.values()))
if down_time and not options.go and not options.early:
wfo.status = 'considered'
session.commit()
wfh.sendLog(
'assignor',
"sending back to considered because of site downtime, instead of waiting"
)
#sendEmail( "cannot be assigned due to downtime","%s is not sufficiently available, due to down time of a site in the whitelist. check the assignor logs. sending back to considered."% wfo.name)
sendLog(
'assignor',
'%s is not sufficiently available, due to down time of a site in the whitelist. sending back to considered.'
% (wfo.name),
level='delay')
continue
#pass
print json.dumps(available_fractions)
if (options.go and not_even_once) or not options.go:
known = []
try:
known = json.loads(open('cannot_assign.json').read())
except:
pass
if not wfo.name in known and not options.limit and not options.go and not options.early and not options.partial:
wfh.sendLog(
'assignor',
"cannot be assigned, %s is not sufficiently available.\n %s"
% (wfo.name, json.dumps(available_fractions)))
sendEmail(
"cannot be assigned",
"%s is not sufficiently available.\n %s" %
(wfo.name, json.dumps(available_fractions)))
known.append(wfo.name)
open('cannot_assign.json',
'w').write(json.dumps(known, indent=2))
n_stalled += 1
if options.early:
if wfo.status == 'considered':
wfh.sendLog('assignor', "setting considered-tried")
wfo.status = 'considered-tried'
session.commit()
else:
print "tried but status is", wfo.status
if options.partial:
print "Will move on with partial locations"
else:
continue
## default back to white list to original white list with any data
print "Allowed", sorted(sites_allowed)
if primary_aaa:
sites_allowed = initial_sites_allowed
options.TrustSitelists = True
wfh.sendLog(
'assignor', "Selected to read primary through xrootd %s" %
sorted(sites_allowed))
else:
sites_allowed = sites_with_any_data
wfh.sendLog('assignor',
"Selected for any data %s" % sorted(sites_allowed))
if secondary_aaa:
options.TrustPUSitelists = True
wfh.sendLog(
'assignor', "Reading secondary through xrootd from %s" %
sorted(sites_allowed))
### check on endpoints for on-going transfers
if endpoints and options.partial:
sites_allowed = list(
set(sites_allowed + [SI.SE_to_CE(s) for s in endpoints]))
print "with added endpoints", sorted(sites_allowed)
if not len(sites_allowed):
wfh.sendLog('assignor', "cannot be assign with no matched sites")
sendLog('assignor',
'%s has no whitelist' % wfo.name,
level='critical')
n_stalled += 1
continue
t1_only = [ce for ce in sites_allowed if ce.startswith('T1')]
if t1_only:
# try to pick from T1 only first
sites_out = [SI.pick_dSE([SI.CE_to_SE(ce) for ce in t1_only])]
else:
# then pick any otherwise
sites_out = [
SI.pick_dSE([SI.CE_to_SE(ce) for ce in sites_allowed])
]
wfh.sendLog('assignor', "Placing the output on %s" % sites_out)
parameters = {
'SiteWhitelist': sites_allowed,
'NonCustodialSites': sites_out,
'AutoApproveSubscriptionSites': list(set(sites_out)),
'AcquisitionEra': wfh.acquisitionEra(),
'ProcessingString': wfh.processingString(),
'MergedLFNBase': set_lfn,
'ProcessingVersion': version,
}
## plain assignment here
team = 'production'
if os.getenv('UNIFIED_TEAM'): team = os.getenv('UNIFIED_TEAM')
if options and options.team:
team = options.team
if False and 'T2_CH_CERN' in parameters['SiteWhitelist']:
## add some check on
### the amount pending to HLT
### the size of the request
### the priority of the request (maybe not if we decide to overflow during runs)
parameters['SiteWhitelist'] = ['T2_CH_CERN_HLT']
team = 'hlt'
## reduce the splitting by factor of 4, regardless of type of splitting
sendEmail("sending work to HLT",
"%s was assigned to HLT" % wfo.name)
##parse options entered in command line if any
if options:
for key in reqMgrClient.assignWorkflow.keys:
v = getattr(options, key)
if v != None:
if type(v) == str and ',' in v:
parameters[key] = filter(None, v.split(','))
else:
parameters[key] = v
if lheinput:
## throttle reading LHE article
wfh.sendLog('assignor',
'Setting the number of events per job to 500k max')
parameters['EventsPerJob'] = 500000
## pick up campaign specific assignment parameters
parameters.update(CI.parameters(wfh.request['Campaign']))
if not options.test:
parameters['execute'] = True
split_check = wfh.checkWorkflowSplitting()
if split_check != True:
parameters.update(split_check)
if 'EventBased' in split_check.values():
wfh.sendLog('assignor', "Falling back to event splitting.")
#sendEmail("Fallback to EventBased","the workflow %s is too heavy to be processed as it is. Fallback to EventBased splitting"%wfo.name)
sendLog(
'assignor',
'the workflow %s is too heavy to be processed as it is. Fallback to EventBased splitting'
% wfo.name,
level='critical')
elif 'EventsPerJob' in split_check.values():
wfh.sendLog('assignor',
"Modifying the number of job per event")
#sendEmail("Modifying the job per events","the workflow %s is too heavy in number of jobs explosion"%wfo.name)
sendLog(
'assignor',
"the workflow %s is too heavy in number of jobs explosion"
% wfo.name,
level='critical')
# Handle run-dependent MC
pstring = wfh.processingString()
if 'PU_RD' in pstring:
numEvents = wfh.getRequestNumEvents()
eventsPerLumi = [getDatasetEventsPerLumi(prim) for prim in primary]
eventsPerLumi = sum(eventsPerLumi) / float(len(eventsPerLumi))
reqJobs = 500
if 'PU_RD2' in pstring:
reqJobs = 2000
eventsPerJob = int(numEvents / (reqJobs * 1.4))
lumisPerJob = int(eventsPerJob / eventsPerLumi)
if lumisPerJob == 0:
#sendEmail("issue with event splitting for run-dependent MC","%s needs to be split by event with %s per job"%(wfo.name, eventsPerJob))
sendLog('assignor',
"%s needs to be split by event with %s per job" %
(wfo.name, eventsPerJob),
level='critical')
wfh.sendLog(
'assignor',
"%s needs to be split by event with %s per job" %
(wfo.name, eventsPerJob))
parameters['EventsPerJob'] = eventsPerJob
else:
spl = wfh.getSplittings()[0]
eventsPerJobEstimated = spl[
'events_per_job'] if 'events_per_job' in spl else None
eventsPerJobEstimated = spl[
'avg_events_per_job'] if 'avg_events_per_job' in spl else None
if eventsPerJobEstimated and eventsPerJobEstimated > eventsPerJob:
#sendEmail("setting lumi splitting for run-dependent MC","%s was assigned with %s lumis/job"%( wfo.name, lumisPerJob))
sendLog('assignor',
"%s was assigned with %s lumis/job" %
(wfo.name, lumisPerJob),
level='critical')
wfh.sendLog(
'assignor', "%s was assigned with %s lumis/job" %
(wfo.name, lumisPerJob))
parameters['LumisPerJob'] = lumisPerJob
else:
#sendEmail("leaving splitting untouched for PU_RD*","please check on "+wfo.name)
sendLog(
'assignor',
"leaving splitting untouched for %s, please check on %s"
% (pstring, wfo.name),
level='critical')
wfh.sendLog(
'assignor',
"leaving splitting untouched for PU_RD*, please check."
)
result = reqMgrClient.assignWorkflow(url, wfo.name, team, parameters)
# set status
if not options.test:
if result:
wfo.status = 'away'
session.commit()
n_assigned += 1
wfh.sendLog(
'assignor', "Properly assigned\n%s" %
(json.dumps(parameters, indent=2)))
try:
## refetch information and lock output
new_wfi = workflowInfo(url, wfo.name)
(_, prim, _, sec) = new_wfi.getIO()
for secure in list(prim) + list(
sec) + new_wfi.request['OutputDatasets']:
## lock all outputs flat
NLI.lock(secure)
#for site in [SI.CE_to_SE(site) for site in sites_allowed]:
# for output in new_wfi.request['OutputDatasets']:
# LI.lock( output, site, 'dataset in production')
# for primary in prim:
# LI.lock( primary, site, 'dataset used in input')
# for secondary in sec:
# LI.lock( secondary, site, 'required for mixing' )
except Exception as e:
print "fail in locking output"
print str(e)
sendEmail("failed locking of output", str(e))
else:
print "ERROR could not assign", wfo.name
else:
pass
print "Assignment summary:"
sendLog('assignor', "Assigned %d Stalled %s" % (n_assigned, n_stalled))
def 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 transferor(url, specific=None, talk=True, options=None):
if userLock(): return
if duplicateLock(): return
use_mcm = True
up = componentInfo(mcm=use_mcm, soft=['mcm'])
if not up.check(): return
use_mcm = up.status['mcm']
if options and options.test:
execute = False
else:
execute = True
SI = siteInfo()
CI = campaignInfo()
NLI = newLockInfo()
mcm = McMClient(dev=False)
dss = DSS()
#allowed_secondary = UC.get('')
print "counting all being handled..."
being_handled = len(
session.query(Workflow).filter(Workflow.status == 'away').all())
being_handled += len(
session.query(Workflow).filter(
Workflow.status.startswith('stag')).all())
being_transfered = len(
session.query(Workflow).filter(Workflow.status == 'staging').all())
being_handled += len(
session.query(Workflow).filter(
Workflow.status.startswith('assistance-')).all())
max_to_handle = options.maxworkflows
max_to_transfer = options.maxstaging
allowed_to_handle = max(0, max_to_handle - being_handled)
allowed_to_transfer = max(0, max_to_transfer - being_transfered)
wf_buffer = 5
if allowed_to_handle <= wf_buffer: ## buffer for having several wf per transfer
print "Not allowed to run more than", max_to_handle, "at a time. Currently", being_handled, "and", wf_buffer, "buffer"
else:
print being_handled, "already being handled", max_to_handle, "max allowed,", allowed_to_handle, "remaining", "and", wf_buffer, "buffer"
if allowed_to_transfer <= wf_buffer:
print "Not allowed to transfer more than", max_to_transfer, "at a time. Currently", being_transfered, "and", wf_buffer, "buffer"
else:
print being_transfered, "already being transfered", max_to_transfer, "max allowed,", allowed_to_transfer, "remaining", "and", wf_buffer, "buffer"
print "... done"
all_transfers = defaultdict(list)
workflow_dependencies = defaultdict(
set) ## list of wf.id per input dataset
wfs_and_wfh = []
print "getting all wf to consider ..."
cache = getWorkflows(url, 'assignment-approved', details=True)
for wfo in session.query(Workflow).filter(
Workflow.status.startswith('considered')).all():
print "\t", wfo.name
if specific and not specific in wfo.name: continue
cache_r = filter(lambda d: d['RequestName'] == wfo.name, cache)
if len(cache_r):
wfs_and_wfh.append((wfo,
workflowInfo(url,
wfo.name,
spec=False,
request=cache_r[0])))
else:
wfs_and_wfh.append((wfo, workflowInfo(url, wfo.name, spec=False)))
print "... done"
transfers_per_sites = defaultdict(int)
input_sizes = {}
ignored_input_sizes = {}
input_cput = {}
input_st = {}
## list the size of those in transfer already
in_transfer_priority = None
min_transfer_priority = None
print "getting all wf in staging ..."
stucks = json.loads(open('%s/stuck_transfers.json' % monitor_dir).read())
for wfo in session.query(Workflow).filter(
Workflow.status == 'staging').all():
wfh = workflowInfo(url, wfo.name, spec=False)
#(lheinput,primary,parent,secondary) = wfh.getIO()
#sites_allowed = getSiteWhiteList( (lheinput,primary,parent,secondary) )
(lheinput, primary, parent, secondary,
sites_allowed) = wfh.getSiteWhiteList()
for site in sites_allowed: ## we should get the actual transfer destination instead of the full white list
transfers_per_sites[site] += 1
#input_cput[wfo.name] = wfh.getComputingTime()
#input_st[wfo.name] = wfh.getSystemTime()
for prim in primary:
ds_s = dss.get(prim)
if prim in stucks:
sendLog('transferor',
"%s appears stuck, so not counting it %s [GB]" %
(prim, ds_s),
wfi=wfh)
ignored_input_sizes[prim] = ds_s
else:
input_sizes[prim] = ds_s
sendLog('transferor',
"%s needs %s [GB]" % (wfo.name, ds_s),
wfi=wfh)
if in_transfer_priority == None:
in_transfer_priority = int(wfh.request['RequestPriority'])
else:
in_transfer_priority = max(in_transfer_priority,
int(wfh.request['RequestPriority']))
if min_transfer_priority == None:
min_transfer_priority = int(wfh.request['RequestPriority'])
else:
min_transfer_priority = min(min_transfer_priority,
int(wfh.request['RequestPriority']))
if min_transfer_priority == None or in_transfer_priority == None:
print "nothing is lining up for transfer"
sendEmail("no request in staging", "no request in staging")
return
pass
try:
print "Ignored input sizes"
ignored_values = list(ignored_input_sizes.items())
ignored_values.sort(key=lambda i: i[1])
print "\n".join(map(str, ignored_values))
print "Considered input sizes"
considered_values = list(input_sizes.items())
considered_values.sort(key=lambda i: i[1])
print "\n".join(map(str, considered_values))
except Exception as e:
print "trying to print the summary of input size"
print str(e)
print "... done"
print "Max priority in transfer already", in_transfer_priority
print "Min priority in transfer already", min_transfer_priority
print "transfers per sites"
print json.dumps(transfers_per_sites, indent=2)
in_transfer_already = sum(input_sizes.values())
cput_in_transfer_already = sum(input_cput.values())
st_in_transfer_already = sum(input_st.values())
## list the size of all inputs
primary_input_per_workflow_gb = defaultdict(float)
print "getting all input sizes ..."
for (wfo, wfh) in wfs_and_wfh:
(_, primary, _, _) = wfh.getIO()
#input_cput[wfo.name] = wfh.getComputingTime()
#input_st[wfo.name] = wfh.getSystemTime()
for prim in primary:
## do not count it if it appears stalled !
prim_size = dss.get(prim)
input_sizes[prim] = prim_size
primary_input_per_workflow_gb[wfo.name] += prim_size
print "... done"
# shuffle first by name
random.shuffle(wfs_and_wfh)
# Sort smallest transfers first; allows us to transfer as many as possible workflows.
def prio_and_size(i, j):
if int(i[1].request['RequestPriority']) == int(
j[1].request['RequestPriority']):
return cmp(int(primary_input_per_workflow_gb.get(j[0].name, 0)),
int(primary_input_per_workflow_gb.get(i[0].name, 0)))
else:
return cmp(int(i[1].request['RequestPriority']),
int(j[1].request['RequestPriority']))
#wfs_and_wfh.sort(cmp = prio_and_size, reverse=True)
#wfs_and_wfh.sort(cmp = lambda i,j : cmp(int(primary_input_per_workflow_gb.get(i[0].name, 0)), int(primary_input_per_workflow_gb.get(j[0].name, 0)) ))
#sort by priority higher first
wfs_and_wfh.sort(cmp=lambda i, j: cmp(int(i[1].request[
'RequestPriority']), int(j[1].request['RequestPriority'])),
reverse=True)
cput_grand_total = sum(input_cput.values())
cput_to_transfer = cput_grand_total - cput_in_transfer_already
st_grand_total = sum(input_st.values())
st_to_transfer = st_grand_total - st_in_transfer_already
print "%15.4f [CPU h] worth already in transfer" % cput_in_transfer_already
print "%15.4f [CPU h] worth is the current requested transfer load" % cput_to_transfer
print "%15.4f [h] worth of absolute system time in transfer" % (
cput_in_transfer_already / SI.availableSlots())
print "%15.4f [h] worth of absolute system time is the current requested transfer load" % (
cput_to_transfer / SI.availableSlots())
print "%15.4f [h] worth of theoritical system time in transfer" % (
st_in_transfer_already)
print "%15.4f [h] worth of theoritical system time is the current requested transfer load" % (
st_to_transfer)
grand_total = sum(input_sizes.values())
to_transfer = grand_total - in_transfer_already
grand_transfer_limit = options.maxtransfer
#grand_transfer_limit = SI.total_disk()*0.25*1024## half of the free sapce in TB->GB
transfer_limit = grand_transfer_limit - in_transfer_already
print "%15.4f GB already being transfered" % in_transfer_already
print "%15.4f GB is the current requested transfer load" % to_transfer
print "%15.4f GB is the global transfer limit" % grand_transfer_limit
print "%15.4f GB is the available limit" % transfer_limit
max_staging_per_site = options.maxstagingpersite
# the max priority value per dataset.
max_priority = defaultdict(int)
needs_transfer = 0 ## so that we can count'em
passing_along = 0
transfer_sizes = {}
went_over_budget = False
destination_cache = {}
no_goes = set()
max_per_round = UC.get('max_per_round').get('transferor', None)
if max_per_round and not spec:
wfs_and_wfh = wfs_and_wfh[:max_per_round]
for (wfo, wfh) in wfs_and_wfh:
print wfo.name, "to be transfered with priority", wfh.request[
'RequestPriority']
if wfh.request['RequestStatus'] != 'assignment-approved':
if wfh.request['RequestStatus'] in [
'aborted', 'rejected', 'rejected-archived',
'aborted-archived'
]:
wfo.status = 'trouble' ## so that we look or a replacement
else:
wfo.status = 'away'
wfh.sendLog(
'transferor', '%s in status %s, setting %s' %
(wfo.name, wfh.request['RequestStatus'], wfo.status))
continue
(_, primary, _, _) = wfh.getIO()
this_load = sum([input_sizes[prim] for prim in primary])
no_budget = False
if (this_load
and (sum(transfer_sizes.values()) + this_load > transfer_limit
or went_over_budget)):
if went_over_budget:
wfh.sendLog('transferor', "Transfer has gone over bubget.")
else:
wfh.sendLog('transferor', "Transfer will go over bubget.")
wfh.sendLog(
'transferor',
"%15.4f GB this load, %15.4f GB already this round, %15.4f GB is the available limit"
% (this_load, sum(transfer_sizes.values()), transfer_limit))
#if sum(transfer_sizes.values()) > transfer_limit:
went_over_budget = True
if in_transfer_priority != None and min_transfer_priority != None:
if int(
wfh.request['RequestPriority']
) >= in_transfer_priority and min_transfer_priority != in_transfer_priority:
wfh.sendLog(
'transferor',
"Higher priority sample %s >= %s go-on over budget" %
(wfh.request['RequestPriority'], in_transfer_priority))
else:
if not options.go:
wfh.sendLog(
'transferor',
"%s minimum priority %s < %s : stop" %
(min_transfer_priority,
wfh.request['RequestPriority'],
in_transfer_priority))
no_budget = True
## throtlle by campaign go
no_go = False
if not wfh.go(log=True) and not options.go:
no_go = True
no_goes.add(wfo.name)
allowed_secondary = set()
for campaign in wfh.getCampaigns():
if campaign in CI.campaigns and 'secondaries' in CI.campaigns[
campaign]:
allowed_secondary.update(CI.campaigns[campaign]['secondaries'])
if secondary:
if (secondary and allowed_secondary) and (
set(secondary) & allowed_secondary != set(secondary)):
wfh.sendLog(
'assignor', '%s is not an allowed secondary' %
(', '.join(set(secondary) - allowed_secondary)))
no_go = True
if no_go:
continue
## check if the batch is announced
def check_mcm(wfn):
announced = False
is_real = False
if not wfn.startswith('pdmvserv'):
is_real = True
try:
for b in mcm.getA('batches', query='contains=%s' % wfo.name):
is_real = True
if b['status'] == 'announced':
announced = True
break
except:
try:
for b in mcm.getA('batches',
query='contains=%s' % wfo.name):
is_real = True
if b['status'] == 'announced':
announced = True
break
except:
print "could not get mcm batch announcement, assuming not real"
return announced, is_real
if not use_mcm:
announced, is_real = False, True
else:
if wfh.request['RequestType'] in ['ReReco']:
announced, is_real = True, True
else:
announced, is_real = check_mcm(wfo.name)
if not announced:
wfh.sendLog('transferor', "does not look announced.")
if not is_real:
wfh.sendLog('transferor', "does not appear to be genuine.")
## prevent any duplication. if the wf is not mentioned in any batch, regardless of status
continue
## check on a grace period
injection_time = time.mktime(
time.strptime('.'.join(map(str, wfh.request['RequestDate'])),
"%Y.%m.%d.%H.%M.%S")) / (60. * 60.)
now = time.mktime(time.gmtime()) / (60. * 60.)
if float(now - injection_time) < 4.:
if not options.go and not announced:
wfh.sendLog(
'transferor',
"It is too soon to start transfer: %3.2fH remaining" %
(now - injection_time))
continue
if passing_along >= allowed_to_handle:
#if int(wfh.request['RequestPriority']) >= in_transfer_priority and min_transfer_priority!=in_transfer_priority:
if in_transfer_priority != None and min_transfer_priority != None:
if int(wfh.request['RequestPriority']
) >= in_transfer_priority and int(
wfh.request['RequestPriority']
) != min_transfer_priority:
## higher priority, and not only this priority being transfered
wfh.sendLog(
'transferor',
"Higher priority sample %s >= %s go-on over %s" %
(wfh.request['RequestPriority'], in_transfer_priority,
max_to_handle))
else:
wfh.sendLog(
'transferor',
" Not allowed to pass more than %s at a time. Currently %s handled, and adding %s"
% (max_to_handle, being_handled, passing_along))
if not options.go:
## should not allow to jump that fence
break
if this_load and needs_transfer >= allowed_to_transfer:
if in_transfer_priority != None and min_transfer_priority != None:
if int(wfh.request['RequestPriority']
) >= in_transfer_priority and int(
wfh.request['RequestPriority']
) != min_transfer_priority:
## higher priority, and not only this priority being transfered
wfh.sendLog(
'transferor',
"Higher priority sample %s >= %s go-on over %s" %
(wfh.request['RequestPriority'], in_transfer_priority,
max_to_transfer))
else:
wfh.sendLog(
'transferor',
"Not allowed to transfer more than %s at a time. Currently %s transfering, and adding %s"
% (max_to_transfer, being_transfered, needs_transfer))
if not options.go:
no_budget = True
if no_budget:
continue
## the site white list considers site, campaign, memory and core information
(lheinput, primary, parent, secondary,
sites_allowed) = wfh.getSiteWhiteList()
if options and options.tosites:
sites_allowed = options.tosites.split(',')
for dataset in list(primary) + list(parent) + list(secondary):
## lock everything flat
NLI.lock(dataset)
if not sites_allowed:
wfh.sendLog('transferor', "not possible site to run at")
#sendEmail("no possible sites","%s has no possible sites to run at"%( wfo.name ))
sendLog('transferor',
"%s has no possible sites to run at" % (wfo.name),
level='critical')
continue
blocks = []
if 'BlockWhitelist' in wfh.request and wfh.request['BlockWhitelist']:
blocks = wfh.request['BlockWhitelist']
if 'RunWhitelist' in wfh.request and wfh.request['RunWhitelist']:
## augment with run white list
for dataset in primary:
blocks = list(
set(blocks + getDatasetBlocks(
dataset, runs=wfh.request['RunWhitelist'])))
if 'LumiList' in wfh.request and wfh.request['LumiList']:
## augment with the lumi white list
blocks = list(
set(blocks +
getDatasetBlocks(dataset, lumis=wfh.request['LumiList'])))
if blocks:
print "Reading", len(blocks), "in block whitelist"
can_go = True
staging = False
allowed = True
primary_destinations = set()
if primary:
copies_needed_from_CPUh, CPUh = wfh.getNCopies()
if talk:
print wfo.name, 'reads', ', '.join(primary), 'in primary'
## chope the primary dataset
for prim in primary:
## keep track of what needs what
workflow_dependencies[prim].add(wfo.id)
max_priority[prim] = max(max_priority[prim],
int(wfh.request['RequestPriority']))
wfh.sendLog(
'transferor', "Would make %s from cpu requirement %s" %
(copies_needed_from_CPUh, CPUh))
copies_needed = copies_needed_from_CPUh
if 'Campaign' in wfh.request and wfh.request[
'Campaign'] in CI.campaigns and 'maxcopies' in CI.campaigns[
wfh.request['Campaign']]:
copies_needed_from_campaign = CI.campaigns[
wfh.request['Campaign']]['maxcopies']
copies_needed = min(copies_needed_from_campaign,
copies_needed)
wfh.sendLog(
'transferor',
"Maxed to %s by campaign configuration %s" %
(copies_needed, wfh.request['Campaign']))
### new ways of making the whole thing
destinations, all_block_names = getDatasetDestinations(
url,
prim,
within_sites=[SI.CE_to_SE(site) for site in sites_allowed],
only_blocks=blocks)
print json.dumps(destinations, indent=2)
## get where the dataset is in full and completed
prim_location = [
site for (site, info) in destinations.items()
if info['completion'] == 100 and info['data_fraction'] == 1
]
## the rest is places it is going to be
prim_destination = [
site for site in destinations.keys()
if not site in prim_location
]
if len(prim_location) >= copies_needed:
wfh.sendLog(
'transferor',
"The input is all fully in place at %s sites %s" %
(len(prim_location), sorted(prim_location)))
continue
copies_needed = max(0, copies_needed - len(prim_location))
wfh.sendLog(
'transferor',
"not counting existing copies ; now need %s" %
copies_needed)
copies_being_made = [
sum([
info['blocks'].keys().count(block)
for site, info in destinations.items()
if site in prim_destination
]) for block in all_block_names
]
latching_on_transfers = set()
[
latching_on_transfers.update(info['blocks'].values())
for site, info in destinations.items()
if site in prim_destination
]
latching_on_transfers = list(latching_on_transfers)
#print latching_on_transfers
## figure out where all this is going to go
prim_to_distribute = [
site for site in sites_allowed
if not SI.CE_to_SE(site) in prim_location
]
prim_to_distribute = [
site for site in prim_to_distribute
if not SI.CE_to_SE(site) in prim_destination
]
## take out the ones that cannot receive transfers
prim_to_distribute = [
site for site in prim_to_distribute if not any([
osite.startswith(site)
for osite in SI.sites_veto_transfer
])
]
wfh.sendLog(
'transferor',
"Could be going to: %s" % sorted(prim_to_distribute))
if not prim_to_distribute or any([
transfers_per_sites[site] < max_staging_per_site
for site in prim_to_distribute
]):
## means there is openings let me go
print "There are transfer slots available:", [
(site, transfers_per_sites[site])
for site in prim_to_distribute
]
#for site in sites_allowed:
# #increment accross the board, regardless of real destination: could be changed
# transfers_per_sites[site] += 1
else:
if int(
wfh.request['RequestPriority']
) >= in_transfer_priority and min_transfer_priority != in_transfer_priority:
wfh.sendLog(
'transferor',
"Higher priority sample %s >= %s go-on over transfer slots available"
% (wfh.request['RequestPriority'],
in_transfer_priority))
else:
wfh.sendLog(
'transferor',
"Not allowed to transfer more than %s per site at a time. Going overboard for %s"
% (max_staging_per_site,
sorted([
site for site in prim_to_distribute
if transfers_per_sites[site] >=
max_staging_per_site
])))
if not options.go:
allowed = False
break
for latching in latching_on_transfers:
tfo = session.query(Transfer).filter(
Transfer.phedexid == int(latching)).first()
if not tfo:
tfo = session.query(Transfer).filter(
Transfer.phedexid == -int(latching)).first()
if not tfo:
tfo = Transfer(phedexid=latching)
tfo.workflows_id = []
session.add(tfo)
else:
tfo.phedexid = latching ## make it positive ever
if not wfo.id in tfo.workflows_id:
print "adding", wfo.id, "to", tfo.id, "with phedexid", latching
l = copy.deepcopy(tfo.workflows_id)
l.append(wfo.id)
tfo.workflows_id = l
if not options.test:
session.commit()
else:
session.flush(
) ## regardless of commit later on, we need to let the next wf feeding on this transfer to see it in query
can_go = False
transfer_sizes[prim] = input_sizes[prim]
staging = True
# reduce the number of copies required by the on-going full transfer : how do we bootstrap on waiting for them ??
#copies_needed = max(0,copies_needed - len(prim_destination))
copies_needed = max(0, copies_needed - min(copies_being_made))
wfh.sendLog(
'transferor',
"Not counting the copies being made ; then need %s" %
copies_needed)
if copies_needed == 0:
wfh.sendLog(
'transferor',
"The output is either fully in place or getting in full somewhere with %s"
% latching_on_transfers)
can_go = True
continue
elif len(prim_to_distribute) == 0:
wfh.sendLog(
'transferor',
"We are going to need extra copies, but no destinations seems available"
)
prim_to_distribute = [
site for site in sites_allowed
if not SI.CE_to_SE(site) in prim_location
]
prim_to_distribute = [
site for site in prim_to_distribute if not any([
osite.startswith(site)
for osite in SI.sites_veto_transfer
])
]
if len(
prim_to_distribute
) > 0: ## maybe that a parameter we can play with to limit the
if not options or options.chop:
### hard include the tape disk andpoint ?
#tapes = [site for site in getDatasetPresence( url, prim, vetos=['T0','T2','T3','Disk']) if site.endswith('MSS')]
chops, sizes = getDatasetChops(
prim,
chop_threshold=options.chopsize,
only_blocks=blocks)
spreading = distributeToSites(chops,
prim_to_distribute,
n_copies=copies_needed,
weights=SI.cpu_pledges,
sizes=sizes)
transfer_sizes[prim] = sum(sizes)
if not spreading:
sendLog(
'transferor',
'cannot send %s to any site, it cannot fit anywhere'
% prim,
level='critical')
wfh.sendLog(
'transferor',
"cannot send to any site. %s cannot seem to fit anywhere"
% (prim))
staging = False
can_go = False
else:
spreading = {}
for site in prim_to_distribute:
if blocks:
spreading[site] = blocks
else:
spreading[site] = [prim]
transfer_sizes[prim] = input_sizes[
prim] ## this is approximate if blocks are specified
can_go = False
wfh.sendLog(
'transferor', "selected CE destinations %s" %
(sorted(spreading.keys())))
for (site, items) in spreading.items():
all_transfers[site].extend(items)
transfers_per_sites[site] += 1
primary_destinations.add(site)
if not allowed:
wfh.sendLog('transferor', "Not allowed to move on with")
continue
if secondary:
override_sec_destination = []
if 'SecondaryLocation' in CI.campaigns[wfh.request['Campaign']]:
override_sec_destination = CI.campaigns[
wfh.request['Campaign']]['SecondaryLocation']
print wfo.name, 'reads', ', '.join(secondary), 'in secondary'
for sec in secondary:
workflow_dependencies[sec].add(wfo.id)
if True:
## new style, failing on minbias
if not sec in destination_cache:
## this is barbbaric, and does not show the correct picture on workflow by workflow with different whitelist
destination_cache[sec], _ = getDatasetDestinations(
url, sec) ## NO SITE WHITE LIST ADDED
#destination_cache[sec],_ = getDatasetDestinations(url, sec, within_sites = [SI.CE_to_SE(site) for site in sites_allowed])
## limit to the site whitelist NOW
se_allowed = [SI.CE_to_SE(site) for site in sites_allowed]
destinations = dict([
(k, v) for (k, v) in destination_cache[sec].items()
if site in se_allowed
])
## truncate location/destination to those making up for >90% of the dataset
bad_destinations = [
destinations.pop(site)
for (site, info) in destinations.items()
if info['data_fraction'] < 0.9
]
sec_location = [
site for (site, info) in destinations.items()
if info['completion'] >= 95
]
sec_destination = [
site for site in destinations.keys()
if not site in sec_location
]
else:
## old style
presence = getDatasetPresence(url, sec)
sec_location = [
site for site, pres in presence.items()
if pres[1] > 90.
] ## more than 90% of the minbias at sites
subscriptions = listSubscriptions(url, sec)
sec_destination = [site for site in subscriptions]
sec_to_distribute = [
site for site in sites_allowed if
not any([osite.startswith(site) for osite in sec_location])
]
sec_to_distribute = [
site for site in sec_to_distribute if not any(
[osite.startswith(site) for osite in sec_destination])
]
sec_to_distribute = [
site for site in sec_to_distribute if not any([
osite.startswith(site)
for osite in SI.sites_veto_transfer
])
]
if override_sec_destination:
## intersect with where we want the PU to be
not_needed_anymore = list(
set(sec_to_distribute) - set(override_sec_destination))
#sendEmail("secondary superfluous","the dataset %s could be removed from %s"%( sec, not_needed_anymore ))
sendLog(
'transferor',
"the dataset %s could be removed from %s" %
(sec, not_needed_anymore))
sec_to_distribute = list(
set(sec_to_distribute) & set(override_sec_destination))
if len(sec_to_distribute) > 0:
print "secondary could go to", sorted(sec_to_distribute)
sec_size = dss.get(sec)
for site in sec_to_distribute:
site_se = SI.CE_to_SE(site)
if (SI.disk[site_se] * 1024.) > sec_size:
all_transfers[site].append(sec)
can_go = False
else:
print "could not send the secondary input to", site_se, "because it is too big for the available disk", SI.disk[
site_se] * 1024, "GB need", sec_size
if primary_destinations and site in primary_destinations:
#sendEmail('secondary input too big','%s is too big (%s) for %s (%s)'%( sec, sec_size, site_se, SI.disk[site_se]*1024))
sendLog('transferor',
'%s is too big (%s) for %s (%s)' %
(sec, sec_size, site_se,
SI.disk[site_se] * 1024),
level='critical')
else:
print "the secondary input does not have to be send to site"
## is that possible to do something more
if can_go:
## no explicit transfer required this time
if staging:
## but using existing ones
wfh.sendLog(
'transferor',
"latches on existing transfers, and nothing else, settin staging"
)
wfo.status = 'staging'
needs_transfer += 1
else:
wfh.sendLog(
'transferor', "should just be assigned now to %s" %
sorted(sites_allowed))
wfo.status = 'staged'
passing_along += 1
wfh.sendLog('transferor', "setting status to %s" % wfo.status)
session.commit()
continue
else:
## there is an explicit transfer required
if staging:
## and also using an existing one
wfh.sendLog('transferor', "latches on existing transfers")
if not options.test:
wfo.status = 'staging'
wfh.sendLog('transferor',
"setting status to %s" % wfo.status)
session.commit()
wfh.sendLog('transferor', "needs a transfer")
needs_transfer += 1
passing_along += 1
if no_goes:
#sendEmail("no go for managing","No go for \n"+"\n".join( no_goes ))
sendLog('transferor',
"No go for \n" + "\n".join(no_goes),
level='critical')
print "accumulated transfers"
print json.dumps(all_transfers, indent=2)
fake_id = -1
wf_id_in_prestaging = set()
for (site, items_to_transfer) in all_transfers.iteritems():
items_to_transfer = list(set(items_to_transfer))
## convert to storage element
site_se = SI.CE_to_SE(site)
## site that do not want input datasets
if site in SI.sites_veto_transfer:
print site, "does not want transfers"
continue
## throttle the transfer size to T2s ? we'd be screwed by a noPU sample properly configured.
## massage a bit the items
blocks = [it for it in items_to_transfer if '#' in it]
block_datasets = list(set([it.split('#')[0] for it in blocks]))
datasets = [it for it in items_to_transfer if not '#' in it]
details_text = "Making a replica to %s (CE) %s (SE) for" % (site,
site_se)
#print "\t",len(blocks),"blocks"
## remove blocks if full dataset is send out
blocks = [
block for block in blocks if not block.split('#')[0] in datasets
]
#print "\t",len(blocks),"needed blocks for",list(set([block.split('#')[0] for block in blocks]))
#print "\t",len(datasets),"datasets"
#print "\t",datasets
details_text += '\n\t%d blocks' % len(blocks)
details_text += '\n\t%d needed blocks for %s' % (
len(blocks),
sorted(list(set([block.split('#')[0] for block in blocks]))))
details_text += '\n\t%d datasets' % len(datasets)
details_text += '\n\t%s' % sorted(datasets)
items_to_transfer = blocks + datasets
if execute:
sendLog('transferor', details_text)
else:
print "Would make a replica to", site, "(CE)", site_se, "(SE) for"
print details_text
## operate the transfer
if options and options.stop:
## ask to move-on
answer = raw_input('Continue with that ?')
if not answer.lower() in ['y', 'yes', 'go']:
continue
if execute:
priority = 'normal'
cds = [
ds for ds in datasets + block_datasets if ds in max_priority
]
if cds and False: ## I don't think this is working. subscription should be updated on the fly and regularly for raising the priority if needed
## decide on an overall priority : that's a bit too large though
if any([max_priority[ds] >= 90000 for ds in cds]):
priority = 'high'
elif all([max_priority[ds] < 80000 for ds in cds]):
priority = 'low'
result = makeReplicaRequest(url,
site_se,
items_to_transfer,
'prestaging',
priority=priority)
else:
result = {'phedex': {'request_created': []}}
fake_id -= 1
if not result:
print "ERROR Could not make a replica request for", site, items_to_transfer, "pre-staging"
continue
for phedexid in [o['id'] for o in result['phedex']['request_created']]:
new_transfer = session.query(Transfer).filter(
Transfer.phedexid == int(phedexid)).first()
if not new_transfer:
new_transfer = session.query(Transfer).filter(
Transfer.phedexid == -int(phedexid)).first()
print phedexid, "transfer created"
if not new_transfer:
new_transfer = Transfer(phedexid=phedexid)
session.add(new_transfer)
else:
new_transfer.phedexid = phedexid ## make it positive again
new_transfer.workflows_id = set()
for transfering in list(
set(map(lambda it: it.split('#')[0], items_to_transfer))):
new_transfer.workflows_id.update(
workflow_dependencies[transfering])
new_transfer.workflows_id = list(new_transfer.workflows_id)
wf_id_in_prestaging.update(new_transfer.workflows_id)
session.commit()
## auto approve it
if execute:
approved = approveSubscription(url, phedexid, [site_se])
for wfid in wf_id_in_prestaging:
tr_wf = session.query(Workflow).get(wfid)
if tr_wf and tr_wf.status != 'staging':
if execute:
tr_wf.status = 'staging'
if talk:
print "setting", tr_wf.name, "to staging"
session.commit()
def equalizor(url, specific=None, options=None):
up = componentInfo(mcm=False, soft=['mcm'])
if not up.check(): return
if not specific:
workflows = getWorkflows(url, status='running-closed', details=True)
workflows.extend(getWorkflows(url, status='running-open',
details=True))
## start from scratch
modifications = defaultdict(dict)
## define regionality site => fallback allowed. feed on an ssb metric ??
mapping = defaultdict(list)
reversed_mapping = defaultdict(list)
regions = defaultdict(list)
SI = siteInfo()
CI = campaignInfo()
UC = unifiedConfiguration()
for site in SI.sites_ready:
region = site.split('_')[1]
if not region in ['US', 'DE', 'IT']: continue
regions[region] = [region]
def site_in_depletion(s):
return True
if s in SI.sites_pressure:
(m, r, pressure) = SI.sites_pressure[s]
if float(m) < float(r):
print s, m, r, "lacking pressure"
return True
else:
print s, m, r, "pressure"
pass
return False
for site in SI.sites_ready:
region = site.split('_')[1]
## fallback to the region, to site with on-going low pressure
mapping[site] = [
fb for fb in SI.sites_ready
if any([('_%s_' %
(reg) in fb and fb != site and site_in_depletion(fb))
for reg in regions[region]])
]
use_T0 = ('T0_CH_CERN' in UC.get("site_for_overflow"))
if options.t0: use_T0 = True
#if options.augment : use_T0 = True
use_HLT = ('T2_CH_CERN_HLT' in UC.get("site_for_overflow"))
if options.hlt: use_HLT = True
#if options.augment : use_HLT=True
if use_HLT:
mapping['T2_CH_CERN'].append('T2_CH_CERN_HLT')
if use_T0:
mapping['T2_CH_CERN'].append('T0_CH_CERN')
#mapping['T1_FR_CCIN2P3'].append('T0_CH_CERN')
#mapping['T2_IT_Legnaro'].append('T1_IT_CNAF')
for reg in ['IT', 'DE', 'UK']:
mapping['T2_CH_CERN'].extend(
[fb for fb in SI.sites_ready if '_%s_' % reg in fb])
## make them appear as OK to use
force_sites = []
## overflow CERN to underutilized T1s
upcoming = json.loads(open('%s/GQ.json' % monitor_dir).read())
for possible in SI.sites_T1s:
if not possible in upcoming:
mapping['T2_CH_CERN'].append(possible)
## remove add-hoc sites from overflow mapping
prevent_sites = ['T2_US_Purdue']
for prevent in prevent_sites:
if prevent in mapping: mapping.pop(prevent)
for src in mapping:
for prevent in prevent_sites:
if prevent in mapping[src]:
mapping[src].remove(prevent)
## create the reverse mapping for the condor module
for site, fallbacks in mapping.items():
for fb in fallbacks:
reversed_mapping[fb].append(site)
## this is the fallback mapping
print "Direct mapping : site => overflow"
print json.dumps(mapping, indent=2)
print "Reverse mapping : dest <= from origin"
print json.dumps(reversed_mapping, indent=2)
altered_tasks = set()
def running_idle(wfi, task_name):
gmon = wfi.getGlideMon()
#print gmon
if not gmon: return (0, 0)
if not task_name in gmon: return (0, 0)
return (gmon[task_name]['Running'], gmon[task_name]['Idle'])
def needs_action(wfi, task, min_idled=100, pressure=0.2):
task_name = task.pathName.split('/')[-1]
running, idled = running_idle(wfi, task_name)
go = True
if not idled and not running:
go = False
if idled < 100:
go = False
if (not running and idled) or (running and
(idled / float(running) > pressure)):
go = True
else:
go = False
return go, task_name, running, idled
def getPerf(task):
task = task.split('/')[1] + '/' + task.split('/')[-1]
try:
u = 'http://cms-gwmsmon.cern.ch/prodview/json/history/memoryusage720/%s' % task
print u
perf_data = json.loads(os.popen('curl -s --retry 5 %s' % u).read())
except Exception as e:
print str(e)
return (None, None)
buckets = perf_data['aggregations']["2"]['buckets']
s_m = sum(bucket['key'] * bucket['doc_count'] for bucket in buckets)
w_m = sum(bucket['doc_count'] for bucket in buckets)
m_m = max(bucket['key'] for bucket in buckets) if buckets else None
b_m = None
if w_m > 100:
b_m = m_m
try:
perf_data = json.loads(
os.popen(
'curl -s --retry 5 http://cms-gwmsmon.cern.ch/prodview/json/history/runtime720/%s'
% task).read())
except Exception as e:
print str(e)
return (b_m, None)
buckets = perf_data['aggregations']["2"]['buckets']
s_t = sum(bucket['key'] * bucket['doc_count'] for bucket in buckets)
w_t = sum(bucket['doc_count'] for bucket in buckets)
m_t = max(bucket['key'] for bucket in buckets) if buckets else None
b_t = None
if w_t > 100:
b_t = m_t
return (b_m, b_t)
def getcampaign(task):
taskname = task.pathName.split('/')[-1]
if hasattr(task, 'prepID'):
return task.prepID.split('-')[1]
elif taskname.count('-') >= 1:
return taskname.split('-')[1]
else:
return None
def close(interface):
open('%s/equalizor.json.new' % monitor_dir,
'w').write(json.dumps(interface, indent=2))
os.system('mv %s/equalizor.json.new %s/equalizor.json' %
(monitor_dir, monitor_dir))
os.system('cp %s/equalizor.json %s/logs/equalizor/equalizor.%s.json' %
(monitor_dir, monitor_dir, time.mktime(time.gmtime())))
interface = {'reversed_mapping': reversed_mapping, 'modifications': {}}
if options.augment or options.remove:
interface['modifications'] = json.loads(
open('%s/equalizor.json' % monitor_dir).read())['modifications']
if options.remove:
if specific in interface['modifications']:
print "poping", specific
interface['modifications'].pop(specific)
close(interface)
return
PU_locations = {}
PU_overflow = {}
LHE_overflow = {}
tune_performance = []
pending_HLT = 0
max_HLT = 60000
pending_T0 = 0
max_T0 = 60000
try:
gmon = json.loads(
os.popen(
'curl -s http://cms-gwmsmon.cern.ch/prodview/json/T2_CH_CERN_HLT'
).read())
pending_HLT += gmon["Running"]
pending_HLT += gmon["MatchingIdle"]
except:
pass
stay_within_site_whitelist = False
specific_task = None
if specific and ":" in specific:
specific, specific_task = specific.split(':')
if specific:
wfs = session.query(Workflow).filter(
Workflow.name.contains(specific)).all()
else:
wfs = session.query(Workflow).filter(Workflow.status == 'away').all()
performance = {}
no_routing = []
random.shuffle(wfs)
for wfo in wfs:
if wfo.name in no_routing and not options.augment:
continue
if specific and not specific in wfo.name:
continue
if specific:
wfi = workflowInfo(url, wfo.name)
else:
cached = filter(lambda d: d['RequestName'] == wfo.name, workflows)
if not cached: continue
wfi = workflowInfo(url, wfo.name, request=cached[0])
## only running should get re-routed
if not wfi.request['RequestStatus'] in [
'running-open', 'running-closed'
] and not specific:
continue
tasks_and_campaigns = []
for task in wfi.getWorkTasks():
tasks_and_campaigns.append((task, getcampaign(task)))
_, _, _, sec = wfi.getIO()
## check needs override
needs_overide = False
if not needs_overide and options.augment: needs_overide = True
def overide_from_agent(wfi, needs_overide):
bad_agents = [] #'http://cmssrv219.fnal.gov:5984']
if not bad_agents: return needs_overide
if needs_overide: return True
agents = wfi.getAgents()
wqss = ['Running', 'Acquired']
if any([
agent in agents.get(wqs, {}).keys()
for wqs, agent in itertools.product(wqss, bad_agents)
]):
print "overriding the need for bad agent"
needs_overide = True
return needs_overide
## now parse this for action
for i_task, (task, campaign) in enumerate(tasks_and_campaigns):
if options.augment:
print task.pathName
print campaign
tune = CI.get(campaign, 'tune', options.tune)
if tune and not campaign in tune_performance:
tune_performance.append(campaign)
overflow = CI.get(campaign, 'overflow', {})
if overflow:
if "PU" in overflow and not campaign in PU_overflow:
PU_overflow[campaign] = copy.deepcopy(overflow['PU'])
print "adding", campaign, "to PU overflow rules"
if "LHE" in overflow and not campaign in LHE_overflow:
print "adding", campaign, "to light input overflow rules"
site_list = overflow['LHE']['site_list']
LHE_overflow[campaign] = copy.deepcopy(
getattr(SI, site_list))
### get the task performance, for further massaging.
if campaign in tune_performance or options.tune:
print "performance", task.taskType, task.pathName
if task.taskType in ['Processing', 'Production']:
set_memory, set_time = getPerf(task.pathName)
#print "Performance %s GB %s min"%( set_memory,set_time)
wfi.sendLog(
'equalizor', 'Performance tuning to %s GB %s min' %
(set_memory, set_time))
## get values from gmwsmon
# massage the values : 95% percentile
performance[task.pathName] = {}
if set_memory:
performance[task.pathName]['memory'] = set_memory
if set_time and False:
performance[task.pathName]['time'] = set_time
### rule to avoid the issue of taskchain secondary jobs being stuck at sites processing the initial step
if campaign in LHE_overflow:
if task.taskType in ['Processing']:
needs, task_name, running, idled = needs_action(wfi, task)
needs_overide = overide_from_agent(wfi, needs_overide)
extend_to = list(set(copy.deepcopy(
LHE_overflow[campaign])))
if stay_within_site_whitelist:
extend_to = list(
set(extend_to) & set(wfi.request['SiteWhitelist'])
) ## restrict to stupid-site-whitelist
extend_to = list(
set(extend_to) & set(SI.sites_ready + force_sites))
if extend_to and needs or needs_overide:
modifications[wfo.name][task.pathName] = {
"ReplaceSiteWhitelist": extend_to,
"Running": running,
"Pending": idled,
"Priority": wfi.request['RequestPriority']
}
wfi.sendLog(
'equalizor',
'%s of %s is running %d and pending %d, taking action : ReplaceSiteWhitelist \n %s'
%
(task_name, wfo.name, running, idled,
json.dumps(
sorted(modifications[wfo.name][task.pathName]
['ReplaceSiteWhitelist']))))
altered_tasks.add(task.pathName)
else:
wfi.sendLog(
'equalizor',
'%s of %s is running %d and pending %d' %
(task_name, wfo.name, running, idled))
### overflow the 76 digi-reco to the site holding the pileup
if campaign in PU_overflow:
force = PU_overflow[campaign][
'force'] if 'force' in PU_overflow[campaign] else False
secondary_locations = set(SI.sites_ready + force_sites)
for s in sec:
if not s in PU_locations:
presence = getDatasetPresence(url, s)
#one_secondary_locations = [site for (site,(there,frac)) in presence.items() if there]
one_secondary_locations = [
site for (site, (there, frac)) in presence.items()
if frac > 98.
]
PU_locations[s] = one_secondary_locations
print "secondary is at", sorted(PU_locations[s])
secondary_locations = set(
[SI.SE_to_CE(site)
for site in PU_locations[s]]) & secondary_locations
## we should add all sites that hold the secondary input if any
### given that we have the secondary location available, it is not necessary to use the add-hoc list
##secondary_locations = list(set(PU_overflow[campaign]['sites']) & set( SI.sites_ready ))
if any([
task.pathName.endswith(finish)
for finish in ['_0', 'StepOneProc', 'Production']
]):
needs, task_name, running, idled = needs_action(wfi, task)
## removing the ones in the site whitelist already since they encode the primary input location
if stay_within_site_whitelist:
original_site_in_use = set(
wfi.request['SiteWhitelist'])
else:
original_site_in_use = set(secondary_locations)
## remove the sites that have already running jobs
gmon = wfi.getGlideMon()
if gmon and task_name in gmon and 'Sites' in gmon[
task_name]:
site_in_use = set(gmon[task_name]['Sites'])
print "removing", sorted(site_in_use)
## that determines where you want to run in addition
augment_by = list((set(secondary_locations) -
site_in_use) & original_site_in_use)
else:
print "no existing running site"
augment_by = list(original_site_in_use)
needs_overide = overide_from_agent(wfi, needs_overide)
if augment_by and (
needs or needs_overide
or force) and PU_overflow[campaign][
'pending'] < PU_overflow[campaign]['max']:
PU_overflow[campaign]['pending'] += idled
print "raising overflow to", PU_overflow[campaign][
'pending'], "for", PU_overflow[campaign]['max']
## the step with an input ought to be the digi part : make this one go anywhere
modifications[wfo.name][task.pathName] = {
"AddWhitelist": augment_by,
"Running": running,
"Pending": idled,
"Priority": wfi.request['RequestPriority']
}
altered_tasks.add(task.pathName)
wfi.sendLog(
'equalizor',
'%s of %s is running %d and pending %d, taking action : AddWhitelist \n %s'
% (task_name, wfo.name, running, idled,
json.dumps(sorted(augment_by), indent=2)))
else:
print task_name, "of", wfo.name, "running", running, "and pending", idled
### overflow the skims back to multi-core
if campaign in ['Run2015D', 'Run2015C_25ns'
] and task.taskType == 'Skim':
original_swl = wfi.request['SiteWhitelist']
needs, task_name, running, idled = needs_action(wfi, task)
if (needs or needs_overide):
modifications[wfo.name][task.pathName] = {
'AddWhitelist': original_swl,
"Running": running,
"Pending": idled,
"Priority": wfi.request['RequestPriority']
}
altered_tasks.add(task.pathName)
wfi.sendLog(
'equalizor',
'%s of %s is running %d and pending %d, taking action : AddWhitelist \n %s'
% (task_name, wfo.name, running, idled,
json.dumps(sorted(original_swl), indent=2)))
if options.augment:
print sorted(wfi.request['SiteWhitelist']), i_task, use_HLT
### add the HLT at partner of CERN
if 'T2_CH_CERN' in wfi.request['SiteWhitelist'] and i_task in [
0, 1
] and use_HLT:
needs, task_name, running, idled = needs_action(wfi, task)
if options.augment: needs = True
needs = True
##needs = random.random()<0.40 remove the random, just add up to a limit
if (needs or needs_overide) and pending_HLT < max_HLT:
pending_HLT += idled
if task.pathName in modifications[
wfo.name] and 'AddWhitelist' in modifications[
wfo.name][task.pathName]:
modifications[wfo.name][task.pathName][
"AddWhitelist"].append("T2_CH_CERN_HLT")
print "\t", wfo.name, "adding addHLT up to", pending_HLT, "for", max_HLT
print task.pathName
## this Replace does not work at all for HLT
#elif task.pathName in modifications[wfo.name] and 'ReplaceSiteWhitelist' in modifications[wfo.name][task.pathName]:
#modifications[wfo.name][task.pathName]["ReplaceSiteWhitelist"].append( "T2_CH_CERN_HLT" )
#print "\t",wfo.name,"adding replace HLT up to",pending_HLT,"for",max_HLT
else:
modifications[wfo.name][task.pathName] = {
"AddWhitelist": ["T2_CH_CERN_HLT"],
"Priority": wfi.request['RequestPriority'],
"Running": running,
"Pending": idled
}
wfi.sendLog(
'equalizor',
'adding the HLT in whitelist of %s to %d for %d' %
(task.pathName, pending_HLT, max_HLT))
if i_task == 0 and not sec and use_T0:
needs, task_name, running, idled = needs_action(wfi, task)
if options.augment: needs = True
#needs = True
good_type = wfi.request['RequestType'] in [
'MonteCarlo', 'MonteCarloFromGEN'
]
read_lhe = ((not 'LheInputFiles' in wfi.request)
or bool(wfi.request['LheInputFiles']))
good_type &= not read_lhe
if not good_type and not options.augment: needs = False
##needs = random.random()<0.40 remove the random, just add up to a limit
if (needs or needs_overide):
pending_T0 += idled
if task.pathName in modifications[
wfo.name] and 'AddWhitelist' in modifications[
wfo.name][task.pathName]:
if not "T0_CH_CERN" in modifications[wfo.name][
task.pathName]["AddWhitelist"]:
modifications[wfo.name][task.pathName][
"AddWhitelist"].append("T0_CH_CERN")
wfi, sendLog(
'equalizor',
'adding the T0 for %s to %d for %d' %
(task.pathName, pending_T0, max_T0))
elif task.pathName in modifications[
wfo.
name] and 'ReplaceSiteWhitelist' in modifications[
wfo.name][task.pathName]:
if not "T0_CH_CERN" in modifications[wfo.name][
task.pathName]["ReplaceSiteWhitelist"]:
modifications[wfo.name][task.pathName][
"ReplaceSiteWhitelist"].append("T0_CH_CERN")
wfi, sendLog(
'equalizor',
'adding the T0 to replacement for %s to %d for %d'
% (task.pathName, pending_T0, max_T0))
else:
modifications[wfo.name][task.pathName] = {
"AddWhitelist": ["T0_CH_CERN"],
"Priority": wfi.request['RequestPriority'],
"Running": running,
"Pending": idled
}
wfi, sendLog(
'equalizor', 'adding the T0 for %s to %d for %d' %
(task.pathName, pending_T0, max_T0))
interface['modifications'].update(modifications)
### manage the number of core and job resizing
interface['cores'] = {
'T2_CH_CERN_HLT': {
'min': 4,
'max': 16
},
'default': {
'min': 1,
'max': 4
}
}
interface['resizes'] = ['RunIISpring16DR80']
### manage the modification of the memory and target time
interface['time'] = defaultdict(list)
interface['memory'] = defaultdict(list)
max_N_mem = 10
max_N_time = 10
## discretize the memory to 10 at most values
mems = set([o['memory'] for t, o in performance.items() if 'memory' in o])
times = set([o['time'] for t, o in performance.items() if 'time' in o])
if len(mems) > max_N_mem:
mem_step = int((max(mems) - min(mems)) / float(max_N_mem))
for t in performance:
if not 'memory' in performance[t]: continue
(m, r) = divmod(performance[t]['memory'], mem_step)
performance[t]['memory'] = (m + 1) * mem_step
if len(times) > max_N_time:
time_step = int((max(times) - min(times)) / float(max_N_time))
for t in performance:
if not 'time' in performance[t]: continue
(m, r) = divmod(performance[t]['time'], time_step)
performance[t]['time'] = (m + 1) * time_step
for t, o in performance.items():
if 'time' in o:
interface['time'][str(o['time'])].append(t)
if 'memory' in o:
interface['memory'][str(o['memory'])].append(t)
## close and save
close(interface)
def assignor(url, specific=None, talk=True, options=None):
if userLock() and not options.manual: return
mlock = moduleLock()
if mlock() and not options.manual: return
if not componentInfo().check() and not options.manual: return
UC = unifiedConfiguration()
CI = campaignInfo()
SI = siteInfo()
SI = global_SI()
###NLI = newLockInfo()
###if not NLI.free() and not options.go: return
LI = lockInfo()
#if not LI.free() and not options.go and not options.manual: return
n_assigned = 0
n_stalled = 0
wfos = []
fetch_from = []
if specific or options.early:
fetch_from.extend(['considered', 'staging'])
if specific:
fetch_from.extend(['considered-tried'])
if options.early:
print "Option Early is on"
fetch_from.extend(['staged'])
if options.from_status:
fetch_from = options.from_status.split(',')
print "Overriding to read from", fetch_from
for status in fetch_from:
print "getting wf in", status
wfos.extend(
session.query(Workflow).filter(Workflow.status == status).all())
print len(wfos)
## in case of partial, go for fetching a list from json ?
#if options.partial and not specific:
# pass
aaa_mapping = json.loads(eosRead('%s/equalizor.json' %
monitor_pub_dir))['mapping']
all_stuck = set()
all_stuck.update(
json.loads(eosRead('%s/stuck_transfers.json' % monitor_pub_dir)))
max_per_round = UC.get('max_per_round').get('assignor', None)
max_cpuh_block = UC.get('max_cpuh_block')
# Temporarily switch off prioritization
random.shuffle(wfos)
##order by priority instead of random
"""
if options.early:
cache = sorted(getWorkflows(url, 'assignment-approved', details=True), key = lambda r : r['RequestPriority'])
cache = [r['RequestName'] for r in cache]
def rank( wfn ):
return cache.index( wfn ) if wfn in cache else 0
wfos = sorted(wfos, key = lambda wfo : rank( wfo.name ),reverse=True)
print "10 first",[wfo.name for wfo in wfos[:10]]
print "10 last",[wfo.name for wfo in wfos[-10:]]
else:
random.shuffle( wfos )
"""
for wfo in wfos:
if options.limit and (n_stalled + n_assigned) > options.limit:
break
if max_per_round and (n_stalled + n_assigned) > max_per_round:
break
if specific:
if not any(map(lambda sp: sp in wfo.name, specific.split(','))):
continue
#if not specific in wfo.name: continue
if not options.manual and 'rucio' in (wfo.name).lower(): continue
print "\n\n"
wfh = workflowInfo(url, wfo.name)
if wfh.request['RequestStatus'] in [
'rejected', 'aborted', 'aborted-completed', 'aborted-archived',
'rejected-archived'
] and wfh.isRelval():
wfo.status = 'forget'
session.commit()
n_stalled += 1
continue
if options.priority and int(
wfh.request['RequestPriority']) < options.priority:
continue
options_text = ""
if options.early: options_text += ", early option is ON"
wfh.sendLog('assignor',
"%s to be assigned %s" % (wfo.name, options_text))
## the site whitelist takes into account siteInfo, campaignInfo, memory and cores
(lheinput, primary, parent, secondary, sites_allowed,
sites_not_allowed) = wfh.getSiteWhiteList()
output_tiers = list(
set([o.split('/')[-1] for o in wfh.request['OutputDatasets']]))
if not output_tiers:
n_stalled += 1
wfh.sendLog('assignor', 'There is no output at all')
sendLog('assignor',
'Workflow %s has no output at all' % (wfo.name),
level='critical')
continue
is_stuck = (all_stuck & primary)
if is_stuck:
wfh.sendLog('assignor',
"%s are stuck input" % (','.join(is_stuck)))
## check if by configuration we gave it a GO
no_go = False
if not wfh.go(log=True) and not options.go:
no_go = True
allowed_secondary = {}
assign_parameters = {}
check_secondary = (not wfh.isRelval())
for campaign in wfh.getCampaigns():
if campaign in CI.campaigns:
assign_parameters.update(CI.campaigns[campaign])
if campaign in CI.campaigns and 'secondaries' in CI.campaigns[
campaign]:
if CI.campaigns[campaign]['secondaries']:
allowed_secondary.update(
CI.campaigns[campaign]['secondaries'])
check_secondary = True
if campaign in CI.campaigns and 'banned_tier' in CI.campaigns[
campaign]:
banned_tier = list(
set(CI.campaigns[campaign]['banned_tier'])
& set(output_tiers))
if banned_tier:
no_go = True
wfh.sendLog(
'assignor', 'These data tiers %s are not allowed' %
(','.join(banned_tier)))
sendLog('assignor',
'These data tiers %s are not allowed' %
(','.join(banned_tier)),
level='critical')
if secondary and check_secondary:
if (set(secondary) & set(allowed_secondary.keys()) !=
set(secondary)):
msg = '%s is not an allowed secondary' % (
', '.join(set(secondary) - set(allowed_secondary.keys())))
wfh.sendLog('assignor', msg)
critical_msg = msg + '\nWorkflow URL: https://dmytro.web.cern.ch/dmytro/cmsprodmon/workflows.php?prep_id=task_{}'.format(
wfh.getPrepIDs()[0])
sendLog('assignor', critical_msg, level='critical')
if not options.go:
no_go = True
## then get whether there is something more to be done by secondary
for sec in secondary:
if sec in allowed_secondary: # and 'parameters' in allowed_secondary[sec]:
assign_parameters.update(allowed_secondary[sec])
if no_go:
n_stalled += 1
## make a very loud noise if >100k priority stalled
continue
## check on current status for by-passed assignment
if wfh.request['RequestStatus'] != 'assignment-approved':
if not options.test:
wfh.sendLog('assignor',
"setting %s away and skipping" % wfo.name)
## the module picking up from away will do what is necessary of it
wfo.wm_status = wfh.request['RequestStatus']
wfo.status = 'away'
session.commit()
continue
else:
print wfo.name, wfh.request['RequestStatus']
## retrieve from the schema, dbs and reqMgr what should be the next version
version = wfh.getNextVersion()
if not version:
if options and options.ProcessingVersion:
version = options.ProcessingVersion
else:
wfh.sendLog('assignor', "cannot decide on version number")
n_stalled += 1
wfo.status = 'trouble'
session.commit()
continue
wfh.sendLog('assignor', "Site white list %s" % sorted(sites_allowed))
blocks = wfh.getBlocks()
if blocks:
wfh.sendLog(
'assignor',
"Needs {} blocks in input {}".format(len(blocks),
'\n'.join(blocks)))
wfh.sendLog('assignor', "Allowed %s" % sorted(sites_allowed))
primary_aaa = options.primary_aaa
secondary_aaa = options.secondary_aaa
if 'Campaign' in wfh.request and wfh.request[
'Campaign'] in CI.campaigns:
assign_parameters.update(CI.campaigns[wfh.request['Campaign']])
if 'primary_AAA' in assign_parameters and primary:
primary_aaa = primary_aaa or assign_parameters['primary_AAA']
if 'secondary_AAA' in assign_parameters:
secondary_aaa = secondary_aaa or assign_parameters['secondary_AAA']
wfh.sendLog(
'assignor',
"Initial values for primary_AAA=%s and secondary_AAA=%s" %
(primary_aaa, secondary_aaa))
if primary_aaa:
if "T2_CH_CERN_HLT" in sites_allowed:
sites_allowed.remove("T2_CH_CERN_HLT")
if "T2_CH_CERN_HLT" not in sites_not_allowed:
sites_not_allowed.append("T2_CH_CERN_HLT")
## keep track of this, after secondary input location restriction : that's how you want to operate it
initial_sites_allowed = copy.deepcopy(sites_allowed)
set_lfn = '/store/mc' ## by default
for prim in list(primary):
set_lfn = getLFNbase(prim)
## if they are requested for processing, they should bbe all closed already
# FIXME: remove this closeAllBlocks
#closeAllBlocks(url, prim, blocks)
## should be 2 but for the time-being let's lower it to get things going
_copies_wanted, cpuh = wfh.getNCopies()
wfh.sendLog('assignor', "we need %s CPUh" % cpuh)
if cpuh > max_cpuh_block and not options.go:
#sendEmail('large workflow','that wf %s has a large number of CPUh %s, not assigning, please check the logs'%(wfo.name, cpuh))#,destination=['*****@*****.**'])
sendLog(
'assignor',
'%s requires a large numbr of CPUh %s , not assigning, please check with requester'
% (wfo.name, cpuh),
level='critical')
wfh.sendLog(
'assignor',
"Requiring a large number of CPUh %s, not assigning" % cpuh)
continue
## should also check on number of sources, if large enough, we should be able to overflow most, efficiently
## default back to white list to original white list with any data
wfh.sendLog('assignor', "Allowed sites :%s" % sorted(sites_allowed))
# TODO Alan on 1/april/2020: keep the AAA functionality
if primary_aaa:
## remove the sites not reachable localy if not in having the data
if not sites_allowed:
wfh.sendLog('assignor',
"Overiding the primary on AAA setting to Off")
primary_aaa = False
else:
aaa_grid = set(sites_allowed)
for site in list(aaa_grid):
aaa_grid.update(aaa_mapping.get(site, []))
sites_allowed = list(set(initial_sites_allowed) & aaa_grid)
wfh.sendLog(
'assignor', "Selected to read primary through xrootd %s" %
sorted(sites_allowed))
isStoreResults = ('StoreResults' == wfh.request.setdefault(
'RequestType', None))
if isStoreResults:
if 'MergedLFNBase' in wfh.request:
set_lfn = wfh.request['MergedLFNBase']
else:
n_stalled += 1
wfh.sendLog(
'assignor',
"Cannot assign StoreResults request because MergedLFN is missing"
)
sendLog(
'assignor',
'Cannot assign StoreResults request because MergedLFN is missing',
level='critical')
continue
if not primary_aaa:
if isStoreResults:
## if we are dealing with a StoreResults request, we don't need to check dataset availability and
## should use the SiteWhiteList set in the original request
if 'SiteWhitelist' in wfh.request:
sites_allowed = wfh.request['SiteWhitelist']
else:
wfh.sendLog(
'assignor',
"Cannot assign StoreResults request because SiteWhitelist is missing"
)
sendLog(
'assignor',
'Cannot assign StoreResults request because SiteWhitelist is missing',
level='critical')
n_stalled += 1
continue
wfh.sendLog('assignor',
"Selected for any data %s" % sorted(sites_allowed))
#if not len(sites_allowed):
# if not options.early:
# wfh.sendLog('assignor',"cannot be assign with no matched sites")
# sendLog('assignor','%s has no whitelist'% wfo.name, level='critical')
# n_stalled+=1
# continue
if not len(sites_allowed) and not options.SiteWhitelist:
if not options.early:
wfh.sendLog('assignor',
"cannot be assign with no matched sites")
sendLog('assignor',
'%s has no whitelist' % wfo.name,
level='critical')
n_stalled += 1
continue
t1t2_only = [
ce for ce in sites_allowed
if [ce.startswith('T1') or ce.startswith('T2')]
]
if t1t2_only:
# try to pick from T1T2 only first
sites_out = [SI.pick_dSE([SI.CE_to_SE(ce) for ce in t1t2_only])]
# then pick any otherwise
else:
sites_out = [
SI.pick_dSE([SI.CE_to_SE(ce) for ce in sites_allowed])
]
print "available=", SI.disk[sites_out[0]]
wfh.sendLog('assignor', "Placing the output on %s" % sites_out)
parameters = {
'SiteWhitelist': sites_allowed,
'SiteBlacklist': sites_not_allowed,
'NonCustodialSites': sites_out,
'AutoApproveSubscriptionSites': list(set(sites_out)),
'AcquisitionEra': wfh.acquisitionEra(),
'ProcessingString': wfh.processingString(),
'MergedLFNBase': set_lfn,
'ProcessingVersion': version,
}
if primary_aaa:
parameters['TrustSitelists'] = True
wfh.sendLog(
'assignor',
"Reading primary through xrootd at %s" % sorted(sites_allowed))
if secondary_aaa:
# Do not set TrustPUSitelist to True if there is no secondary
if secondary:
parameters['TrustPUSitelists'] = True
wfh.sendLog(
'assignor', "Reading secondary through xrootd at %s" %
sorted(sites_allowed))
## plain assignment here
team = 'production'
if os.getenv('UNIFIED_TEAM'): team = os.getenv('UNIFIED_TEAM')
if options and options.team:
team = options.team
parameters['Team'] = team
if lheinput:
## throttle reading LHE article
wfh.sendLog('assignor',
'Setting the number of events per job to 500k max')
parameters['EventsPerJob'] = 500000
def pick_options(options, parameters):
##parse options entered in command line if any
if options:
for key in reqMgrClient.assignWorkflow.keys:
v = getattr(options, key)
if v != None:
if type(v) == str and ',' in v:
parameters[key] = filter(None, v.split(','))
else:
parameters[key] = v
def pick_campaign(assign_parameters, parameters):
## pick up campaign specific assignment parameters
parameters.update(assign_parameters.get('parameters', {}))
if options.force_options:
pick_campaign(assign_parameters, parameters)
pick_options(options, parameters)
else:
## campaign parameters update last
pick_options(options, parameters)
pick_campaign(assign_parameters, parameters)
if not options.test:
parameters['execute'] = True
hold_split, split_check = wfh.checkSplitting()
if hold_split and not options.go:
if split_check:
wfh.sendLog(
'assignor', 'Holding on to the change in splitting %s' %
('\n\n'.join([str(i) for i in split_check])))
else:
wfh.sendLog('assignor', 'Change of splitting is on hold')
n_stalled += 1
continue
if split_check == None or split_check == False:
n_stalled += 1
continue
elif split_check:
## operate all recommended changes
reqMgrClient.setWorkflowSplitting(url, wfo.name, split_check)
wfh.sendLog(
'assignor', 'Applying the change in splitting %s' %
('\n\n'.join([str(i) for i in split_check])))
split_check = True ## bypass completely and use the above
# Handle run-dependent MC
pstring = wfh.processingString()
if 'PU_RD' in pstring:
numEvents = wfh.getRequestNumEvents()
eventsPerLumi = [getDatasetEventsPerLumi(prim) for prim in primary]
eventsPerLumi = sum(eventsPerLumi) / float(len(eventsPerLumi))
reqJobs = 500
if 'PU_RD2' in pstring:
reqJobs = 2000
eventsPerJob = int(numEvents / (reqJobs * 1.4))
lumisPerJob = int(eventsPerJob / eventsPerLumi)
if lumisPerJob == 0:
#sendEmail("issue with event splitting for run-dependent MC","%s needs to be split by event with %s per job"%(wfo.name, eventsPerJob))
sendLog('assignor',
"%s needs to be split by event with %s per job" %
(wfo.name, eventsPerJob),
level='critical')
wfh.sendLog(
'assignor',
"%s needs to be split by event with %s per job" %
(wfo.name, eventsPerJob))
parameters['EventsPerJob'] = eventsPerJob
else:
spl = wfh.getSplittings()[0]
# FIXME: decide which of the lines below needs to remain...
eventsPerJobEstimated = spl[
'events_per_job'] if 'events_per_job' in spl else None
eventsPerJobEstimated = spl[
'avg_events_per_job'] if 'avg_events_per_job' in spl else None
if eventsPerJobEstimated and eventsPerJobEstimated > eventsPerJob:
#sendEmail("setting lumi splitting for run-dependent MC","%s was assigned with %s lumis/job"%( wfo.name, lumisPerJob))
sendLog('assignor',
"%s was assigned with %s lumis/job" %
(wfo.name, lumisPerJob),
level='critical')
wfh.sendLog(
'assignor', "%s was assigned with %s lumis/job" %
(wfo.name, lumisPerJob))
parameters['LumisPerJob'] = lumisPerJob
else:
#sendEmail("leaving splitting untouched for PU_RD*","please check on "+wfo.name)
sendLog(
'assignor',
"leaving splitting untouched for %s, please check on %s"
% (pstring, wfo.name),
level='critical')
wfh.sendLog(
'assignor',
"leaving splitting untouched for PU_RD*, please check."
)
## make sure to autoapprove all NonCustodialSites
parameters['AutoApproveSubscriptionSites'] = list(
set(parameters['NonCustodialSites'] +
parameters.get('AutoApproveSubscriptionSites', [])))
result = reqMgrClient.assignWorkflow(
url, wfo.name, None,
parameters) ## team is not relevant anymore here
# set status
if not options.test:
if result:
wfo.status = 'away'
session.commit()
n_assigned += 1
wfh.sendLog(
'assignor', "Properly assigned\n%s" %
(json.dumps(parameters, indent=2)))
if wfh.producePremix() and (not wfh.isRelval()):
title = "Heavy workflow assigned to {}".format(
parameters['SiteWhitelist'])
body = "Workflow name: {}".format(
wfh.request['RequestName'])
body += "\nOutput dataset(s): {}".format(
wfh.request['OutputDatasets'])
body += "\nAssigned to: {}".format(
parameters['SiteWhitelist'])
sendEmail(
title,
body,
destination=[
'*****@*****.**'
])
try:
## refetch information and lock output
new_wfi = workflowInfo(url, wfo.name)
(_, prim, _, sec) = new_wfi.getIO()
for secure in list(prim) + list(
sec) + new_wfi.request['OutputDatasets']:
## lock all outputs
LI.lock(secure, reason='assigning')
except Exception as e:
print "fail in locking output"
print str(e)
sendEmail("failed locking of output", str(e))
else:
wfh.sendLog(
'assignor',
"Failed to assign %s.\n%s \n Please check the logs" %
(wfo.name, reqMgrClient.assignWorkflow.errorMessage))
sendLog('assignor',
"Failed to assign %s.\n%s \n Please check the logs" %
(wfo.name, reqMgrClient.assignWorkflow.errorMessage),
level='critical')
print "ERROR could not assign", wfo.name
else:
pass
print "Assignment summary:"
sendLog('assignor', "Assigned %d Stalled %s" % (n_assigned, n_stalled))
if n_stalled and not options.go and not options.early:
sendLog('assignor',
"%s workflows cannot be assigned. Please take a look" %
(n_stalled),
level='critical')
def completor(url, specific):
mlock = moduleLock(silent=True)
if mlock(): return
use_mcm = True
up = componentInfo(soft=['mcm','wtc','jira'])
if not up.check(): return
use_mcm = up.status['mcm']
if use_mcm:
mcm = McMClient(dev=False)
safe_mode = False
CI = campaignInfo()
SI = siteInfo()
UC = unifiedConfiguration()
JC = JIRAClient() if up.status.get('jira',False) else None
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
for itry in range(5):
try:
all_stuck.update( json.loads( eosRead('%s/stuck_transfers.json'%monitor_pub_dir)))
break
except:
time.sleep(2)
for itry in range(5):
try:
## take into account the block that needed to be repositioned recently
all_stuck.update( [b.split('#')[0] for b in json.loads( eosRead('%s/missing_blocks.json'%monitor_dir)) ] )
break
except:
time.sleep(2)
## take into account all stuck block and dataset from transfer team
all_stuck.update( getAllStuckDataset())
good_fractions = {}
overdoing_fractions = {}
truncate_fractions = {}
timeout = {}
campaign_injection_delay = {}
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']
if 'injection-delay' in CI.campaigns[c]:
campaign_injection_delay[c] = CI.campaigns[c]['injection-delay']
if 'overdoing-complete' in CI.campaigns[c]:
overdoing_fractions[c] = CI.campaigns[c]['overdoing-complete']
long_lasting = {}
WI = wtcInfo()
overrides = WI.getForce()
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")
max_priority = UC.get("max_tail_priority")
injection_delay_threshold = UC.get("injection_delay_threshold")
injection_delay_priority = UC.get("injection_delay_priority")
delay_priority_increase = UC.get("delay_priority_increase")
default_fraction_overdoing = UC.get('default_fraction_overdoing')
set_force_complete = set()
# priority and time above which to fire a JIRA
jira_priority_and_delays = { 110000 : 21,
90000 : 28,
# 80000 : 60,
#0 : 90
}
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
campaigns = wfi.getCampaigns()
#if not any([c in good_fractions.keys() for c in campaigns]): skip=True
#if not any([c in truncate_fractions.keys() for c in campaigns]): skip=True
for user,spec in overrides.items():
if not spec: continue
spec = filter(None, spec)
if not wfi.request['RequestStatus'] in ['force-complete', 'completed']:
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
## until we can map the output to task ...
output_per_task = wfi.getOutputPerTask() ## can use that one, and follow mapping
good_fraction_per_out = {}
good_fraction_nodelay_per_out = {}
truncate_fraction_per_out = {}
#allowed_delay_per_out = {}
for task,outs in output_per_task.items():
task_campaign = wfi.getCampaignPerTask( task )
for out in outs:
good_fraction_per_out[out] = good_fractions.get(task_campaign,1000.)
good_fraction_nodelay_per_out[out] = overdoing_fractions.get(task_campaign,default_fraction_overdoing)
truncate_fraction_per_out[out] = truncate_fractions.get(task_campaign,1000.)
#allowed_delay_per_out[out] = timeout.get(task_campaign, 14)
#print "force at", json.dumps( good_fraction_per_out, indent=2)
#print "truncate at",json.dumps( truncate_fraction_per_out, indent=2)
now = time.mktime(time.gmtime()) / (60*60*24.)
priority_log = filter(lambda change: change['Priority'] == priority,wfi.request.get('PriorityTransition',[]))
if not priority_log:
print "\tHas no priority log"
priority_delay = 0
else:
then = max([change['UpdateTime'] for change in priority_log]) / (60.*60.*24.)
priority_delay = now - then ## in days
print "priority was set to",priority,priority_delay,"[days] ago"
running_log = filter(lambda change : change["Status"] in ["running-open","running-closed"],wfi.request['RequestTransition'])
if not running_log:
print "\tHas no running log"
delay = 0
else:
then = max([change['UpdateTime'] for change in running_log]) / (60.*60.*24.)
delay = now - then ## in days
#further check on delays
cpuh = wfi.getComputingTime(unit='d')
wfi.sendLog('completor',"Requires %.1f [CPUd], running since %.1f [day], last priority change to %d done %.1f [day] ago"%( cpuh, delay, priority, priority_delay))
if priority_delay!=0 and priority_delay < delay:
## regardless when it started running, set the delay to when priority was changed last
delay = priority_delay
## this is supposed to be the very initial request date, inherited from clones
injection_delay = None
original = wfi
if 'OriginalRequestName' in original.request:
## go up the clone chain
original = workflowInfo(url, original.request['OriginalRequestName'])
injected_log = filter(lambda change : change["Status"] in ["assignment-approved"],original.request['RequestTransition'])
if injected_log:
injected_on = injected_log[-1]['UpdateTime'] / (60.*60.*24.)
injection_delay = now - injected_on
delay_for_priority_increase = injection_delay
#delay_for_priority_increase = delay
(w,d) = divmod(delay, 7 )
print "\t"*int(w)+"Running since",delay,"[days] priority=",priority
pop_a_jira = False
ping_on_jira = 7 *(24*60*60) # 7 days
for jp,jd in jira_priority_and_delays.items():
if priority >= jp and delay >= jd: pop_a_jira = True
if pop_a_jira and JC:
j,reopened,just_created = JC.create_or_last( prepid = wfi.request['PrepID'],
priority = wfi.request['RequestPriority'],
label = 'Late',
reopen = True)
last_time = JC.last_time( j )
since_last_ping = time.mktime(time.gmtime()) - last_time
if since_last_ping > ping_on_jira or just_created:
j_comment = "Running since %.1f [days] at priority %d"%( delay, priority)
JC.comment(j.key, j_comment)
if delay_for_priority_increase!=None and delay_for_priority_increase > injection_delay_threshold and priority >= injection_delay_priority:
quantized = 5000 ## quantize priority
tail_cutting_priority = wfi.request['InitialPriority']+ int((delay_priority_increase * (delay_for_priority_increase - injection_delay_threshold) / 7) / quantized) * quantized
tail_cutting_priority += 101 ## to signal it is from this mechanism
tail_cutting_priority = min(400000, tail_cutting_priority) ## never go above 400k priority
tail_cutting_priority = max(tail_cutting_priority, priority) ## never go below the current value
if priority < tail_cutting_priority:
if max_priority:
sendLog('completor',"%s Injected since %s [days] priority=%s, increasing to %s"%(wfo.name,delay_for_priority_increase,priority, tail_cutting_priority), level='critical')
wfi.sendLog('completor','bumping priority to %d for being injected since %s'%( tail_cutting_priority, delay_for_priority_increase))
reqMgrClient.changePriorityWorkflow(url, wfo.name, tail_cutting_priority)
max_priority-=1
else:
sendLog('completor',"%s Injected since %s [days] priority=%s, would like to increase to %s"%(wfo.name,delay_for_priority_increase,priority, tail_cutting_priority), level='critical')
wfi.sendLog('completor','would like to bump priority to %d for being injected since %s'%( tail_cutting_priority, delay_for_priority_increase))
print "Could be changing the priority to higher value, but too many already were done"
_,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 = max([timeout.get(c,14) for c in campaigns])
monitor_delay = min(monitor_delay, allowed_delay)
### just skip if too early, just for the sake of not computing the completion fraction just now.
# maybe this is fast enough that we can do it for all
if delay <= monitor_delay:
print "not enough time has passed yet"
continue
long_lasting[wfo.name] = { "delay" : delay,
"injection_delay" : injection_delay }
percent_completions = wfi.getCompletionFraction(caller='completor')
if not percent_completions:
sendLog('completor','%s has no output at all'% wfo.name, level='critical')
continue
is_over_allowed_delay = (all([percent_completions[out] >= good_fraction_per_out.get(out,1000.) for out in percent_completions]) and delay >= allowed_delay)
is_over_truncation_delay = (is_stuck and (all([percent_completions[out] >= truncate_fraction_per_out.get(out,1000.) for out in percent_completions])) and delay >= allowed_delay)
is_over_completion = (all([percent_completions[out] >= good_fraction_nodelay_per_out.get(out,1000.) for out in percent_completions]))
if is_over_completion:
wfi.sendLog('completor', "all is over completed %s\n %s"%( json.dumps( good_fraction_nodelay_per_out, indent=2 ),
json.dumps( percent_completions, indent=2 )
))
elif is_over_allowed_delay:
wfi.sendLog('completor', "all is above %s \n%s"%( json.dumps(good_fraction_per_out, indent=2 ),
json.dumps( percent_completions, indent=2 )
))
elif is_over_truncation_delay:
wfi.sendLog('completor', "all is above %s truncation level, and the input is stuck\n%s"%( json.dumps(truncate_fraction_per_out, indent=2 ),
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 \ncomplete at %s \n truncate at %s \nRunning %s"%(json.dumps( percent_completions, indent=2),
json.dumps(good_fraction_per_out, indent=2),
json.dumps( truncate_fraction_per_out, indent=2),
json.dumps( long_lasting[wfo.name]['agents'], indent=2) ))
continue
#for output in percent_completions:
# completions[output]['injected'] = then
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
## find ACDCs that might be running
if max_force>0:
print "going for force-complete of",wfo.name
if not safe_mode:
forceComplete(url, wfi )
set_force_complete.add( 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:
sendEmail('completor', 'The workflow %s is ready for force complete, but completor is in safe mode'%wfo.name)
else:
wfi.sendLog('completor',"too many completion this round, cannot force complete")
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 ))
eosFile('%s/longlasting.json'%monitor_dir,'w').write( json.dumps( long_lasting, indent=2 )).close()
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 stagor(url, specific=None, options=None):
if not componentInfo().check(): return
SI = siteInfo()
CI = campaignInfo()
UC = unifiedConfiguration()
TS = transferStatuses()
cached_transfer_statuses = TS.content()
transfer_statuses = {}
done_by_wf_id = {}
done_by_input = {}
completion_by_input = {}
good_enough = 100.0
lost_blocks = json.loads(
eosRead('%s/lost_blocks_datasets.json' % monitor_dir))
lost_files = json.loads(
eosRead('%s/lost_files_datasets.json' % monitor_dir))
known_lost_blocks = {}
known_lost_files = {}
for dataset in set(lost_blocks.keys() + lost_files.keys()):
b, f = findLostBlocksFiles(url, dataset)
if dataset in lost_blocks and not b:
print dataset, "has no really lost blocks"
else:
known_lost_blocks[dataset] = [i['name'] for i in b]
if dataset in lost_files and not f:
print dataset, "has no really lost files"
else:
known_lost_files[dataset] = [i['name'] for i in f]
def time_point(label="", sub_lap=False):
now = time.mktime(time.gmtime())
nows = time.asctime(time.gmtime())
print "Time check (%s) point at : %s" % (label, nows)
print "Since start: %s [s]" % (now - time_point.start)
if sub_lap:
print "Sub Lap : %s [s]" % (now - time_point.sub_lap)
time_point.sub_lap = now
else:
print "Lap : %s [s]" % (now - time_point.lap)
time_point.lap = now
time_point.sub_lap = now
time_point.sub_lap = time_point.lap = time_point.start = time.mktime(
time.gmtime())
time_point("Check cached transfer")
## collect all datasets that are needed for wf in staging, correcting the status of those that are not really in staging
wfois = []
needs = defaultdict(list)
needs_by_priority = defaultdict(list)
for wfo in session.query(Workflow).filter(
Workflow.status == 'staging').all():
wfi = workflowInfo(url, wfo.name)
if wfi.request['RequestStatus'] in [
'running-open', 'running-closed', 'completed', 'assigned',
'acquired'
]:
wfi.sendLog('stagor',
"is in status %s" % wfi.request['RequestStatus'])
wfo.status = 'away'
session.commit()
continue
if not wfi.request['RequestStatus'] in ['assignment-approved']:
## should be setting 'away' too
## that usually happens for relvals
if wfi.request['RequestStatus'] in [
'rejected', 'aborted', 'aborted-completed',
'aborted-archived', 'rejected-archived'
] and wfi.isRelval():
wfo.status = 'forget'
session.commit()
continue
else:
print wfo.name, "is", wfi.request['RequestStatus']
#sendEmail("wrong status in staging. debug","%s is in %s, should set away."%(wfo.name,wfi.request['RequestStatus']))
sendLog("stagor",
"%s is in %s, set away" %
(wfo.name, wfi.request['RequestStatus']),
level='critical')
wfo.status = 'away'
session.commit()
continue
wfois.append((wfo, wfi))
_, primaries, _, secondaries = wfi.getIO()
for dataset in list(primaries) + list(secondaries):
needs[wfo.name].append(dataset)
done_by_input[dataset] = {}
completion_by_input[dataset] = {}
needs_by_priority[wfi.request['RequestPriority']].append(dataset)
wfi.sendLog('stagor', '%s needs %s' % (wfo.name, dataset))
time_point("Check staging workflows")
open('%s/dataset_requirements.json' % monitor_dir,
'w').write(json.dumps(needs, indent=2))
for prio in needs_by_priority:
needs_by_priority[prio] = list(set(needs_by_priority[prio]))
open('%s/dataset_priorities.json' % monitor_dir,
'w').write(json.dumps(needs_by_priority, indent=2))
dataset_endpoints = defaultdict(set)
endpoint_in_downtime = defaultdict(set)
#endpoint_completed = defaultdict(set)
endpoint_incompleted = defaultdict(set)
#endpoint = defaultdict(set)
send_back_to_considered = set()
## first check if anything is inactive
all_actives = set([
transfer.phedexid for transfer in session.query(TransferImp).filter(
TransferImp.active).all()
])
for active_phedexid in all_actives:
skip = True
transfers_phedexid = session.query(TransferImp).filter(
TransferImp.phedexid == active_phedexid).all()
for imp in transfers_phedexid:
if imp.workflow.status == 'staging':
skip = False
sendLog(
'stagor', "\t%s is staging for %s" %
(imp.phedexid, imp.workflow.name))
if skip:
sendLog('stagor', "setting %s inactive" % active_phedexid)
for imp in transfers_phedexid:
imp.active = False
session.commit()
all_actives = sorted(
set([
transfer.phedexid for transfer in session.query(
TransferImp).filter(TransferImp.active).all()
]))
for phedexid in all_actives:
if specific: continue
## check on transfer completion
not_cached = False
if phedexid in cached_transfer_statuses:
### use a cache for transfer that already looked done
sendLog('stagor', "read %s from cache" % phedexid)
checks = cached_transfer_statuses[phedexid]
else:
## I actually would like to avoid that all I can
sendLog('stagor',
'Performing spurious transfer check on %s' % phedexid,
level='critical')
checks = checkTransferStatus(url, phedexid, nocollapse=True)
try:
print json.dumps(checks, indent=2)
except:
print checks
if not checks:
## this is going to bias quite heavily the rest of the code. we should abort here
#sendLog('stagor','Ending stagor because of skewed input from checkTransferStatus', level='critical')
#return False
sendLog(
'stagor',
'Stagor has got a skewed input from checkTransferStatus',
level='critical')
checks = {}
pass
else:
TS.add(phedexid, checks)
time_point("Check transfer status %s" % phedexid, sub_lap=True)
if not specific:
for dsname in checks:
if not dsname in done_by_input: done_by_input[dsname] = {}
if not dsname in completion_by_input:
completion_by_input[dsname] = {}
done_by_input[dsname][phedexid] = all(
map(lambda i: i >= good_enough, checks[dsname].values()))
completion_by_input[dsname][phedexid] = checks[dsname].values()
if checks:
sendLog(
'stagor', "Checks for %s are %s" %
(phedexid, [node.values() for node in checks.values()]))
done = all(
map(
lambda i: i >= good_enough,
list(
itertools.chain.from_iterable(
[node.values() for node in checks.values()]))))
else:
## it is empty, is that a sign that all is done and away ?
if not_cached:
print "Transfer status was not cached"
else:
print "ERROR with the scubscriptions API of ", phedexid
print "Most likely something else is overiding the transfer request. Need to work on finding the replacement automatically, if the replacement exists"
done = False
transfers_phedexid = session.query(TransferImp).filter(
TransferImp.phedexid == phedexid).all()
for imp in transfers_phedexid:
tr_wf = imp.workflow
if tr_wf: # and tr_wf.status == 'staging':
if not tr_wf.id in done_by_wf_id: done_by_wf_id[tr_wf.id] = {}
done_by_wf_id[tr_wf.id][phedexid] = done
if done:
imp.active = False
session.commit()
for ds in checks:
for s, v in checks[ds].items():
dataset_endpoints[ds].add(s)
if done:
sendLog('stagor', "%s is done" % phedexid)
TS.add(phedexid, checks)
else:
sendLog(
'stagor',
"%s is not finished %s" % (phedexid, pprint.pformat(checks)))
##pprint.pprint( checks )
## check if the destination is in down-time
for ds in checks:
sites_incomplete = [
SI.SE_to_CE(s) for s, v in checks[ds].items()
if v < good_enough
]
sites_incomplete_down = [
s for s in sites_incomplete if not s in SI.sites_ready
]
## no space means no transfer should go there : NO, it does not work in the long run
#sites_incomplete_down = [SI.SE_to_CE(s) for s,v in checks[ds].items() if (v<good_enough and (SI.disk[s]==0 or (not SI.SE_to_CE(s) in SI.sites_ready)))]
if sites_incomplete_down:
sendLog(
'stagor',
"%s are in downtime, while waiting for %s to get there"
% (",".join(sites_incomplete_down), ds))
endpoint_in_downtime[ds].update(sites_incomplete_down)
if sites_incomplete:
endpoint_incompleted[ds].update(sites_incomplete)
time_point("Check on-going transfers")
print "End points"
for k in dataset_endpoints:
dataset_endpoints[k] = list(dataset_endpoints[k])
print json.dumps(dataset_endpoints, indent=2)
print "End point in down time"
for k in endpoint_in_downtime:
endpoint_in_downtime[k] = list(endpoint_in_downtime[k])
print json.dumps(endpoint_in_downtime, indent=2)
print "End point incomplete in down time"
for k in endpoint_incompleted:
endpoint_incompleted[k] = list(endpoint_incompleted[k])
print json.dumps(endpoint_incompleted, indent=2)
#open('%s/transfer_statuses.json'%monitor_dir,'w').write( json.dumps( transfer_statuses, indent=2))
eosFile('%s/transfer_statuses.json' % monitor_dir,
'w').write(json.dumps(TS.content(), indent=2)).close()
eosFile('%s/dataset_endpoints.json' % monitor_dir,
'w').write(json.dumps(dataset_endpoints, indent=2)).close()
already_stuck = json.loads(
eosRead('%s/stuck_transfers.json' % monitor_pub_dir)).keys()
already_stuck.extend(getAllStuckDataset())
missing_in_action = defaultdict(list)
print "-" * 10, "Checking on workflows in staging", "-" * 10
#forget_about = ['/MinBias_TuneCUETP8M1_13TeV-pythia8/RunIISummer15GS-MCRUN2_71_V1-v2/GEN-SIM']
#for what in forget_about:
# if not done_by_input[what]:
# done_by_input[what] = {'fake':True}
## come back to workflows and check if they can go
available_cache = defaultdict(lambda: defaultdict(float))
presence_cache = defaultdict(dict)
time_point("Preparing for more")
for wfo, wfi in wfois:
print "#" * 30
time_point("Forward checking %s" % wfo.name, sub_lap=True)
## the site white list takes site, campaign, memory and core information
(_, primaries, _, secondaries,
sites_allowed) = wfi.getSiteWhiteList(verbose=False)
se_allowed = [SI.CE_to_SE(site) for site in sites_allowed]
se_allowed.sort()
se_allowed_key = ','.join(se_allowed)
readys = {}
for need in list(primaries) + list(secondaries):
if not need in done_by_input:
wfi.sendLog('stagor', "missing transfer report for %s" % need)
readys[need] = False
## should warn someone about this !!!
## it cannot happen, by construction
sendEmail('missing transfer report',
'%s does not have a transfer report' % (need))
continue
if not done_by_input[need] and need in list(secondaries):
wfi.sendLog(
'stagor',
"assuming it is OK for secondary %s to have no attached transfers"
% need)
readys[need] = True
done_by_input[need] = {"fake": True}
continue
if len(done_by_input[need]) and all(done_by_input[need].values()):
wfi.sendLog('stagor', "%s is ready" % need)
print json.dumps(done_by_input[need], indent=2)
readys[need] = True
else:
wfi.sendLog(
'stagor', "%s is not ready \n%s" %
(need, json.dumps(done_by_input[need], indent=2)))
readys[need] = False
if readys and all(readys.values()):
if wfo.status == 'staging':
wfi.sendLog('stagor',
"all needs are fullfilled, setting staged")
wfo.status = 'staged'
session.commit()
else:
wfi.sendLog('stagor', "all needs are fullfilled, already")
print json.dumps(readys, indent=2)
else:
wfi.sendLog('stagor', "missing requirements")
copies_needed, _ = wfi.getNCopies()
jump_ahead = False
re_transfer = False
## there is missing input let's do something more elaborated
for need in list(primaries): #+list(secondaries):
if endpoint_in_downtime[need] and endpoint_in_downtime[
need] == endpoint_incompleted[need]:
#print need,"is going to an end point in downtime"
wfi.sendLog(
'stagor',
"%s has only incomplete endpoint in downtime\n%s" %
(need, endpoint_in_downtime[need]))
re_transfer = True
if not se_allowed_key in available_cache[need]:
available_cache[need][
se_allowed_key] = getDatasetBlocksFraction(
url, need, sites=se_allowed)
if available_cache[need][se_allowed_key] >= copies_needed:
wfi.sendLog(
'stagor',
"assuming it is OK to move on like this already for %s"
% need)
jump_ahead = True
else:
wfi.sendLog(
'stagor', "Available %s times" %
available_cache[need][se_allowed_key])
missing_and_downtime = list(
set(endpoint_in_downtime[need])
& set(endpoint_incompleted[need]))
if missing_and_downtime:
wfi.sendLog(
'stagor',
"%s is incomplete at %s which is in downtime, trying to move along"
% (need, ','.join(missing_and_downtime)))
jump_ahead = True
else:
wfi.sendLog(
'stagor',
"continue waiting for transfers for optimum production performance."
)
## compute a time since staging to filter jump starting ?
# check whether the inputs is already in the stuck list ...
for need in list(primaries) + list(secondaries):
if need in already_stuck:
wfi.sendLog('stagor',
"%s is stuck, so try to jump ahead" % need)
jump_ahead = True
if jump_ahead or re_transfer:
details_text = "checking on availability for %s to jump ahead" % wfo.name
details_text += '\n%s wants %s copies' % (wfo.name,
copies_needed)
copies_needed = max(1, copies_needed - 1)
details_text += '\nlowering by one unit to %s' % copies_needed
wfi.sendLog('stagor', details_text)
all_check = True
prim_where = set()
for need in list(primaries):
if not se_allowed_key in presence_cache[need]:
presence_cache[need][
se_allowed_key] = getDatasetPresence(
url, need, within_sites=se_allowed)
presence = presence_cache[need][se_allowed_key]
prim_where.update(presence.keys())
available = available_cache[need][se_allowed_key]
this_check = (available >= copies_needed)
wfi.sendLog(
'stagor', "%s is available %s times (%s), at %s" %
(need, available, this_check, se_allowed_key))
all_check &= this_check
if not all_check: break
for need in list(secondaries):
## I do not want to check on the secon
## this below does not function because the primary could be all available, and the secondary not complete at a certain site that does not matter at that point
this_check = all(done_by_input[need].values())
wfi.sendLog(
'stagor', "%s is this much transfered %s" %
(need, json.dumps(done_by_input[need], indent=2)))
all_check &= this_check
#if not se_allowed_key in presence_cache[need]:
# presence_cache[need][se_allowed_key] = getDatasetPresence( url, need , within_sites=se_allowed)
## restrict to where the primary is
#presence = dict([(k,v) for (k,v) in presence_cache[need][se_allowed_key].items() if k in prim_where])
#this_check = all([there for (there,frac) in presence.values()])
#print need,"is present at all sites:",this_check
#all_check&= this_check
if all_check and not re_transfer:
wfi.sendLog(
'stagor',
"needs are sufficiently fullfilled, setting staged")
wfo.status = 'staged'
session.commit()
else:
print wfo.name, "has to wait a bit more"
wfi.sendLog('stagor', "needs to wait a bit more")
else:
wfi.sendLog('stagor', "not checking availability")
if re_transfer:
wfi.sendLog(
'stagor',
"Sending back to considered because of endpoint in downtime"
)
if wfo.status == 'staging':
wfo.status = 'considered'
session.commit()
send_back_to_considered.add(wfo.name)
time_point("Checked affected workflows")
if send_back_to_considered:
#sendEmail("transfer to endpoint in downtime","sending back to considered the following workflows \n%s"%('\n'.join( send_back_to_considered)))
sendLog('stagor',
"sending back to considered the following workflows \n%s" %
('\n'.join(send_back_to_considered)),
level='critical')
print "-" * 10, "Checking on non-available datasets", "-" * 10
## now check on those that are not fully available
for dsname in available_cache.keys():
## squash the se_allowed_key key
available_cache[dsname] = min(available_cache[dsname].values())
really_stuck_dataset = set()
for dsname, available in available_cache.items():
using_its = getWorkflowByInput(url, dsname)
#print using_its
using_wfos = []
for using_it in using_its:
wf = session.query(Workflow).filter(
Workflow.name == using_it).first()
if wf:
using_wfos.append(wf)
if not len(done_by_input[dsname]):
print "For dataset", dsname, "there are no transfer report. That's an issue."
for wf in using_wfos:
if wf.status == 'staging':
if UC.get("stagor_sends_back"):
print "sending", wf.name, "back to considered"
wf.status = 'considered'
session.commit()
#sendEmail( "send back to considered","%s was send back and might be trouble"% wf.name)
sendLog('stagor',
"%s was send back and might be trouble" %
wf.name,
level='critical')
else:
print "would send", wf.name, "back to considered"
#sendEmail( "subscription lagging behind","susbscriptions to get %s running are not appearing in phedex. I would have send it back to considered but that's not good."% wf.name)
sendLog(
'stagor',
"susbscriptions to get %s running are not appearing in phedex. I would have send it back to considered but that's not good."
% wf.name,
level='critical')
continue
## not compatible with checking on secondary availability
#if all([wf.status != 'staging' for wf in using_wfos]):
# ## means despite all checks that input is not needed
# continue
if available < 1.:
print "incomplete", dsname
## there is a problem in the method below that it does not account for files stuck in T1*Buffer only
lost_blocks, lost_files = findLostBlocksFiles(
url, dsname) if (not dsname.endswith('/RAW')) else ([], [])
lost_block_names = [item['name'] for item in lost_blocks]
lost_file_names = [item['name'] for item in lost_files]
if lost_blocks:
#print json.dumps( lost , indent=2 )
## estimate for how much !
fraction_loss, _, n_missing = getDatasetBlockFraction(
dsname, lost_block_names)
print "We have lost", len(
lost_block_names
), "blocks", lost_block_names, "for %f%%" % (100. *
fraction_loss)
if fraction_loss > 0.05: ## 95% completion mark
#sendEmail('we have lost too many blocks','%s is missing %d blocks, for %d events, %f %% loss'%(dsname, len(lost_block_names), n_missing, fraction_loss))
sendLog(
'stagor',
'%s is missing %d blocks, for %d events, %3.2f %% loss'
% (dsname, len(lost_block_names), n_missing,
100 * fraction_loss),
level='critical')
## the workflow should be rejected !
for wf in using_wfos:
if wf.status == 'staging':
print wf.name, "is doomed. setting to trouble"
wf.status = 'trouble'
session.commit()
sendLog(
'stagor',
'%s has too much loss on the input dataset %s. Missing %d blocks, for %d events, %3.2f %% loss'
% (wf.name, dsname, len(lost_block_names),
n_missing, 100 * fraction_loss),
level='critical')
else:
## probably enough to make a ggus and remove
if not dsname in known_lost_blocks:
#sendEmail('we have lost a few blocks', '%s is missing %d blocks, for %d events, %f %% loss\n\n%s'%(dsname, len(lost_block_names), n_missing, fraction_loss, '\n'.join( lost_block_names ) ))
sendLog(
'stagor',
'%s is missing %d blocks, for %d events, %f %% loss\n\n%s'
% (dsname, len(lost_block_names), n_missing,
fraction_loss, '\n'.join(lost_block_names)),
level='critical')
known_lost_blocks[dsname] = [
i['name'] for i in lost_blocks
]
really_stuck_dataset.add(dsname)
if lost_files:
fraction_loss, _, n_missing = getDatasetFileFraction(
dsname, lost_file_names)
print "We have lost", len(
lost_file_names
), "files", lost_file_names, "for %f%%" % fraction_loss
if fraction_loss > 0.05:
#sendEmail('we have lost too many files','%s is missing %d files, for %d events, %f %% loss'%(dsname, len(lost_file_names),n_missing, fraction_loss))
sendLog(
'stagor',
'%s is missing %d files, for %d events, %f %% loss' %
(dsname, len(lost_file_names), n_missing,
fraction_loss),
level='critical')
for wf in using_wfos:
if wf.status == 'staging':
print wf.name, "is doomed. setting to trouble"
wf.status = 'trouble'
session.commit()
else:
## probably enough to make a ggus and remove
if not dsname in known_lost_files:
#sendEmail('we have lost a few files','%s is missing %d files, for %d events, %f %% loss\n\n%s'%(dsname, len(lost_file_names),n_missing, fraction_loss, '\n'.join(lost_file_names)))
sendLog(
'stagor',
'%s is missing %d files, for %d events, %f %% loss\n\n%s'
% (dsname, len(lost_file_names), n_missing,
fraction_loss, '\n'.join(lost_file_names)),
level='critical')
known_lost_files[dsname] = [
i['name'] for i in lost_files
]
## should the status be change to held-staging and pending on a ticket
missings = [
pid for (pid, d) in done_by_input[dsname].items() if d == False
]
print "\t", done_by_input[dsname]
print "\tneeds", len(done_by_input[dsname])
print "\tgot", done_by_input[dsname].values().count(True)
print "\tmissing", missings
missing_in_action[dsname].extend(missings)
rr = eosFile('%s/lost_blocks_datasets.json' % monitor_dir, 'w')
rr.write(json.dumps(known_lost_blocks, indent=2))
rr.close()
rr = eosFile('%s/lost_files_datasets.json' % monitor_dir, 'w')
rr.write(json.dumps(known_lost_files, indent=2))
rr.close()
eosFile('%s/incomplete_transfers.json' % monitor_dir,
'w').write(json.dumps(missing_in_action, indent=2)).close()
print "Stuck transfers and datasets"
print json.dumps(missing_in_action, indent=2)
TD = transferDataset()
datasets_by_phid = defaultdict(set)
for dataset in missing_in_action:
for phid in missing_in_action[dataset]:
#print dataset,"stuck through",phid
datasets_by_phid[phid].add(dataset)
for k in datasets_by_phid:
#datasets_by_phid[k] = list(datasets_by_phid[k])
TD.add(k, list(datasets_by_phid[k]))
#eosFile('%s/datasets_by_phid.json'%base_eos_dir,'w').write( json.dumps(datasets_by_phid, indent=2 )).close()
eosFile('%s/really_stuck_dataset.json' % base_eos_dir,
'w').write(json.dumps(list(really_stuck_dataset),
indent=2)).close()
print '\n' * 2, "Datasets really stuck"
print '\n'.join(really_stuck_dataset)
#############
## not going further for what matters
#############
return
def rejector(url, specific, options=None):
up = componentInfo(soft=['wtc', 'jira'])
#if not up.check(): return
if specific and specific.startswith('/'):
## this is for a dataset
print setDatasetStatus(specific, 'INVALID')
return
if options.filelist:
wfs = []
for line in filter(None, open(options.filelist).read().split('\n')):
print line
wfs.extend(
session.query(Workflow).filter(
Workflow.name.contains(line)).all())
elif specific:
wfs = session.query(Workflow).filter(
Workflow.name.contains(specific)).all()
if not wfs:
batches = batchInfo().content()
for bname in batches:
if specific == bname:
for pid in batches[bname]:
b_wfs = getWorkflowById(url, pid)
for wf in b_wfs:
wfs.append(
session.query(Workflow).filter(
Workflow.name == wf).first())
break
else:
wfs = session.query(Workflow).filter(
Workflow.status == 'assistance-clone').all()
#wfs.extend( session.query(Workflow).filter(Workflow.status == 'assistance-reject').all())
## be careful then on clone case by case
options.clone = True
print "not supposed to function yet"
return
print len(wfs), "to reject"
if len(wfs) > 1:
print "\n".join([wfo.name for wfo in wfs])
answer = raw_input('Reject these')
if not answer.lower() in ['y', 'yes']:
return
for wfo in wfs:
#wfo = session.query(Workflow).filter(Workflow.name == specific).first()
if not wfo:
print "cannot reject", spec
return
wfi = workflowInfo(url, wfo.name)
comment = ""
if options.comments: comment = ", reason: " + options.comments
if options.keep:
wfi.sendLog(
'rejector',
'invalidating the workflow by unified operator%s' % comment)
else:
wfi.sendLog(
'rejector',
'invalidating the workflow and outputs by unified operator%s' %
comment)
results = invalidate(url,
wfi,
only_resub=True,
with_output=(not options.keep))
if all(results):
print wfo.name, "rejected"
if options and options.clone:
wfo.status = 'trouble'
session.commit()
schema = wfi.getSchema()
schema['Requestor'] = os.getenv('USER')
schema['Group'] = 'DATAOPS'
schema['OriginalRequestName'] = wfo.name
if 'ProcessingVersion' in schema:
schema['ProcessingVersion'] = int(
schema['ProcessingVersion']
) + 1 ## dubious str->int conversion
else:
schema['ProcessingVersion'] = 2
for k in schema.keys():
if k.startswith('Team'):
schema.pop(k)
if k.startswith('checkbox'):
schema.pop(k)
## a few tampering of the original request
if options.Memory:
if schema['RequestType'] == 'TaskChain':
it = 1
while True:
t = 'Task%d' % it
it += 1
if t in schema:
schema[t]['Memory'] = options.Memory
else:
break
else:
schema['Memory'] = options.Memory
if options.short_task and schema['RequestType'] == 'TaskChain':
translate = {}
it = 1
while True:
tt = 'Task%d' % it
if tt in schema:
tname = schema[tt]['TaskName']
ntname = 'T%d' % it
translate[tname] = ntname
it += 1
schema[tt]['TaskName'] = ntname
if 'InputTask' in schema[tt]:
itname = schema[tt]['InputTask']
schema[tt]['InputTask'] = translate[itname]
else:
break
for k in schema.get('ProcessingString', {}).keys():
schema['ProcessingString'][
translate[k]] = schema['ProcessingString'].pop(k)
for k in schema.get('AcquisitionEra', {}).keys():
schema['AcquisitionEra'][
translate[k]] = schema['AcquisitionEra'].pop(k)
if options.Multicore:
## to do : set it properly in taskchains
if schema['RequestType'] == 'TaskChain':
tasks, set_to = options.Multicore.split(
':') if ':' in options.Multicore else (
"", options.Multicore)
set_to = int(set_to)
tasks = tasks.split(',') if tasks else ['Task1']
it = 1
while True:
tt = 'Task%d' % it
it += 1
if tt in schema:
tname = schema[tt]['TaskName']
if tname in tasks or tt in tasks:
mem = schema[tt]['Memory']
mcore = schema[tt].get('Multicore', 1)
factor = (set_to / float(mcore))
fraction_constant = 0.4
mem_per_core_c = int(
(1 - fraction_constant) * mem /
float(mcore))
print "mem per core", mem_per_core_c
print "base mem", mem
## adjusting the parameter in the clone
schema[tt]['Memory'] += (
set_to - mcore) * mem_per_core_c
schema[tt]['Multicore'] = set_to
schema[tt]['TimePerEvent'] /= factor
else:
break
else:
schema['Multicore'] = options.Multicore
if options.deterministic:
if schema['RequestType'] == 'TaskChain':
schema['Task1']['DeterministicPileup'] = True
if options.EventsPerJob:
if schema['RequestType'] == 'TaskChain':
schema['Task1']['EventsPerJob'] = options.EventsPerJob
else:
schema['EventsPerJob'] = options.EventsPerJob
if options.EventAwareLumiBased:
schema['SplittingAlgo'] = 'EventAwareLumiBased'
if options.TimePerEvent:
schema['TimePerEvent'] = options.TimePerEvent
if options.ProcessingString:
schema['ProcessingString'] = options.ProcessingString
if options.AcquisitionEra:
schema['AcquisitionEra'] = options.AcquisitionEra
if options.runs:
schema['RunWhitelist'] = map(int, options.runs.split(','))
if options.PrepID:
schema['PrepID'] = options.PrepID
if schema['RequestType'] == 'TaskChain' and options.no_output:
ntask = schema['TaskChain']
for it in range(1, ntask - 1):
schema['Task%d' % it]['KeepOutput'] = False
schema['TaskChain'] = ntask - 1
schema.pop('Task%d' % ntask)
if options.priority:
schema['RequestPriority'] = options.priority
## update to the current priority
schema['RequestPriority'] = wfi.request['RequestPriority']
## drop shit on the way to reqmgr2
schema = reqMgrClient.purgeClonedSchema(schema)
print "submitting"
if (options.to_stepchain
and (schema['RequestType'] == 'TaskChain')):
## transform the schema into StepChain schema
print "Transforming a TaskChain into a StepChain"
mcore = 0
mem = 0
schema['RequestType'] = 'StepChain'
schema['StepChain'] = schema.pop('TaskChain')
schema['SizePerEvent'] = 0
schema['TimePerEvent'] = 0
step = 1
s_n = {}
while True:
if 'Task%d' % step in schema:
sname = 'Step%d' % step
schema[sname] = schema.pop('Task%d' % step)
tmcore = schema[sname].pop('Multicore')
tmem = schema[sname].pop('Memory')
if mcore and tmcore != mcore:
wfi.sendLog(
'rejector',
'the conversion to stepchain encoutered different value of Multicore %d != %d'
% (tmcore, mcore))
sendLog(
'rejector',
'the conversion of %s to stepchain encoutered different value of Multicore %d != %d'
% (wfo.name, tmcore, mcore),
level='critical')
mcore = max(mcore, tmcore)
mem = max(mem, tmem)
schema[sname]['StepName'] = schema[sname].pop(
'TaskName')
s_n[schema[sname]['StepName']] = sname
if 'InputTask' in schema[sname]:
schema[sname]['InputStep'] = schema[sname].pop(
'InputTask')
eff = 1.
up_s = sname
while True:
## climb up a step. supposedely already all converted
up_s = s_n.get(
schema[up_s].get('InputStep', None), None)
if up_s:
## multiply with the efficiency
eff *= schema[up_s].get(
'FilterEfficiency', 1.)
else:
## or stop there
break
if not 'KeepOutput' in schema[sname]:
## this is a weird translation capability. Absence of keepoutput in step means : keep the output. while in TaskChain absence means : drop
schema[sname]['KeepOutput'] = False
schema['TimePerEvent'] += eff * schema[sname].pop(
'TimePerEvent')
schema['SizePerEvent'] += eff * schema[sname].pop(
'SizePerEvent')
step += 1
else:
break
schema['Multicore'] = mcore
schema['Memory'] = mem
print json.dumps(schema, indent=2)
newWorkflow = reqMgrClient.submitWorkflow(url, schema)
if not newWorkflow:
msg = "Error in cloning {}".format(wfo.name)
print(msg)
wfi.sendLog('rejector', msg)
# Get the error message
time.sleep(5)
data = reqMgrClient.requestManagerPost(
url, "/reqmgr2/data/request", schema)
wfi.sendLog('rejector', data)
print json.dumps(schema, indent=2)
return
print newWorkflow
data = reqMgrClient.setWorkflowApproved(url, newWorkflow)
print data
wfi.sendLog(
'rejector', 'Cloned into %s by unified operator %s' %
(newWorkflow, comment))
#wfi.notifyRequestor('Cloned into %s by unified operator %s'%( newWorkflow, comment ),do_batch=False)
else:
wfo.status = 'trouble' if options.set_trouble else 'forget'
wfi.notifyRequestor('Rejected by unified operator %s' %
(comment),
do_batch=False)
session.commit()
else:
msg = "Error in rejecting {}: {}".format(wfo.name, results)
print(msg)
wfi.sendLog('rejector', msg)
def injector(url, options, specific):
use_mcm = True
up = componentInfo(mcm=use_mcm, soft=['mcm'])
if not up.check(): return
use_mcm = up.status['mcm']
workflows = getWorkflows(url, status=options.wmstatus, user=options.user)
workflows.extend(
getWorkflows(url,
status=options.wmstatus,
user='******',
rtype="ReReco")
) ## regardless of users, pick up all ReReco on the table
print len(workflows), "in line"
cannot_inject = set()
## browse for assignment-approved requests, browsed for ours, insert the diff
for wf in workflows:
if specific and not specific in wf: continue
exists = session.query(Workflow).filter(Workflow.name == wf).first()
if not exists:
wfi = workflowInfo(url, wf)
#wl = getWorkLoad(url, wf)
## check first that there isn't related here with something valid
can_add = True
## first try at finding a match
# print wfi.request
familly = session.query(Workflow).filter(
Workflow.name.contains(wfi.request['PrepID'])).all()
if not familly:
#req_familly = getWorkflowById( url, wl['PrepID'])
#familly = [session.query(Workflow).filter(Workflow.name == member).first() for member in req_familly]
pids = wfi.getPrepIDs()
req_familly = []
for pid in pids:
req_familly.extend(getWorkflowById(url, pid, details=True))
familly = []
print len(req_familly), "members"
for req_member in req_familly:
#print "member",req_member['RequestName']
owfi = workflowInfo(url,
req_member['RequestName'],
request=req_member)
other_pids = owfi.getPrepIDs()
if set(pids) == set(other_pids):
## this is a real match
familly.extend(
session.query(Workflow).filter(
Workflow.name ==
req_member['RequestName']).all())
for lwfo in familly:
if lwfo:
## we have it already
if not lwfo.status in [
'forget', 'trouble', 'forget-unlock',
'forget-out-unlock'
]:
sendLog(
'injector', "Should not put %s because of %s %s" %
(wf, lwfo.name, lwfo.status))
print "Should not put", wf, "because of", lwfo.name, lwfo.status
cannot_inject.add(wf)
can_add = False
if not can_add: continue
wfi.sendLog('injector', "considering %s" % wf)
new_wf = Workflow(name=wf,
status=options.setstatus,
wm_status=options.wmstatus)
session.add(new_wf)
session.commit()
time.sleep(0.5)
else:
#print "already have",wf
pass
if cannot_inject:
#sendEmail('workflow duplicates','These workflow cannot be added in because of duplicates \n\n %s'%( '\n'.join(cannot_inject)))
sendLog(
'injector',
'These workflow cannot be added in because of duplicates \n\n %s' %
('\n'.join(cannot_inject)),
level='warning')
## passing a round of invalidation of what needs to be invalidated
if use_mcm and (options.invalidate or True):
invalidator(url)
no_replacement = set()
## pick up replacements
for wf in session.query(Workflow).filter(
Workflow.status == 'trouble').all():
print wf.name
if specific and not specific in wf.name: continue
print wf.name
wfi = workflowInfo(url, wf.name)
wl = wfi.request #getWorkLoad(url, wf.name)
familly = getWorkflowById(url, wl['PrepID'])
true_familly = []
for member in familly:
if member == wf.name: continue
fwl = getWorkLoad(url, member)
if options.replace:
if member != options.replace: continue
else:
if fwl['RequestDate'] < wl['RequestDate']: continue
if fwl['RequestType'] == 'Resubmission': continue
if fwl['RequestStatus'] in ['None', None, 'new']: continue
if fwl['RequestStatus'] in [
'rejected', 'rejected-archived', 'aborted',
'aborted-archived'
]:
continue
true_familly.append(fwl)
if len(true_familly) == 0:
#sendLog('injector','%s had no replacement'%wf.name, level='critical')
wfi.sendLog(
'injector',
'the workflow was found in trouble with no replacement')
no_replacement.add(wf.name)
continue
else:
wfi.sendLog(
'injector',
'the workflow was found in trouble and has a replacement')
print wf.name, "has", len(familly), "familly members"
print wf.name, "has", len(true_familly), "true familly members"
##we cannot have more than one of them !!! pick the last one
if len(true_familly) > 1:
#sendEmail('multiple wf','please take a look at injector for %s'%wf.name)
sendLog('injector',
'Multiple wf in line, will take the last one for %s \n%s' %
(wf.name, ', '.join(fwl['RequestName']
for fwl in true_familly)),
level='critical')
for fwl in true_familly[-1:]:
member = fwl['RequestName']
new_wf = session.query(Workflow).filter(
Workflow.name == member).first()
if not new_wf:
sendLog('injector',
"putting %s as replacement of %s" % (member, wf.name))
status = 'away'
if fwl['RequestStatus'] in ['assignment-approved']:
status = 'considered'
new_wf = Workflow(name=member,
status=status,
wm_status=fwl['RequestStatus'])
wf.status = 'forget'
session.add(new_wf)
else:
if new_wf.status == 'forget': continue
sendLog(
'injector',
"getting %s as replacement of %s" % (new_wf.name, wf.name))
wf.status = 'forget'
for tr in session.query(Transfer).all():
if wf.id in tr.workflows_id:
sw = copy.deepcopy(tr.workflows_id)
sw.remove(wf.id)
sw.append(new_wf.id)
tr.workflows_id = sw
print tr.phedexid, "got", new_wf.name
if new_wf.status != 'away':
print "\t setting it considered"
new_wf.status = 'considered'
if tr.phedexid < 0: ## set it back to positive
tr.phedexid = -tr.phedexid
session.commit()
## don't do that automatically
#wf.status = 'forget'
session.commit()
if no_replacement:
#sendEmail('workflow with no replacement','%s \n are dangling there'%( '\n'.join(no_replacement)))
sendLog('injector',
'workflow with no replacement, %s \n are dangling there' %
('\n'.join(no_replacement)),
level='critical')
def assignor(url, specific=None, talk=True, options=None):
if userLock(): return
if duplicateLock(): return
#if notRunningBefore( 'stagor' ): return
if not componentInfo().check(): return
UC = unifiedConfiguration()
CI = campaignInfo()
SI = global_SI
#LI = lockInfo()
NLI = newLockInfo()
n_assigned = 0
n_stalled = 0
wfos = []
if specific or options.early:
wfos.extend(
session.query(Workflow).filter(
Workflow.status == 'considered').all())
wfos.extend(
session.query(Workflow).filter(Workflow.status == 'staging').all())
if specific:
wfos.extend(
session.query(Workflow).filter(
Workflow.status == 'considered-tried').all())
wfos.extend(
session.query(Workflow).filter(Workflow.status == 'staged').all())
#if specific:
# #wfos = session.query(Workflow).filter(Workflow.name==specific).all()
# wfos = session.query(Workflow).filter(Workflow.name.contains(specific)).all()
#if not wfos:
# if specific:
# wfos = session.query(Workflow).filter(Workflow.status=='considered').all()
# wfos.extend( session.query(Workflow).filter(Workflow.status=='staging').all())
# wfos.extend(session.query(Workflow).filter(Workflow.status=='staged').all())
random.shuffle(wfos)
for wfo in wfos:
if options.limit and (n_stalled + n_assigned) > options.limit:
break
if specific:
if not any(map(lambda sp: sp in wfo.name, specific.split(','))):
continue
#if not specific in wfo.name: continue
print "\n\n"
wfh = workflowInfo(url, wfo.name)
wfh.sendLog('assignor', "%s to be assigned" % wfo.name)
## check if by configuration we gave it a GO
if not CI.go(wfh.request['Campaign']) and not options.go:
wfh.sendLog('assignor', "No go for %s" % wfh.request['Campaign'])
n_stalled += 1
continue
## check on current status for by-passed assignment
if wfh.request['RequestStatus'] != 'assignment-approved':
if not options.test:
wfh.sendLog('assignor',
"setting %s away and skipping" % wfo.name)
## the module picking up from away will do what is necessary of it
wfo.wm_status = wfh.request['RequestStatus']
wfo.status = 'away'
session.commit()
continue
else:
print wfo.name, wfh.request['RequestStatus']
## retrieve from the schema, dbs and reqMgr what should be the next version
version = wfh.getNextVersion()
if not version:
if options and options.ProcessingVersion:
version = options.ProcessingVersion
else:
wfh.sendLog('assignor', "cannot decide on version number")
n_stalled += 1
wfo.status = 'trouble'
session.commit()
continue
## the site whitelist takes into account siteInfo, campaignInfo, memory and cores
(lheinput, primary, parent, secondary,
sites_allowed) = wfh.getSiteWhiteList()
original_sites_allowed = copy.deepcopy(sites_allowed)
wfh.sendLog('assignor', "Site white list %s" % sorted(sites_allowed))
override_sec_location = CI.get(wfh.request['Campaign'],
'SecondaryLocation', [])
blocks = []
if 'BlockWhitelist' in wfh.request:
blocks = wfh.request['BlockWhitelist']
if 'RunWhitelist' in wfh.request and wfh.request['RunWhitelist']:
## augment with run white list
for dataset in primary:
blocks = list(
set(blocks + getDatasetBlocks(
dataset, runs=wfh.request['RunWhitelist'])))
wfh.sendLog('assignor', "Allowed %s" % sorted(sites_allowed))
secondary_locations = None
for sec in list(secondary):
if override_sec_location:
print "We don't care where the secondary is"
print "Cannot pass for now"
sendEmail("tempting to pass sec location check",
"but we cannot yet IMO")
#pass
presence = getDatasetPresence(url, sec)
print sec
print json.dumps(presence, indent=2)
one_secondary_locations = [
site for (site, (there, frac)) in presence.items()
if frac > 98.
]
#one_secondary_locations = [site for (site,(there,frac)) in presence.items() if there]
if secondary_locations == None:
secondary_locations = one_secondary_locations
else:
secondary_locations = list(
set(secondary_locations) & set(one_secondary_locations))
## reduce the site white list to site with secondary only
#sites_allowed = [site for site in sites_allowed if any([osite.startswith(site) for osite in one_secondary_locations])]
sites_allowed = [
site for site in sites_allowed
if SI.CE_to_SE(site) in one_secondary_locations
]
wfh.sendLog(
'assignor', "From secondary requirement, now Allowed%s" %
sorted(sites_allowed))
initial_sites_allowed = copy.deepcopy(
sites_allowed
) ## keep track of this, after secondary input location restriction : that's how you want to operate it
sites_all_data = copy.deepcopy(sites_allowed)
sites_with_data = copy.deepcopy(sites_allowed)
sites_with_any_data = copy.deepcopy(sites_allowed)
primary_locations = None
available_fractions = {}
set_lfn = '/store/mc' ## by default
for prim in list(primary):
set_lfn = getLFNbase(prim)
presence = getDatasetPresence(url, prim, only_blocks=blocks)
if talk:
print prim
print json.dumps(presence, indent=2)
available_fractions[prim] = getDatasetBlocksFraction(
url,
prim,
sites=[SI.CE_to_SE(site) for site in sites_allowed],
only_blocks=blocks)
#sites_all_data = [site for site in sites_with_data if any([osite.startswith(site) for osite in [psite for (psite,(there,frac)) in presence.items() if there]])]
#sites_with_data = [site for site in sites_with_data if any([osite.startswith(site) for osite in [psite for (psite,frac) in presence.items() if frac[1]>90.]])]
sites_all_data = [
site for site in sites_with_data if SI.CE_to_SE(site) in [
psite for (psite, (there, frac)) in presence.items()
if there
]
]
sites_with_data = [
site for site in sites_with_data if SI.CE_to_SE(site) in
[psite for (psite, frac) in presence.items() if frac[1] > 90.]
]
sites_with_any_data = [
site for site in sites_with_any_data
if SI.CE_to_SE(site) in presence.keys()
]
wfh.sendLog(
'assignor', "Holding the data but not allowed %s" % sorted(
list(
set([
se_site for se_site in presence.keys()
if not SI.SE_to_CE(se_site) in sites_allowed
]))))
if primary_locations == None:
primary_locations = presence.keys()
else:
primary_locations = list(
set(primary_locations) & set(presence.keys()))
sites_with_data = list(set(sites_with_data))
sites_with_any_data = list(set(sites_with_any_data))
opportunistic_sites = []
down_time = False
## opportunistic running where any piece of data is available
if secondary_locations or primary_locations:
## intersection of both any pieces of the primary and good IO
#opportunistic_sites = [SI.SE_to_CE(site) for site in list((set(secondary_locations) & set(primary_locations) & set(SI.sites_with_goodIO)) - set(sites_allowed))]
if secondary_locations and primary_locations:
opportunistic_sites = [
SI.SE_to_CE(site) for site in
list((set(secondary_locations) & set(primary_locations)) -
set([SI.CE_to_SE(site) for site in sites_allowed]))
]
elif primary_locations:
opportunistic_sites = [
SI.SE_to_CE(site) for site in list(
set(primary_locations) -
set([SI.CE_to_SE(site) for site in sites_allowed]))
]
else:
opportunistic_sites = []
wfh.sendLog(
'assignor', "We could be running in addition at %s" %
sorted(opportunistic_sites))
if any(
[osite in SI.sites_not_ready
for osite in opportunistic_sites]):
wfh.sendLog(
'assignor', "One of the usable site is in downtime %s" % ([
osite in SI.sites_not_ready
for osite in opportunistic_sites
]))
down_time = True
## should this be send back to considered ?
"""
if available_fractions and not all([available>=1. for available in available_fractions.values()]):
print "The input dataset is not located in full over sites"
print json.dumps(available_fractions)
if not options.test and not options.go:
known = []
try:
known = json.loads(open('cannot_assign.json').read())
except:
pass
if not wfo.name in known:
sendEmail( "cannot be assigned","%s is not full over sites \n %s"%(wfo.name,json.dumps(available_fractions)))
known.append( wfo.name )
open('cannot_assign.json','w').write(json.dumps( known, indent=2))
n_stalled+=1
continue ## skip skip skip
"""
## should be 2 but for the time-being let's lower it to get things going
copies_wanted, cpuh = wfh.getNCopies()
if 'Campaign' in wfh.request and wfh.request[
'Campaign'] in CI.campaigns and 'maxcopies' in CI.campaigns[
wfh.request['Campaign']]:
copies_needed_from_campaign = CI.campaigns[
wfh.request['Campaign']]['maxcopies']
copies_wanted = min(copies_needed_from_campaign, copies_wanted)
if not options.early:
less_copies_than_requested = UC.get("less_copies_than_requested")
copies_wanted = max(
1, copies_wanted -
less_copies_than_requested) # take one out for the efficiency
wfh.sendLog('assignor',
"needed availability fraction %s" % copies_wanted)
## should also check on number of sources, if large enough, we should be able to overflow most, efficiently
if available_fractions and not all([
available >= copies_wanted
for available in available_fractions.values()
]):
not_even_once = not all([
available >= 1. for available in available_fractions.values()
])
wfh.sendLog(
'assignor',
"The input dataset is not available %s times, only %s" %
(copies_wanted, available_fractions.values()))
if down_time and not options.go and not options.early:
wfo.status = 'considered'
session.commit()
wfh.sendLog(
'assignor',
"sending back to considered because of site downtime, instead of waiting"
)
sendEmail(
"cannot be assigned due to downtime",
"%s is not sufficiently available, due to down time of a site in the whitelist. check the assignor logs. sending back to considered."
% wfo.name)
continue
#pass
print json.dumps(available_fractions)
if (options.go and not_even_once) or not options.go:
known = []
try:
known = json.loads(open('cannot_assign.json').read())
except:
pass
if not wfo.name in known and not options.limit and not options.go and not options.early:
wfh.sendLog(
'assignor',
"cannot be assigned, %s is not sufficiently available.\n %s"
% (wfo.name, json.dumps(available_fractions)))
sendEmail(
"cannot be assigned",
"%s is not sufficiently available.\n %s" %
(wfo.name, json.dumps(available_fractions)))
known.append(wfo.name)
open('cannot_assign.json',
'w').write(json.dumps(known, indent=2))
n_stalled += 1
if options.early:
if wfo.status == 'considered':
wfh.sendLog('assignor', "setting considered-tried")
wfo.status = 'considered-tried'
session.commit()
else:
print "tried but status is", wfo.status
continue
## default back to white list to original white list with any data
print "Allowed", sites_allowed
if options.primary_aaa:
sites_allowed = initial_sites_allowed
#options.useSiteListAsLocation = True
options.TrustSitelists = True
else:
sites_allowed = sites_with_any_data
wfh.sendLog('assignor',
"Selected for any data %s" % sorted(sites_allowed))
if options.restrict:
print "Allowed", sites_allowed
sites_allowed = sites_with_any_data
print "Selected", sites_allowed
else:
if set(sites_with_data) != set(sites_allowed):
## the data is not everywhere we wanted to run at : enable aaa
print "Sites with 90% data not matching site white list (block choping!)"
print "Resorting to AAA reading for", list(
set(sites_allowed) - set(sites_with_data)), "?"
print "Whitelist site with any data", list(
set(sites_allowed) - set(sites_with_any_data))
#options.useSiteListAsLocation = True
#print "Not commissioned yet"
#continue
#print "We could be running at",opportunistic_sites,"in addition"
##sites_allowed = list(set(sites_allowed+ opportunistic_sites))
if not len(sites_allowed):
wfh.sendLog('assignor', "cannot be assign with no matched sites")
sendEmail("cannot be assigned", "%s has no whitelist" % (wfo.name))
n_stalled += 1
continue
t1_only = [ce for ce in sites_allowed if ce.startswith('T1')]
if t1_only:
# try to pick from T1 only first
sites_out = [SI.pick_dSE([SI.CE_to_SE(ce) for ce in t1_only])]
else:
# then pick any otherwise
sites_out = [
SI.pick_dSE([SI.CE_to_SE(ce) for ce in sites_allowed])
]
## one last modification now that we know we can assign, and to make sure all ressource can be used by the request : set all ON sites to whitelist
###sites_allowed = original_sites_allowed ## not needed, afterall as secondary jobs go their own ways
wfh.sendLog('assignor', "Placing the output on %s" % sites_out)
parameters = {
'SiteWhitelist': sites_allowed,
#'CustodialSites' : sites_custodial,
'NonCustodialSites': sites_out,
'AutoApproveSubscriptionSites': list(set(sites_out)),
'AcquisitionEra': wfh.acquisitionEra(),
'ProcessingString': wfh.processingString(),
'MergedLFNBase': set_lfn,
'ProcessingVersion': version,
}
## plain assignment here
team = 'production'
if os.getenv('UNIFIED_TEAM'): team = os.getenv('UNIFIED_TEAM')
if options and options.team:
team = options.team
## high priority team agent
#if wfh.request['RequestPriority'] >= 100000 and (wfh.request['TimePerEvent']*int(wfh.getRequestNumEvents()))/(8*3600.) < 10000:
# team = 'highprio'
# sendEmail("sending work with highprio team","%s"% wfo.name, destination=['*****@*****.**'])
## SDSC redirection
#if "T2_US_UCSD" in sites_with_data and random.random() < -0.5 and wfh.request['Campaign']=='RunIISpring15DR74' and int(wfh.getRequestNumEvents()) < 600000 and not any([out.endswith('RAW') for out in wfh.request['OutputDatasets']]):
# ## consider SDSC
# parameters['SiteWhitelist'] = ['T2_US_UCSD','T3_US_SDSC']
# parameters['useSiteListAsLocation'] = True
# team = 'allocation-based'
# sendEmail("sending work to SDSC","%s was assigned to SDSC/UCSD"% wfo.name, destination=['*****@*****.**'])
## SDSC redirection
#if wfh.request['Campaign']==R'unIIWinter15GS' and random.random() < -1.0:
# parameters['SiteWhitelist'] = ['T3_US_SDSC']
# team = 'allocation-based'
# sendEmail("sending work to SDSC","%s was assigned to SDSC"% wfo.name, destination=['*****@*****.**'])
if False and 'T2_CH_CERN' in parameters['SiteWhitelist']:
## add some check on
### the amount pending to HLT
### the size of the request
### the priority of the request (maybe not if we decide to overflow during runs)
parameters['SiteWhitelist'] = ['T2_CH_CERN_HLT']
team = 'hlt'
## reduce the splitting by factor of 4, regardless of type of splitting
sendEmail("sending work to HLT",
"%s was assigned to HLT" % wfo.name)
##parse options entered in command line if any
if options:
for key in reqMgrClient.assignWorkflow.keys:
v = getattr(options, key)
if v != None:
if type(v) == str and ',' in v:
parameters[key] = filter(None, v.split(','))
else:
parameters[key] = v
## pick up campaign specific assignment parameters
parameters.update(CI.parameters(wfh.request['Campaign']))
if not options.test:
parameters['execute'] = True
split_check = wfh.checkWorkflowSplitting()
if split_check != True:
parameters.update(split_check)
if 'EventBased' in split_check.values():
wfh.sendLog('assignor', "Falling back to event splitting.")
sendEmail(
"Fallback to EventBased",
"the workflow %s is too heavy to be processed as it is. Fallback to EventBased splitting"
% wfo.name)
elif 'EventsPerJob' in split_check.values():
wfh.sendLog('assignor',
"Modifying the number of job per event")
sendEmail(
"Modifying the job per events",
"the workflow %s is too heavy in number of jobs explosion"
% wfo.name)
# Handle run-dependent MC
pstring = wfh.processingString()
if 'PU_RD' in pstring:
numEvents = wfh.getRequestNumEvents()
eventsPerLumi = [getDatasetEventsPerLumi(prim) for prim in primary]
eventsPerLumi = sum(eventsPerLumi) / float(len(eventsPerLumi))
reqJobs = 500
if 'PU_RD2' in pstring:
reqJobs = 2000
eventsPerJob = int(numEvents / (reqJobs * 1.4))
lumisPerJob = int(eventsPerJob / eventsPerLumi)
if lumisPerJob == 0:
sendEmail(
"issue with event splitting for run-dependent MC",
"%s needs to be split by event with %s per job" %
(wfo.name, eventsPerJob))
wfh.sendLog(
'assignor',
"%s needs to be split by event with %s per job" %
(wfo.name, eventsPerJob))
parameters['EventsPerJob'] = eventsPerJob
else:
spl = wfh.getSplittings()[0]
eventsPerJobEstimated = spl[
'events_per_job'] if 'events_per_job' in spl else None
eventsPerJobEstimated = spl[
'avg_events_per_job'] if 'avg_events_per_job' in spl else None
if eventsPerJobEstimated and eventsPerJobEstimated > eventsPerJob:
sendEmail(
"setting lumi splitting for run-dependent MC",
"%s was assigned with %s lumis/job" %
(wfo.name, lumisPerJob))
wfh.sendLog(
'assignor', "%s was assigned with %s lumis/job" %
(wfo.name, lumisPerJob))
parameters['LumisPerJob'] = lumisPerJob
else:
sendEmail("leaving splitting untouched for PU_RD*",
"please check on " + wfo.name)
wfh.sendLog(
'assignor',
"leaving splitting untouched for PU_RD*, please check."
)
result = reqMgrClient.assignWorkflow(url, wfo.name, team, parameters)
# set status
if not options.test:
if result:
wfo.status = 'away'
session.commit()
n_assigned += 1
try:
## refetch information and lock output
new_wfi = workflowInfo(url, wfo.name)
(_, prim, _, sec) = new_wfi.getIO()
for secure in list(prim) + list(
sec) + new_wfi.request['OutputDatasets']:
## lock all outputs flat
NLI.lock(secure)
#for site in [SI.CE_to_SE(site) for site in sites_allowed]:
# for output in new_wfi.request['OutputDatasets']:
# LI.lock( output, site, 'dataset in production')
# for primary in prim:
# LI.lock( primary, site, 'dataset used in input')
# for secondary in sec:
# LI.lock( secondary, site, 'required for mixing' )
except Exception as e:
print "fail in locking output"
print str(e)
sendEmail("failed locking of output", str(e))
else:
print "ERROR could not assign", wfo.name
else:
pass
print "Assignment summary:"
sendLog('assignor', "Assigned %d Stalled %s" % (n_assigned, n_stalled))
def singleClone(url, wfname, actions, comment, do=False):
wfi = workflowInfo(url, wfname)
payload = wfi.getSchema()
initial = wfi.request
payload['Requestor'] = os.getenv('USER')
payload['Group'] = 'DATAOPS'
payload['OriginalRequestName'] = initial['RequestName']
payload['RequestPriority'] = initial['RequestPriority']
if 'ProcessingVersion' in initial:
payload['ProcessingVersion'] = int(initial['ProcessingVersion']) + 1
else:
payload['ProcessingVersion'] = 2
payload = reqMgrClient.purgeClonedSchema(payload)
if actions:
for action in actions:
if action.startswith('mem') and actions[action] != "" and actions[
action] != 'Same':
if 'TaskChain' in payload:
print "Setting memory for clone of task chain"
mem_dict = {}
it = 1
while True:
t = 'Task%d' % it
it += 1
if t in payload:
tname = payload[t]['TaskName']
mem_dict[tname] = int(actions[action])
print "Memory set for Task%d" % it
else:
break
payload['Memory'] = mem_dict
else:
print "Setting memory for non-taskchain workflow"
payload['Memory'] = int(actions[action])
print "Memory set to " + actions[action]
print "Clone payload"
# print json.dumps( payload , indent=2)
print actions
#Create clone
clone = reqMgrClient.submitWorkflow(url, payload)
if not clone:
print "Error in making clone for", initial["RequestName"]
clone = reqMgrClient.submitWorkflow(url, payload)
if not clone:
print "Error twice in making clone for", initial["RequestName"]
sendLog('actor',
'Failed to make a clone twice for %s!' %
initial["RequestName"],
level='critical')
wfi.sendLog(
'actor', 'Failed to make a clone twice for %s!' %
initial["RequestName"])
return None
if actions:
for action in actions:
if action.startswith('split'):
cloneinfo = workflowInfo(url, clone)
splittings = cloneinfo.getSplittingsNew(strip=True)
if actions[action] != 'Same' and actions[
action] != 'max' and actions[action] != '':
factor = int(
actions[action][0:-1]) if 'x' in actions[action] else 2
for split in splittings:
split_par = split['splitParams']
for act in [
'avg_events_per_job', 'events_per_job',
'lumis_per_job'
]:
if act in split_par:
wfi.sendLog(
'actor',
'Changing %s (%d) by a factor %d' %
(act, split_par[act], factor))
split_par[act] /= factor
print "to", split_par[act]
break
#split['requestName'] = clone
#print "changing the splitting of",clone
#print json.dumps( split, indent=2 )
#print reqMgrClient.setWorkflowSplitting(url, clone, split )
elif 'max' in actions[action]:
for split in splittings:
split_par = split['splitParams']
for act in [
'avg_events_per_job', 'events_per_job',
'lumis_per_job'
]:
if act in split_par:
wfi.sendLog(
'actor', 'Max splitting set for %s (%d' %
(act, split_par[act]))
print "Changing %s (%d) " % (act,
split_par[act]),
split_par[act] = 1
print "to max splitting ", split_par[act]
break
#split['requestName'] = clone
#print "changing the splitting of",clone
#print json.dumps( split, indent=2 )
#print reqMgrClient.setWorkflowSplitting(url, clone, split )
print "changing the splitting of", clone
print json.dumps(splittings, indent=2)
print reqMgrClient.setWorkflowSplitting(url, clone, splittings)
#Approve
data = reqMgrClient.setWorkflowApproved(url, clone)
#wfi.sendLog('actor','Cloned into %s'%clone)
# wfi.sendLog('actor','Cloned into %s by unified operator %s'%( clone, comment ))
# wfi.notifyRequestor('Cloned into %s by unified operator %s'%( clone, comment ),do_batch=False)
print data
return clone
def actor(url,options=None):
if userLock('actor'): return
up = componentInfo(mcm=False, soft=['mcm'])
if not up.check(): return
# CI = campaignInfo()
SI = siteInfo()
UC = unifiedConfiguration()
# Need to look at the actions page https://vocms0113.cern.ch:80/getaction (can add ?days=20) and perform any actions listed
try:
action_list = json.loads(os.popen('curl -s -k https://vocms0113.cern.ch:80/getaction?days=15').read())
## now we have a list of things that we can take action on
except:
print "Not able to load action list :("
sendLog('actor','Not able to load action list', level='critical')
return
print action_list
if not action_list:
print "EMPTY!"
return
for wfname in action_list:
print '-'*100
print "Looking at",wfname,"for recovery options"
to_clone = False
to_acdc = False
for key in action_list[wfname]:
if key == 'Parameters':
tasks = action_list[wfname][key]
elif key == 'Action' and action_list[wfname][key] == 'acdc':
print "Going to create ACDCs for ", wfname
to_acdc = True
elif key == 'Action' and action_list[wfname][key] == 'clone':
print "Going to clone ", wfname
to_clone = True
if not to_acdc and not to_clone:
sendLog('actor','Action submitted for something other than acdc and clone for workflow %s'%wfname,level='critical')
print "Can only do acdcs and clones! Skipping workflow ",wfname
continue
if not tasks:
sendLog('actor','Empty action submitted for workflow %s'%wfname,level='critical')
print "Moving on. Parameters is blank for " + wfname
continue
wfi = workflowInfo(url, wfname)
recover = True
message_to_ops = ""
message_to_user = ""
#===========================================================
if to_clone and options.do:
print "Let's try kill and clone: "
wfi.sendLog('actor','Going to clone %s'%wfname)
results=[]
datasets = set(wfi.request['OutputDatasets'])
comment=""
if 'comment' in tasks: comment = ", reason: "+ tasks['comment']
wfi.sendLog('actor',"invalidating the workflow by traffic controller %s"%comment)
#Reject all workflows in the family
#first reject the original workflow.
reqMgrClient.invalidateWorkflow(url, wfi.request['RequestName'], current_status=wfi.request['RequestStatus'], cascade=False)
#Then reject any ACDCs associated with that workflow
if 'ACDCs' in action_list[wfname]:
children = action_list[wfname]['ACDCs']
for child in children:
wfi.sendLog('actor',"rejecting %s"%child)
wfi_acdc = workflowInfo(url, child)
reqMgrClient.invalidateWorkflow(url, wfi_acdc.request['RequestName'], current_status=wfi_acdc.request['RequestStatus'], cascade=False)
datasets.update( wfi_acdc.request['OutputDatasets'] )
#Invalidate all associated output datasets
for dataset in datasets:
results.append( setDatasetStatus(dataset, 'INVALID') )
if all(map(lambda result : result in ['None',None,True],results)):
wfi.sendLog('actor',"%s and children are rejected"%wfname)
cloned = None
try:
cloned = singleClone(url, wfname, tasks, comment, options.do)
except:
sendLog('actor','Failed to create clone for %s! Check logs for more information. Action will need to be resubmitted.'%wfname,level='critical')
wfi.sendLog('actor','Failed to create clone for %s!'%wfname)
remove_action(wfname)
if not cloned:
recover = False
wfi.sendLog('actor','Failed to create clone for %s!'%wfname)
sendLog('actor','Failed to create clone for %s!'%wfname,level='critical')
else:
wfi.sendLog('actor',"Workflow %s cloned"%wfname)
#===========================================================
elif to_acdc:
if 'AllSteps' in tasks:
allTasksDefaults = tasks['AllSteps']
tasks.pop('AllSteps')
for setting in allTasksDefaults:
for task in tasks:
if setting in tasks[task]:
tasks[task][setting] = allTasksDefaults[setting]
else:
tasks[task].append({setting:allTasksDefaults[setting]})
print "Tasks is "
print tasks
all_tasks = wfi.getAllTasks()
## need a way to verify that this is the first round of ACDC, since the second round will have to be on the ACDC themselves
try:
WMErr = wfi.getWMErrors()
# print WMErr
except:
sendLog('actor','Cannot create ACDCS for %s because WMErr cannot be reached.'%wfname,level='critical')
continue
if not WMErr:
sendLog('actor','Cannot create ACDCS for %s because WMErr is blank.'%wfname,level='critical')
print "Moving on. WMErr is blank"
continue
try:
where_to_run, missing_to_run,missing_to_run_at = wfi.getRecoveryInfo()
print "Where to run = "
print where_to_run
except:
sendLog('actor','Cannot create ACDCS for %s because recovery info cannot be found.'%wfname,level='critical')
print "Moving on. Cannot access recovery info for " + wfname
continue
if not where_to_run:
sendLog('actor','Cannot create ACDCS for %s because site list cannot be found.'%wfname,level='critical')
print "Moving on. where to run is blank"
continue
message_to_ops = ""
message_to_user = ""
num_tasks_to_recover = 0
for task in WMErr:
if 'LogCollect' in task: continue
if 'Cleanup' in task: continue
if not 'jobfailed' in WMErr[task]:
continue
else:
num_tasks_to_recover += 1
# print "Task to recover: " + task
if not num_tasks_to_recover:
print "\tno error for",wfname
# recover = False
if 'LheInputFiles' in wfi.request and wfi.request['LheInputFiles']:
## we do not try to recover pLHE
sendLog('actor','Cannot create ACDCS for %s because it is a pLHE workflow.'%wfname,level='critical')
print "We don't try to recover pLHE. Moving on."
recover = False
# sendEmail('cannot submit action', '%s is a pLHE workflow. We do not try to recover pLHE'%wfname)
# if wfi.request['RequestType'] in ['ReReco']:
# recover= False
# print 'cannot submit action. ReReco'
# sendEmail('cannot submit action', '%s is request type ReReco'%wfname)
recovering = set()
for task in tasks:
assign_to_sites = set()
print "Task names is " + task
fulltaskname = '/' + wfname + '/' + task
# print "Full task name is " + fulltaskname
wrong_task = False
for task_info in all_tasks:
if fulltaskname == task_info.pathName:
if task_info.taskType not in ['Processing','Production','Merge']:
wrong_task=True
wfi.sendLog('actor', "Skipping task %s because the taskType is %s. Can only ACDC Processing, Production, or Merge tasks"%( fulltaskname, task_info.taskType))
if wrong_task:
continue
print tasks[task]
actions = tasks[task]
for action in actions:
if action.startswith('sites'):
if type(actions[action]) != list:
assign_to_sites=[SI.SE_to_CE(actions[action])]
else:
assign_to_sites=list(set([SI.SE_to_CE(site) for site in actions[action]]))
# if action.startswith('mem') and actions[action] != "" and actions[action] != 'Same' and wfi.request['RequestType'] in ['TaskChain']:
# recover = False;
# print "Skipping %s for now until Allie fixes memory parameter for TaskChain ACDCs."%wfname
# wfi.sendLog('actor',"Skipping %s for now until Allie fixes memory parameter for TaskChain ACDCs."%wfname)
if not 'sites' in actions:
assign_to_sites = list(set([SI.SE_to_CE(site) for site in where_to_run[task]]))
print "Found",sorted(assign_to_sites),"as sites where to run the ACDC at, from the acdc doc of ",wfname
print "Going to run at",sorted(assign_to_sites)
if recover:
print "Initiating recovery"
acdc = singleRecovery(url, fulltaskname, wfi.request, actions, do = options.do)
if not acdc:
if options.do:
if recovering:
print wfname + " has been partially ACDC'ed. Needs manual attention."
sendLog('actor', "%s has had %s/%s recoveries %s only"%( wfname, len(recovering), num_tasks_to_recover, list(recovering)), level='critical')
wfi.sendLog('actor', "%s has had %s/%s recoveries %s only"%( wfname, len(recovering), num_tasks_to_recover, list(recovering)))
break
else:
print wfname + " failed recovery once"
recover = False
break
else:
print "no action to take further"
# sendLog('recoveror', "ACDC for %s can be done automatically"% wfname, level='critical')
continue
else: #ACDC was made correctly. Now we have to assign it.
wfi.sendLog('actor','ACDC created for task %s. Actions taken \n%s'%(fulltaskname,list(actions)))
team = wfi.request['Teams'][0]
parameters={
'SiteWhitelist' : sorted(assign_to_sites),
'AcquisitionEra' : wfi.acquisitionEra(),
'ProcessingString' : wfi.processingString(),
'MergedLFNBase' : wfi.request['MergedLFNBase'],
'ProcessingVersion' : wfi.request['ProcessingVersion'],
}
## hackery for ACDC merge assignment
if wfi.request['RequestType'] == 'TaskChain' and 'Merge' in task.split('/')[-1]:
parameters['AcquisitionEra'] = None
parameters['ProcessingString'] = None
## xrootd setttings on primary and secondary
if 'xrootd' in actions:
if actions['xrootd'] == 'enabled':
print "Going to assign via xrootd"
parameters['TrustSitelists'] = True
elif actions['xrootd'] == 'disabled':
parameters['TrustSitelists'] = False
elif ('TrustSitelists' in wfi.request and wfi.request['TrustSitelists']=='true'):
parameters['TrustSitelists'] = True
else:
parameters['TrustSitelists'] = False
if 'TrustPUSitelists' in wfi.request and wfi.request['TrustPUSitelists']:
parameters['TrustPUSitelists'] = True
if options.ass:
print "really doing the assignment of the ACDC",acdc
parameters['execute']=True
wfi.sendLog('actor',"%s was assigned for recovery"% acdc)
else:
print "no assignment done with this ACDC",acdc
sendLog('actor',"%s needs to be assigned"%(acdc), level='critical')
continue
# print parameters
result = reqMgrClient.assignWorkflow(url, acdc, team, parameters)
if not result:
print acdc,"was not assigned"
sendLog('actor',"%s needs to be assigned"%(acdc), level='critical')
else:
recovering.add( acdc )
wfi.sendLog('actor',"ACDCs created for %s"%wfname)
#===========================================================
if recover and options.do:
remove_action(wfname)
if message_to_user:
print wfname,"to be notified to user(DUMMY)",message_to_user
if message_to_ops:
print 'message'
#sendEmail( "notification in recoveror" , message_to_ops, destination=['*****@*****.**'])
# sendLog('recoveror',message_to_ops,level='warning')
return
def assignor(url, specific=None, talk=True, options=None):
if userLock():
return
if duplicateLock():
return
if not componentInfo().check():
return
UC = unifiedConfiguration()
CI = campaignInfo()
SI = siteInfo()
NLI = newLockInfo()
n_assigned = 0
n_stalled = 0
wfos = []
if specific or options.early:
wfos.extend(session.query(Workflow).filter(Workflow.status == "considered").all())
wfos.extend(session.query(Workflow).filter(Workflow.status == "staging").all())
if specific:
wfos.extend(session.query(Workflow).filter(Workflow.status == "considered-tried").all())
wfos.extend(session.query(Workflow).filter(Workflow.status == "staged").all())
dataset_endpoints = json.loads(open("%s/dataset_endpoints.json" % monitor_dir).read())
max_per_round = UC.get("max_per_round").get("assignor", None)
max_cpuh_block = UC.get("max_cpuh_block")
random.shuffle(wfos)
for wfo in wfos:
if options.limit and (n_stalled + n_assigned) > options.limit:
break
if max_per_round and (n_stalled + n_assigned) > max_per_round:
break
if specific:
if not any(map(lambda sp: sp in wfo.name, specific.split(","))):
continue
# if not specific in wfo.name: continue
print "\n\n"
wfh = workflowInfo(url, wfo.name)
wfh.sendLog("assignor", "%s to be assigned" % wfo.name)
## the site whitelist takes into account siteInfo, campaignInfo, memory and cores
(lheinput, primary, parent, secondary, sites_allowed) = wfh.getSiteWhiteList()
## check if by configuration we gave it a GO
no_go = False
if not wfh.go(log=True) and not options.go:
n_stalled += 1
no_go = True
allowed_secondary = set()
for campaign in wfh.getCampaigns():
if campaign in CI.campaigns and "secondaries" in CI.campaigns[campaign]:
allowed_secondary.update(CI.campaigns[campaign]["secondaries"])
if (secondary and allowed_secondary) and (set(secondary) & allowed_secondary != set(secondary)):
wfh.sendLog("assignor", "%s is not an allowed secondary" % (", ".join(set(secondary) - allowed_secondary)))
# sendEmail('secondary not allowed','%s is not an allowed secondary'%( ', '.join(set(secondary)-allowed_secondary)))
sendLog(
"assignor",
"%s is not an allowed secondary" % (", ".join(set(secondary) - allowed_secondary)),
level="critical",
)
if not options.go:
n_stalled += 1
no_go = True
if no_go:
continue
## check on current status for by-passed assignment
if wfh.request["RequestStatus"] != "assignment-approved":
if not options.test:
wfh.sendLog("assignor", "setting %s away and skipping" % wfo.name)
## the module picking up from away will do what is necessary of it
wfo.wm_status = wfh.request["RequestStatus"]
wfo.status = "away"
session.commit()
continue
else:
print wfo.name, wfh.request["RequestStatus"]
## retrieve from the schema, dbs and reqMgr what should be the next version
version = wfh.getNextVersion()
if not version:
if options and options.ProcessingVersion:
version = options.ProcessingVersion
else:
wfh.sendLog("assignor", "cannot decide on version number")
n_stalled += 1
wfo.status = "trouble"
session.commit()
continue
original_sites_allowed = copy.deepcopy(sites_allowed)
wfh.sendLog("assignor", "Site white list %s" % sorted(sites_allowed))
override_sec_location = CI.get(wfh.request["Campaign"], "SecondaryLocation", [])
blocks = []
if "BlockWhitelist" in wfh.request:
blocks = wfh.request["BlockWhitelist"]
if "RunWhitelist" in wfh.request and wfh.request["RunWhitelist"]:
## augment with run white list
for dataset in primary:
blocks = list(set(blocks + getDatasetBlocks(dataset, runs=wfh.request["RunWhitelist"])))
wfh.sendLog("assignor", "Allowed %s" % sorted(sites_allowed))
secondary_locations = None
primary_aaa = options.primary_aaa
if (
"Campaign" in wfh.request
and wfh.request["Campaign"] in CI.campaigns
and "primary_AAA" in CI.campaigns[wfh.request["Campaign"]]
):
primary_aaa = primary_aaa or CI.campaigns[wfh.request["Campaign"]]["primary_AAA"]
secondary_aaa = options.secondary_aaa
if (
"Campaign" in wfh.request
and wfh.request["Campaign"] in CI.campaigns
and "secondary_AAA" in CI.campaigns[wfh.request["Campaign"]]
):
secondary_aaa = secondary_aaa or CI.campaigns[wfh.request["Campaign"]]["secondary_AAA"]
for sec in list(secondary):
if override_sec_location:
print "We don't care where the secondary is"
print "Cannot pass for now"
sendEmail("tempting to pass sec location check", "but we cannot yet IMO")
# pass
if secondary_aaa:
# just continue without checking
continue
presence = getDatasetPresence(url, sec)
print sec
print json.dumps(presence, indent=2)
one_secondary_locations = [site for (site, (there, frac)) in presence.items() if frac > 98.0]
# one_secondary_locations = [site for (site,(there,frac)) in presence.items() if there]
if secondary_locations == None:
secondary_locations = one_secondary_locations
else:
secondary_locations = list(set(secondary_locations) & set(one_secondary_locations))
## reduce the site white list to site with secondary only
# sites_allowed = [site for site in sites_allowed if any([osite.startswith(site) for osite in one_secondary_locations])]
sites_allowed = [site for site in sites_allowed if SI.CE_to_SE(site) in one_secondary_locations]
wfh.sendLog("assignor", "From secondary requirement, now Allowed%s" % sorted(sites_allowed))
initial_sites_allowed = copy.deepcopy(
sites_allowed
) ## keep track of this, after secondary input location restriction : that's how you want to operate it
sites_all_data = copy.deepcopy(sites_allowed)
sites_with_data = copy.deepcopy(sites_allowed)
sites_with_any_data = copy.deepcopy(sites_allowed)
primary_locations = None
available_fractions = {}
set_lfn = "/store/mc" ## by default
endpoints = set()
for prim in list(primary):
if prim in dataset_endpoints:
print "endpoints from stagor", dataset_endpoints[prim]
endpoints.update(dataset_endpoints[prim])
set_lfn = getLFNbase(prim)
presence = getDatasetPresence(url, prim, only_blocks=blocks)
if talk:
print prim
print json.dumps(presence, indent=2)
available_fractions[prim] = getDatasetBlocksFraction(
url, prim, sites=[SI.CE_to_SE(site) for site in sites_allowed], only_blocks=blocks
)
# sites_all_data = [site for site in sites_with_data if any([osite.startswith(site) for osite in [psite for (psite,(there,frac)) in presence.items() if there]])]
# sites_with_data = [site for site in sites_with_data if any([osite.startswith(site) for osite in [psite for (psite,frac) in presence.items() if frac[1]>90.]])]
sites_all_data = [
site
for site in sites_with_data
if SI.CE_to_SE(site) in [psite for (psite, (there, frac)) in presence.items() if there]
]
sites_with_data = [
site
for site in sites_with_data
if SI.CE_to_SE(site) in [psite for (psite, frac) in presence.items() if frac[1] > 90.0]
]
sites_with_any_data = [site for site in sites_with_any_data if SI.CE_to_SE(site) in presence.keys()]
wfh.sendLog(
"assignor",
"Holding the data but not allowed %s"
% sorted(
list(set([se_site for se_site in presence.keys() if not SI.SE_to_CE(se_site) in sites_allowed]))
),
)
if primary_locations == None:
primary_locations = presence.keys()
else:
primary_locations = list(set(primary_locations) & set(presence.keys()))
sites_with_data = list(set(sites_with_data))
sites_with_any_data = list(set(sites_with_any_data))
opportunistic_sites = []
down_time = False
## opportunistic running where any piece of data is available
if secondary_locations or primary_locations:
## intersection of both any pieces of the primary and good IO
# opportunistic_sites = [SI.SE_to_CE(site) for site in list((set(secondary_locations) & set(primary_locations) & set(SI.sites_with_goodIO)) - set(sites_allowed))]
if secondary_locations and primary_locations:
opportunistic_sites = [
SI.SE_to_CE(site)
for site in list(
(set(secondary_locations) & set(primary_locations))
- set([SI.CE_to_SE(site) for site in sites_allowed])
)
]
elif primary_locations:
opportunistic_sites = [
SI.SE_to_CE(site)
for site in list(set(primary_locations) - set([SI.CE_to_SE(site) for site in sites_allowed]))
]
else:
opportunistic_sites = []
wfh.sendLog("assignor", "We could be running in addition at %s" % sorted(opportunistic_sites))
if any([osite in SI.sites_not_ready for osite in opportunistic_sites]):
wfh.sendLog(
"assignor",
"One of the usable site is in downtime %s"
% ([osite in SI.sites_not_ready for osite in opportunistic_sites]),
)
down_time = True
## should this be send back to considered ?
## should be 2 but for the time-being let's lower it to get things going
copies_wanted, cpuh = wfh.getNCopies()
wfh.sendLog("assignor", "we need %s CPUh" % cpuh)
if cpuh > max_cpuh_block and not options.go:
# sendEmail('large workflow','that wf %s has a large number of CPUh %s, not assigning, please check the logs'%(wfo.name, cpuh))#,destination=['*****@*****.**'])
sendLog(
"assignor",
"%s requires a large numbr of CPUh %s , not assigning, please check with requester" % (wfo.name, cpuh),
level="critical",
)
wfh.sendLog("assignor", "Requiring a large number of CPUh %s, not assigning" % cpuh)
continue
if (
"Campaign" in wfh.request
and wfh.request["Campaign"] in CI.campaigns
and "maxcopies" in CI.campaigns[wfh.request["Campaign"]]
):
copies_needed_from_campaign = CI.campaigns[wfh.request["Campaign"]]["maxcopies"]
copies_wanted = min(copies_needed_from_campaign, copies_wanted)
if not options.early:
less_copies_than_requested = UC.get("less_copies_than_requested")
copies_wanted = max(1, copies_wanted - less_copies_than_requested) # take one out for the efficiency
else:
## find out whether there is a site in the whitelist, that is lacking jobs and reduce to 1 copy needed to get things going
pass
wfh.sendLog("assignor", "needed availability fraction %s" % copies_wanted)
## should also check on number of sources, if large enough, we should be able to overflow most, efficiently
if available_fractions and not all([available >= copies_wanted for available in available_fractions.values()]):
not_even_once = not all([available >= 1.0 for available in available_fractions.values()])
wfh.sendLog(
"assignor",
"The input dataset is not available %s times, only %s" % (copies_wanted, available_fractions.values()),
)
if down_time and not options.go and not options.early:
wfo.status = "considered"
session.commit()
wfh.sendLog("assignor", "sending back to considered because of site downtime, instead of waiting")
# sendEmail( "cannot be assigned due to downtime","%s is not sufficiently available, due to down time of a site in the whitelist. check the assignor logs. sending back to considered."% wfo.name)
sendLog(
"assignor",
"%s is not sufficiently available, due to down time of a site in the whitelist. sending back to considered."
% (wfo.name),
level="delay",
)
continue
# pass
print json.dumps(available_fractions)
if (options.go and not_even_once) or not options.go:
known = []
try:
known = json.loads(open("cannot_assign.json").read())
except:
pass
if (
not wfo.name in known
and not options.limit
and not options.go
and not options.early
and not options.partial
):
wfh.sendLog(
"assignor",
"cannot be assigned, %s is not sufficiently available.\n %s"
% (wfo.name, json.dumps(available_fractions)),
)
sendEmail(
"cannot be assigned",
"%s is not sufficiently available.\n %s" % (wfo.name, json.dumps(available_fractions)),
)
known.append(wfo.name)
open("cannot_assign.json", "w").write(json.dumps(known, indent=2))
n_stalled += 1
if options.early:
if wfo.status == "considered":
wfh.sendLog("assignor", "setting considered-tried")
wfo.status = "considered-tried"
session.commit()
else:
print "tried but status is", wfo.status
if options.partial:
print "Will move on with partial locations"
else:
continue
## default back to white list to original white list with any data
print "Allowed", sorted(sites_allowed)
if primary_aaa:
sites_allowed = initial_sites_allowed
options.TrustSitelists = True
wfh.sendLog("assignor", "Selected to read primary through xrootd %s" % sorted(sites_allowed))
else:
sites_allowed = sites_with_any_data
wfh.sendLog("assignor", "Selected for any data %s" % sorted(sites_allowed))
if secondary_aaa:
options.TrustPUSitelists = True
wfh.sendLog("assignor", "Reading secondary through xrootd from %s" % sorted(sites_allowed))
### check on endpoints for on-going transfers
if endpoints and options.partial:
sites_allowed = list(set(sites_allowed + [SI.SE_to_CE(s) for s in endpoints]))
print "with added endpoints", sorted(sites_allowed)
if not len(sites_allowed):
wfh.sendLog("assignor", "cannot be assign with no matched sites")
sendLog("assignor", "%s has no whitelist" % wfo.name, level="critical")
n_stalled += 1
continue
t1_only = [ce for ce in sites_allowed if ce.startswith("T1")]
if t1_only:
# try to pick from T1 only first
sites_out = [SI.pick_dSE([SI.CE_to_SE(ce) for ce in t1_only])]
else:
# then pick any otherwise
sites_out = [SI.pick_dSE([SI.CE_to_SE(ce) for ce in sites_allowed])]
wfh.sendLog("assignor", "Placing the output on %s" % sites_out)
parameters = {
"SiteWhitelist": sites_allowed,
"NonCustodialSites": sites_out,
"AutoApproveSubscriptionSites": list(set(sites_out)),
"AcquisitionEra": wfh.acquisitionEra(),
"ProcessingString": wfh.processingString(),
"MergedLFNBase": set_lfn,
"ProcessingVersion": version,
}
## plain assignment here
team = "production"
if os.getenv("UNIFIED_TEAM"):
team = os.getenv("UNIFIED_TEAM")
if options and options.team:
team = options.team
if False and "T2_CH_CERN" in parameters["SiteWhitelist"]:
## add some check on
### the amount pending to HLT
### the size of the request
### the priority of the request (maybe not if we decide to overflow during runs)
parameters["SiteWhitelist"] = ["T2_CH_CERN_HLT"]
team = "hlt"
## reduce the splitting by factor of 4, regardless of type of splitting
sendEmail("sending work to HLT", "%s was assigned to HLT" % wfo.name)
##parse options entered in command line if any
if options:
for key in reqMgrClient.assignWorkflow.keys:
v = getattr(options, key)
if v != None:
if type(v) == str and "," in v:
parameters[key] = filter(None, v.split(","))
else:
parameters[key] = v
if lheinput:
## throttle reading LHE article
wfh.sendLog("assignor", "Setting the number of events per job to 500k max")
parameters["EventsPerJob"] = 500000
## pick up campaign specific assignment parameters
parameters.update(CI.parameters(wfh.request["Campaign"]))
if not options.test:
parameters["execute"] = True
split_check = wfh.checkWorkflowSplitting()
if split_check != True:
parameters.update(split_check)
if "EventBased" in split_check.values():
wfh.sendLog("assignor", "Falling back to event splitting.")
# sendEmail("Fallback to EventBased","the workflow %s is too heavy to be processed as it is. Fallback to EventBased splitting"%wfo.name)
sendLog(
"assignor",
"the workflow %s is too heavy to be processed as it is. Fallback to EventBased splitting"
% wfo.name,
level="critical",
)
elif "EventsPerJob" in split_check.values():
wfh.sendLog("assignor", "Modifying the number of job per event")
# sendEmail("Modifying the job per events","the workflow %s is too heavy in number of jobs explosion"%wfo.name)
sendLog(
"assignor", "the workflow %s is too heavy in number of jobs explosion" % wfo.name, level="critical"
)
# Handle run-dependent MC
pstring = wfh.processingString()
if "PU_RD" in pstring:
numEvents = wfh.getRequestNumEvents()
eventsPerLumi = [getDatasetEventsPerLumi(prim) for prim in primary]
eventsPerLumi = sum(eventsPerLumi) / float(len(eventsPerLumi))
reqJobs = 500
if "PU_RD2" in pstring:
reqJobs = 2000
eventsPerJob = int(numEvents / (reqJobs * 1.4))
lumisPerJob = int(eventsPerJob / eventsPerLumi)
if lumisPerJob == 0:
# sendEmail("issue with event splitting for run-dependent MC","%s needs to be split by event with %s per job"%(wfo.name, eventsPerJob))
sendLog(
"assignor",
"%s needs to be split by event with %s per job" % (wfo.name, eventsPerJob),
level="critical",
)
wfh.sendLog("assignor", "%s needs to be split by event with %s per job" % (wfo.name, eventsPerJob))
parameters["EventsPerJob"] = eventsPerJob
else:
spl = wfh.getSplittings()[0]
eventsPerJobEstimated = spl["events_per_job"] if "events_per_job" in spl else None
eventsPerJobEstimated = spl["avg_events_per_job"] if "avg_events_per_job" in spl else None
if eventsPerJobEstimated and eventsPerJobEstimated > eventsPerJob:
# sendEmail("setting lumi splitting for run-dependent MC","%s was assigned with %s lumis/job"%( wfo.name, lumisPerJob))
sendLog(
"assignor", "%s was assigned with %s lumis/job" % (wfo.name, lumisPerJob), level="critical"
)
wfh.sendLog("assignor", "%s was assigned with %s lumis/job" % (wfo.name, lumisPerJob))
parameters["LumisPerJob"] = lumisPerJob
else:
# sendEmail("leaving splitting untouched for PU_RD*","please check on "+wfo.name)
sendLog(
"assignor",
"leaving splitting untouched for %s, please check on %s" % (pstring, wfo.name),
level="critical",
)
wfh.sendLog("assignor", "leaving splitting untouched for PU_RD*, please check.")
result = reqMgrClient.assignWorkflow(url, wfo.name, team, parameters)
# set status
if not options.test:
if result:
wfo.status = "away"
session.commit()
n_assigned += 1
wfh.sendLog("assignor", "Properly assigned\n%s" % (json.dumps(parameters, indent=2)))
try:
## refetch information and lock output
new_wfi = workflowInfo(url, wfo.name)
(_, prim, _, sec) = new_wfi.getIO()
for secure in list(prim) + list(sec) + new_wfi.request["OutputDatasets"]:
## lock all outputs flat
NLI.lock(secure)
# for site in [SI.CE_to_SE(site) for site in sites_allowed]:
# for output in new_wfi.request['OutputDatasets']:
# LI.lock( output, site, 'dataset in production')
# for primary in prim:
# LI.lock( primary, site, 'dataset used in input')
# for secondary in sec:
# LI.lock( secondary, site, 'required for mixing' )
except Exception as e:
print "fail in locking output"
print str(e)
sendEmail("failed locking of output", str(e))
else:
print "ERROR could not assign", wfo.name
else:
pass
print "Assignment summary:"
sendLog("assignor", "Assigned %d Stalled %s" % (n_assigned, n_stalled))
unlock = True
else:
print "keep a lock on secondary within",delay_days,"days"
unlock = False
newly_locking.add(dataset)
continue
time_point("Checked as useful secondary", sub_lap=True)
tier = dataset.split('/')[-1]
creators = getWorkflowByOutput( url, dataset , details=True)
if not creators and not tier == 'RAW' and not '-PromptReco-' in dataset:
ds_status = getDatasetStatus( dataset )
if not '-v0/' in dataset and ds_status!=None:
#sendEmail('failing get by output','%s has not been produced by anything?'%dataset)
sendLog('lockor','failing get by output, %s has not been produced by anything?'%dataset, level='critical')
newly_locking.add(dataset)
continue
else:
# does not matter, cannot be an OK dataset
unlock = True
bad_ds = True
creators_status = [r['RequestStatus'] for r in creators]
print "Statuses of workflow that made the dataset",dataset,"are",creators_status
if len(creators_status) and all([status in ['failed','aborted','rejected','aborted-archived','rejected-archived'] for status in creators_status]):
## crap
print "\tunlocking",dataset,"for bad workflow statuses"
unlock = True
bad_ds = True
time_point("Check as necessary output", sub_lap=True)
# #os.system('Unified/rejector.py --clone --to_step %s --comments "Transforming into StepChain for efficiency"'%( wfo.name))
# #os.system('Unified/injector.py %s'% wfi.request['PrepID'])
# pass
### all dqmharvest completed to announced right away ###
wfs = getWorkflows(url, 'completed', user=None, rtype='DQMHarvest')
for wf in wfs:
print "closing out",wf
reqMgrClient.closeOutWorkflow(url, wf)
wfs = getWorkflows(url, 'closed-out', user=None, rtype='DQMHarvest')
for wf in wfs:
print "announcing",wf
reqMgrClient.announceWorkflow(url, wf)
wfs = getWorkflows(url, 'failed', user=None, rtype='DQMHarvest')
if len(wfs):
sendLog('addHoc','There are failed Harvesting requests\n%s'%('\n'.join(sorted( wfs))),level='critical')
### clone all DR80 that end up with issues ###
##
#for wfo in session.query(Workflow).filter(Workflow.status == 'assistance-manual').all():
# if not any([c in wfo.name for c in ['RunIISpring16DR80']]): continue
# wfi = workflowInfo(url, wfo.name)
# if wfi.getRequestNumEvents() < 500000:
# ## small workflow that needs recovery : kill-clone
# os.system('Unified/rejector.py --clone %s'%wfo.name)
"""
### add the value of the delay to announcing datasets
data = json.loads(open('%s/announce_delays.json'%monitor_dir).read())
def invalidator(url, invalid_status='INVALID'):
use_mcm = True
up = componentInfo(soft=['wtc','jira'])
if not up.check(): return
mcm = McMClient(dev=False)
invalids = mcm.getA('invalidations',query='status=announced')
if not invalids: return
print len(invalids),"Object to be invalidated"
text_to_batch = defaultdict(str)
text_to_request = defaultdict(str)
for invalid in invalids:
acknowledge= False
pid = invalid['prepid']
batch_lookup = invalid['prepid']
text = ""
if invalid['type'] == 'request':
wfn = invalid['object']
print "need to invalidate the workflow",wfn
wfo = session.query(Workflow).filter(Workflow.name == wfn).first()
if wfo:
## set forget of that thing (although checkor will recover from it)
print "setting the status of",wfo.status,"to forget"
wfo.status = 'forget'
session.commit()
else:
## do not go on like this, do not acknoledge it
print wfn,"is set to be rejected, but we do not know about it yet"
#continue
wfi = workflowInfo(url, wfn)
success = "not rejected"
## to do, we should find a way to reject the workflow and any related acdc
successes = invalidate(url, wfi, only_resub=True, with_output=False)
wfi.sendLog('invalidator',"rejection is performed from McM invalidations request")
acknowledge= all(successes)
text = "The workflow %s (%s) was rejected due to invalidation in McM" % ( wfn, pid )
batch_lookup = wfn ##so that the batch id is taken as the one containing the workflow name
elif invalid['type'] == 'dataset':
dataset = invalid['object']
if '?' in dataset: continue
if 'None' in dataset: continue
if 'None-' in dataset: continue
if 'FAKE-' in dataset: continue
print "setting",dataset,"to",invalid_status
success = setDatasetStatus(dataset , invalid_status )
if success:
acknowledge= True
text = "The dataset %s (%s) was set INVALID due to invalidation in McM" % ( dataset, pid )
else:
msg = "Could not invalidate {}. Please consider contacting data management team for manual intervention.".format(dataset)
print(msg)
sendLog('invalidator', msg, level='critical')
else:
print "\t\t",invalid['type']," type not recognized"
if acknowledge:
## acknoldge invalidation in mcm, provided we can have the api
print "acknowledgment to mcm"
ackno_url = '/restapi/invalidations/acknowledge/%s'%( invalid['_id'] )
print "at",ackno_url
mcm.get(ackno_url)
# prepare the text for batches
batches = []
batches.extend(mcm.getA('batches',query='contains=%s'%batch_lookup))
batches = filter(lambda b : b['status'] in ['announced','done','reset'], batches)
if len(batches):
bid = batches[-1]['prepid']
print "batch nofication to",bid
text_to_batch[bid] += text+"\n\n"
# prepare the text for requests
text_to_request[pid] += text+"\n\n"
for bid,text in text_to_batch.items():
if not text: continue
text += '\n This is an automated message'
mcm.put('/restapi/batches/notify',{ "notes" : text, "prepid" : bid})
pass
for pid,text in text_to_request.items():
if not text: continue
text += '\n This is an automated message'
mcm.put('/restapi/requests/notify',{ "message" : text, "prepids" : [pid]})
def singleClone(url, wfname, actions, comment, do=False):
wfi = workflowInfo(url, wfname)
payload = wfi.getSchema()
initial = wfi.request
payload['Requestor'] = os.getenv('USER')
payload['Group'] = 'DATAOPS'
payload['OriginalRequestName'] = initial['RequestName']
payload['RequestPriority'] = initial['RequestPriority']
if 'ProcessingVersion' in initial:
payload['ProcessingVersion'] = int(initial['ProcessingVersion']) +1
else:
payload['ProcessingVersion'] = 2
## drop parameters on the way to reqmgr2
paramBlacklist = ['BlockCloseMaxEvents', 'BlockCloseMaxFiles', 'BlockCloseMaxSize', 'BlockCloseMaxWaitTime',
'CouchWorkloadDBName', 'CustodialGroup', 'CustodialSubType', 'Dashboard',
'GracePeriod', 'HardTimeout', 'InitialPriority', 'inputMode', 'MaxMergeEvents', 'MaxMergeSize',
'MaxRSS', 'MaxVSize', 'MinMergeSize', 'NonCustodialGroup', 'NonCustodialSubType',
'OutputDatasets', 'ReqMgr2Only', 'RequestDate' 'RequestorDN', 'RequestName', 'RequestStatus',
'RequestTransition', 'RequestWorkflow', 'SiteWhitelist', 'SoftTimeout', 'SoftwareVersions',
'SubscriptionPriority', 'Team', 'timeStamp', 'TrustSitelists', 'TrustPUSitelists',
'TotalEstimatedJobs', 'TotalInputEvents', 'TotalInputLumis', 'TotalInputFiles','checkbox',
'DN', 'AutoApproveSubscriptionSites', 'NonCustodialSites', 'CustodialSites', 'OriginalRequestName', 'Teams', 'OutputModulesLFNBases',
'SiteBlacklist', 'AllowOpportunistic', '_id']
for p in paramBlacklist:
if p in payload:
payload.pop( p )
pass
if actions:
for action in actions:
if action.startswith('mem') and actions[action] != "" and actions[action] != 'Same':
if 'TaskChain' in payload:
print "Setting memory for clone of task chain"
it=1
while True:
t = 'Task%d'%it
it+=1
if t in payload:
payload[t]['Memory'] = actions[action]
print "Memory set for Task%d"%it
else:
break
else:
print "Setting memory for non-taskchain workflow"
payload['Memory'] = actions[action]
print "Memory set to " + actions[action]
#This line is doesn't work for some reason
# wfi.sendLog('actor','Memory of clone set to %d'%actions[action])
print "Clone payload"
# print json.dumps( payload , indent=2)
print actions
#Create clone
clone = reqMgrClient.submitWorkflow(url, payload)
if not clone:
print "Error in making clone for",initial["RequestName"]
clone = reqMgrClient.submitWorkflow(url, payload)
if not clone:
print "Error twice in making clone for",initial["RequestName"]
sendLog('actor','Failed to make a clone twice for %s!'%initial["RequestName"],level='critical')
wfi.sendLog('actor','Failed to make a clone twice for %s!'%initial["RequestName"])
return None
if actions:
for action in actions:
if action.startswith('split'):
cloneinfo = workflowInfo(url, clone)
splittings = cloneinfo.getSplittings()
if actions[action] != 'Same' and actions[action] != 'max' and actions[action] != '':
factor = int(actions[action][0:-1]) if 'x' in actions[action] else 2
for split in splittings:
for act in ['avg_events_per_job','events_per_job','lumis_per_job']:
if act in split:
wfi.sendLog('actor','Changing %s (%d) by a factor %d'%( act, split[act], factor))
print "Changing %s (%d) by a factor %d"%( act, split[act], factor),
split[act] /= factor
print "to",split[act]
break
split['requestName'] = clone
print "changing the splitting of",clone
print json.dumps( split, indent=2 )
print reqMgrClient.setWorkflowSplitting(url, clone, split )
elif 'max' in actions[action]:
for split in splittings:
for act in ['avg_events_per_job','events_per_job','lumis_per_job']:
if act in split:
wfi.sendLog('actor','Max splitting set for %s (%d'%( act, split[act]))
print "Changing %s (%d) "%( act, split[act]),
split[act] = 1
print "to max splitting ",split[act]
break
split['requestName'] = clone
print "changing the splitting of",clone
print json.dumps( split, indent=2 )
print reqMgrClient.setWorkflowSplitting(url, clone, split )
#Approve
data = reqMgrClient.setWorkflowApproved(url, clone)
wfi.sendLog('actor','Cloned into %s'%clone)
# wfi.sendLog('actor','Cloned into %s by unified operator %s'%( clone, comment ))
# wfi.notifyRequestor('Cloned into %s by unified operator %s'%( clone, comment ),do_batch=False)
print data
return clone
task_path_name_limit = UC.get("full_task_path_name_limit")
for wf in wfs:
if spec and not spec in wf['RequestName']: continue
wfi = workflowInfo(url, wf['RequestName'], request=wf)
splits = wfi.getSplittingsNew(all_tasks=True)
if any([len(s['taskName']) > task_path_name_limit for s in splits]):
msg = "The full task path name exceeds the limit of {} characters. The requestor should either shorten task names or have fewer tasks in the chain.".format(
task_path_name_limit)
wfi.sendLog("GQ", msg)
msg += '\nWorkflow URL: https://dmytro.web.cern.ch/dmytro/cmsprodmon/workflows.php?prep_id=task_{}'.format(
wfi.getPrepIDs()[0])
sendLog("GQ", msg, level='critical')
continue
sitewhitelist = wfi.request['SiteWhitelist']
wqs = wfi.getWorkQueue()
stats = wfi.getWMStats()
if not 'AgentJobInfo' in stats: stats['AgentJobInfo'] = {}
## skip wf that unified does not know about, leaves acdc
wfo = session.query(Workflow).filter(
Workflow.name == wf['RequestName']).first()
if not (wfo or wf['RequestType'] == 'Resubmission'):
print "not knonw or not acdc : %s" % (wf['RequestName'])
continue
def singleRecovery(url, task, initial, actions, do=False):
print "Inside single recovery!"
payload = {
"Requestor" : os.getenv('USER'),
"Group" : 'DATAOPS',
"RequestType" : "Resubmission",
"ACDCServer" : initial['CouchURL'],
"ACDCDatabase" : "acdcserver",
"OriginalRequestName" : initial['RequestName'],
"OpenRunningTimeout" : 0
}
copy_over = ['PrepID','Campaign','RequestPriority', 'TimePerEvent', 'SizePerEvent', 'Group', 'Memory', 'RequestString' ,'CMSSWVersion']
for c in copy_over:
if c in initial:
payload[c] = copy.deepcopy(initial[c])
else:
print c,"not in the initial payload"
#a massage ? boost the recovery over the initial wf
# payload['RequestPriority'] *= 10
#Max priority is 1M
payload['RequestPriority'] = min(500000, payload['RequestPriority']*2 ) ## never above 500k
#change parameters based on actions here
if actions:
for action in actions:
if action.startswith('mem') and actions[action] != "" and actions[action] != 'Same':
payload['Memory'] = actions[action]
print "Memory set to " + actions[action]
## Taskchains needs to be treated special to set the memory to all tasks
if 'TaskChain' in initial:
it = 1
while True:
t = 'Task%d'%it
it += 1
if t in initial:
payload[t] = copy.deepcopy(initial[t])
payload[t]['Memory'] = actions[action]
else:
break
if action.startswith('split'):
split_alert = (initial['RequestType'] in ['MonteCarlo'] )
for key in initial:
if key == 'SplittingAlgo' and (initial[key] in ['EventBased']):
split_alert = True
elif key.startswith('Task') and key != 'TaskChain':
for key2 in initial[key]:
if key2 == 'TaskName':
print "task",task.split('/')[-1]
print "TaskName",initial[key][key2]
if (initial[key][key2] == task) and (initial[key][key2] in ['EventBased']):
split_alert = True
if split_alert:
sendLog('actor','Cannot change splitting for %s'%initial['RequestName'],level='warning')
print "I should not be doing splitting for this type of request",initial['RequestName']
return None
acdc_round = 0
initial_string = payload['RequestString']
if initial_string.startswith('ACDC'):
if initial_string[4].isdigit():
acdc_round = int(initial_string[4])
acdc_round += 1
initial_string = initial_string.replace('ACDC_','').replace('ACDC%d_'%(acdc_round-1),'')
payload['RequestString'] = 'ACDC%d_%s'%(acdc_round,initial_string)
payload['InitialTaskPath'] = task
if not do:
print json.dumps( payload, indent=2)
return None
print "ACDC payload"
# print json.dumps( payload , indent=2)
print actions
## submit here
acdc = reqMgrClient.submitWorkflow(url, payload)
if not acdc:
print "Error in making ACDC for",initial["RequestName"]
acdc = reqMgrClient.submitWorkflow(url, payload)
if not acdc:
print "Error twice in making ACDC for",initial["RequestName"]
sendLog('actor','Failed twice in making ACDCs for %s!'%initial['RequestName'],level='critical')
return None
## change splitting if requested
if actions:
for action in actions:
if action.startswith('split'):
acdcInfo = workflowInfo(url, acdc)
splittings = acdcInfo.getSplittings()
if actions[action] != 'Same' and actions[action] != 'max':
factor = int(actions[action][0:-1]) if 'x' in actions[action] else 2
for split in splittings:
for act in ['avg_events_per_job','events_per_job','lumis_per_job']:
if act in split:
print "Changing %s (%d) by a factor %d"%( act, split[act], factor),
split[act] /= factor
print "to",split[act]
break
split['requestName'] = acdc
print "changing the splitting of",acdc
print json.dumps( split, indent=2 )
print reqMgrClient.setWorkflowSplitting(url, acdc, split )
elif 'max' in actions[action]:
for split in splittings:
for act in ['avg_events_per_job','events_per_job','lumis_per_job']:
if act in split:
print "Changing %s (%d) "%( act, split[act]),
split[act] = 1
print "to max splitting ",split[act]
break
split['requestName'] = acdc
print "changing the splitting of",acdc
print json.dumps( split, indent=2 )
print reqMgrClient.setWorkflowSplitting(url, acdc, split )
data = reqMgrClient.setWorkflowApproved(url, acdc)
print data
return acdc
def assignor(url ,specific = None, talk=True, options=None):
if userLock(): return
if duplicateLock(): return
if not componentInfo().check(): return
UC = unifiedConfiguration()
CI = campaignInfo()
SI = global_SI
#LI = lockInfo()
NLI = newLockInfo()
n_assigned = 0
n_stalled = 0
wfos=[]
if specific or options.early:
wfos.extend( session.query(Workflow).filter(Workflow.status=='considered').all())
wfos.extend( session.query(Workflow).filter(Workflow.status=='staging').all())
if specific:
wfos.extend( session.query(Workflow).filter(Workflow.status=='considered-tried').all())
wfos.extend(session.query(Workflow).filter(Workflow.status=='staged').all())
dataset_endpoints = json.loads(open('%s/dataset_endpoints.json'%monitor_dir).read())
max_per_round = UC.get('max_per_round').get('assignor',None)
max_cpuh_block = UC.get('max_cpuh_block')
random.shuffle( wfos )
for wfo in wfos:
if options.limit and (n_stalled+n_assigned)>options.limit:
break
if max_per_round and (n_stalled+n_assigned)>max_per_round:
break
if specific:
if not any(map(lambda sp: sp in wfo.name,specific.split(','))): continue
#if not specific in wfo.name: continue
print "\n\n"
wfh = workflowInfo( url, wfo.name)
wfh.sendLog('assignor',"%s to be assigned"%wfo.name)
## the site whitelist takes into account siteInfo, campaignInfo, memory and cores
(lheinput,primary,parent,secondary, sites_allowed) = wfh.getSiteWhiteList()
## check if by configuration we gave it a GO
no_go = False
allowed_secondary = set()
for campaign in wfh.getCampaigns():
if not CI.go( campaign ):
wfh.sendLog('assignor',"No go for %s"%campaign)
if not options.go:
n_stalled+=1
no_go = True
break
if campaign in CI.campaigns and 'secondaries' in CI.campaigns[campaign]:
allowed_secondary.update( CI.campaigns[campaign]['secondaries'] )
if (secondary and allowed_secondary) and (set(secondary)&allowed_secondary!=set(secondary)):
wfh.sendLog('assignor','%s is not an allowed secondary'%(', '.join(set(secondary)-allowed_secondary)))
#sendEmail('secondary not allowed','%s is not an allowed secondary'%( ', '.join(set(secondary)-allowed_secondary)))
sendLog('assignor','%s is not an allowed secondary'%(', '.join(set(secondary)-allowed_secondary)), level='critical')
if not options.go:
n_stalled+=1
no_go = True
if no_go:
continue
## check on current status for by-passed assignment
if wfh.request['RequestStatus'] !='assignment-approved':
if not options.test:
wfh.sendLog('assignor',"setting %s away and skipping"%wfo.name)
## the module picking up from away will do what is necessary of it
wfo.wm_status = wfh.request['RequestStatus']
wfo.status = 'away'
session.commit()
continue
else:
print wfo.name,wfh.request['RequestStatus']
## retrieve from the schema, dbs and reqMgr what should be the next version
version=wfh.getNextVersion()
if not version:
if options and options.ProcessingVersion:
version = options.ProcessingVersion
else:
wfh.sendLog('assignor',"cannot decide on version number")
n_stalled+=1
wfo.status = 'trouble'
session.commit()
continue
original_sites_allowed = copy.deepcopy( sites_allowed )
wfh.sendLog('assignor',"Site white list %s"%sorted(sites_allowed))
override_sec_location = CI.get(wfh.request['Campaign'], 'SecondaryLocation', [])
blocks = []
if 'BlockWhitelist' in wfh.request:
blocks = wfh.request['BlockWhitelist']
if 'RunWhitelist' in wfh.request and wfh.request['RunWhitelist']:
## augment with run white list
for dataset in primary:
blocks = list(set( blocks + getDatasetBlocks( dataset, runs=wfh.request['RunWhitelist'] ) ))
wfh.sendLog('assignor',"Allowed %s"%sorted(sites_allowed))
secondary_locations=None
for sec in list(secondary):
if override_sec_location:
print "We don't care where the secondary is"
print "Cannot pass for now"
sendEmail("tempting to pass sec location check","but we cannot yet IMO")
#pass
presence = getDatasetPresence( url, sec )
print sec
print json.dumps(presence, indent=2)
one_secondary_locations = [site for (site,(there,frac)) in presence.items() if frac>98.]
#one_secondary_locations = [site for (site,(there,frac)) in presence.items() if there]
if secondary_locations==None:
secondary_locations = one_secondary_locations
else:
secondary_locations = list(set(secondary_locations) & set(one_secondary_locations))
## reduce the site white list to site with secondary only
#sites_allowed = [site for site in sites_allowed if any([osite.startswith(site) for osite in one_secondary_locations])]
sites_allowed = [site for site in sites_allowed if SI.CE_to_SE(site) in one_secondary_locations]
wfh.sendLog('assignor',"From secondary requirement, now Allowed%s"%sorted(sites_allowed))
initial_sites_allowed = copy.deepcopy( sites_allowed ) ## keep track of this, after secondary input location restriction : that's how you want to operate it
sites_all_data = copy.deepcopy( sites_allowed )
sites_with_data = copy.deepcopy( sites_allowed )
sites_with_any_data = copy.deepcopy( sites_allowed )
primary_locations = None
available_fractions = {}
set_lfn = '/store/mc' ## by default
endpoints = set()
for prim in list(primary):
if prim in dataset_endpoints:
print "endpoints from stagor",dataset_endpoints[prim]
endpoints.update( dataset_endpoints[prim] )
set_lfn = getLFNbase( prim )
presence = getDatasetPresence( url, prim , only_blocks=blocks)
if talk:
print prim
print json.dumps(presence, indent=2)
available_fractions[prim] = getDatasetBlocksFraction(url, prim, sites = [SI.CE_to_SE(site) for site in sites_allowed] , only_blocks = blocks)
#sites_all_data = [site for site in sites_with_data if any([osite.startswith(site) for osite in [psite for (psite,(there,frac)) in presence.items() if there]])]
#sites_with_data = [site for site in sites_with_data if any([osite.startswith(site) for osite in [psite for (psite,frac) in presence.items() if frac[1]>90.]])]
sites_all_data = [site for site in sites_with_data if SI.CE_to_SE(site) in [psite for (psite,(there,frac)) in presence.items() if there]]
sites_with_data = [site for site in sites_with_data if SI.CE_to_SE(site) in [psite for (psite,frac) in presence.items() if frac[1]>90.]]
sites_with_any_data = [site for site in sites_with_any_data if SI.CE_to_SE(site) in presence.keys()]
wfh.sendLog('assignor',"Holding the data but not allowed %s"%sorted(list(set([se_site for se_site in presence.keys() if not SI.SE_to_CE(se_site) in sites_allowed]))))
if primary_locations==None:
primary_locations = presence.keys()
else:
primary_locations = list(set(primary_locations) & set(presence.keys() ))
sites_with_data = list(set(sites_with_data))
sites_with_any_data = list(set(sites_with_any_data))
opportunistic_sites=[]
down_time = False
## opportunistic running where any piece of data is available
if secondary_locations or primary_locations:
## intersection of both any pieces of the primary and good IO
#opportunistic_sites = [SI.SE_to_CE(site) for site in list((set(secondary_locations) & set(primary_locations) & set(SI.sites_with_goodIO)) - set(sites_allowed))]
if secondary_locations and primary_locations:
opportunistic_sites = [SI.SE_to_CE(site) for site in list((set(secondary_locations) & set(primary_locations)) - set([SI.CE_to_SE(site) for site in sites_allowed]))]
elif primary_locations:
opportunistic_sites = [SI.SE_to_CE(site) for site in list(set(primary_locations) - set([SI.CE_to_SE(site) for site in sites_allowed]))]
else:
opportunistic_sites = []
wfh.sendLog('assignor',"We could be running in addition at %s"% sorted(opportunistic_sites))
if any([osite in SI.sites_not_ready for osite in opportunistic_sites]):
wfh.sendLog('assignor',"One of the usable site is in downtime %s"%([osite in SI.sites_not_ready for osite in opportunistic_sites]))
down_time = True
## should this be send back to considered ?
## should be 2 but for the time-being let's lower it to get things going
copies_wanted,cpuh = wfh.getNCopies()
wfh.sendLog('assignor',"we need %s CPUh"%cpuh)
if cpuh>max_cpuh_block and not options.go:
#sendEmail('large workflow','that wf %s has a large number of CPUh %s, not assigning, please check the logs'%(wfo.name, cpuh))#,destination=['*****@*****.**'])
sendLog('assignor','%s requires a large numbr of CPUh %s , not assigning, please check with requester'%( wfo.name, cpuh), level='critical')
wfh.sendLog('assignor',"Requiring a large number of CPUh %s, not assigning"%cpuh)
continue
if 'Campaign' in wfh.request and wfh.request['Campaign'] in CI.campaigns and 'maxcopies' in CI.campaigns[wfh.request['Campaign']]:
copies_needed_from_campaign = CI.campaigns[wfh.request['Campaign']]['maxcopies']
copies_wanted = min(copies_needed_from_campaign, copies_wanted)
if not options.early:
less_copies_than_requested = UC.get("less_copies_than_requested")
copies_wanted = max(1,copies_wanted-less_copies_than_requested) # take one out for the efficiency
else:
## find out whether there is a site in the whitelist, that is lacking jobs and reduce to 1 copy needed to get things going
pass
wfh.sendLog('assignor',"needed availability fraction %s"% copies_wanted)
## should also check on number of sources, if large enough, we should be able to overflow most, efficiently
if available_fractions and not all([available>=copies_wanted for available in available_fractions.values()]):
not_even_once = not all([available>=1. for available in available_fractions.values()])
wfh.sendLog('assignor',"The input dataset is not available %s times, only %s"%( copies_wanted, available_fractions.values()))
if down_time and not options.go and not options.early:
wfo.status = 'considered'
session.commit()
wfh.sendLog('assignor',"sending back to considered because of site downtime, instead of waiting")
#sendEmail( "cannot be assigned due to downtime","%s is not sufficiently available, due to down time of a site in the whitelist. check the assignor logs. sending back to considered."% wfo.name)
sendLog('assignor','%s is not sufficiently available, due to down time of a site in the whitelist. sending back to considered.'%( wfo.name ), level='delay')
continue
#pass
print json.dumps(available_fractions)
if (options.go and not_even_once) or not options.go:
known = []
try:
known = json.loads(open('cannot_assign.json').read())
except:
pass
if not wfo.name in known and not options.limit and not options.go and not options.early and not options.partial:
wfh.sendLog('assignor',"cannot be assigned, %s is not sufficiently available.\n %s"%(wfo.name,json.dumps(available_fractions)))
sendEmail( "cannot be assigned","%s is not sufficiently available.\n %s"%(wfo.name,json.dumps(available_fractions)))
known.append( wfo.name )
open('cannot_assign.json','w').write(json.dumps( known, indent=2))
n_stalled+=1
if options.early:
if wfo.status == 'considered':
wfh.sendLog('assignor',"setting considered-tried")
wfo.status = 'considered-tried'
session.commit()
else:
print "tried but status is",wfo.status
if options.partial:
print "Will move on with partial locations"
else:
continue
## default back to white list to original white list with any data
print "Allowed",sorted(sites_allowed)
if options.primary_aaa:
sites_allowed = initial_sites_allowed
#options.useSiteListAsLocation = True
options.TrustSitelists = True
else:
sites_allowed = sites_with_any_data
wfh.sendLog('assignor',"Selected for any data %s"%sorted(sites_allowed))
if options.restrict:
print "Allowed",sites_allowed
sites_allowed = sites_with_any_data
print "Selected",sites_allowed
else:
if set(sites_with_data) != set(sites_allowed):
## the data is not everywhere we wanted to run at : enable aaa
print "Sites with 90% data not matching site white list (block choping!)"
print "Resorting to AAA reading for",list(set(sites_allowed) - set(sites_with_data)),"?"
print "Whitelist site with any data",list(set(sites_allowed) - set(sites_with_any_data))
#options.useSiteListAsLocation = True
#print "Not commissioned yet"
#continue
#print "We could be running at",opportunistic_sites,"in addition"
##sites_allowed = list(set(sites_allowed+ opportunistic_sites))
### check on endpoints for on-going transfers
if endpoints and options.partial:
sites_allowed = list(set(sites_allowed + [SI.SE_to_CE(s) for s in endpoints]))
print "with added endpoints",sorted(sites_allowed)
#if options.partial:
# continue
if not len(sites_allowed):
wfh.sendLog('assignor',"cannot be assign with no matched sites")
#sendEmail( "cannot be assigned","%s has no whitelist"%(wfo.name))
sendLog('assignor','%s has no whitelist'% wfo.name, level='critical')
n_stalled+=1
continue
t1_only = [ce for ce in sites_allowed if ce.startswith('T1')]
if t1_only:
# try to pick from T1 only first
sites_out = [SI.pick_dSE([SI.CE_to_SE(ce) for ce in t1_only])]
else:
# then pick any otherwise
sites_out = [SI.pick_dSE([SI.CE_to_SE(ce) for ce in sites_allowed])]
## one last modification now that we know we can assign, and to make sure all ressource can be used by the request : set all ON sites to whitelist
###sites_allowed = original_sites_allowed ## not needed, afterall as secondary jobs go their own ways
wfh.sendLog('assignor',"Placing the output on %s"%sites_out)
parameters={
'SiteWhitelist' : sites_allowed,
#'CustodialSites' : sites_custodial,
'NonCustodialSites' : sites_out,
'AutoApproveSubscriptionSites' : list(set(sites_out)),
'AcquisitionEra' : wfh.acquisitionEra(),
'ProcessingString' : wfh.processingString(),
'MergedLFNBase' : set_lfn,
'ProcessingVersion' : version,
}
## plain assignment here
team='production'
if os.getenv('UNIFIED_TEAM'): team = os.getenv('UNIFIED_TEAM')
if options and options.team:
team = options.team
## high priority team agent
#if wfh.request['RequestPriority'] >= 100000 and (wfh.request['TimePerEvent']*int(wfh.getRequestNumEvents()))/(8*3600.) < 10000:
# team = 'highprio'
# sendEmail("sending work with highprio team","%s"% wfo.name, destination=['*****@*****.**'])
## SDSC redirection
#if "T2_US_UCSD" in sites_with_data and random.random() < -0.5 and wfh.request['Campaign']=='RunIISpring15DR74' and int(wfh.getRequestNumEvents()) < 600000 and not any([out.endswith('RAW') for out in wfh.request['OutputDatasets']]):
# ## consider SDSC
# parameters['SiteWhitelist'] = ['T2_US_UCSD','T3_US_SDSC']
# parameters['useSiteListAsLocation'] = True
# team = 'allocation-based'
# sendEmail("sending work to SDSC","%s was assigned to SDSC/UCSD"% wfo.name, destination=['*****@*****.**'])
## SDSC redirection
#if wfh.request['Campaign']==R'unIIWinter15GS' and random.random() < -1.0:
# parameters['SiteWhitelist'] = ['T3_US_SDSC']
# team = 'allocation-based'
# sendEmail("sending work to SDSC","%s was assigned to SDSC"% wfo.name, destination=['*****@*****.**'])
if False and 'T2_CH_CERN' in parameters['SiteWhitelist']:
## add some check on
### the amount pending to HLT
### the size of the request
### the priority of the request (maybe not if we decide to overflow during runs)
parameters['SiteWhitelist'] = ['T2_CH_CERN_HLT']
team = 'hlt'
## reduce the splitting by factor of 4, regardless of type of splitting
sendEmail("sending work to HLT","%s was assigned to HLT"%wfo.name)
##parse options entered in command line if any
if options:
for key in reqMgrClient.assignWorkflow.keys:
v=getattr(options,key)
if v!=None:
if type(v)==str and ',' in v:
parameters[key] = filter(None,v.split(','))
else:
parameters[key] = v
## pick up campaign specific assignment parameters
parameters.update( CI.parameters(wfh.request['Campaign']) )
if not options.test:
parameters['execute'] = True
split_check = wfh.checkWorkflowSplitting()
if split_check!=True:
parameters.update( split_check )
if 'EventBased' in split_check.values():
wfh.sendLog('assignor', "Falling back to event splitting.")
#sendEmail("Fallback to EventBased","the workflow %s is too heavy to be processed as it is. Fallback to EventBased splitting"%wfo.name)
sendLog('assignor','the workflow %s is too heavy to be processed as it is. Fallback to EventBased splitting'%wfo.name, level='critical')
elif 'EventsPerJob' in split_check.values():
wfh.sendLog('assignor', "Modifying the number of job per event")
#sendEmail("Modifying the job per events","the workflow %s is too heavy in number of jobs explosion"%wfo.name)
sendLog('assignor',"the workflow %s is too heavy in number of jobs explosion"%wfo.name, level='critical')
# Handle run-dependent MC
pstring = wfh.processingString()
if 'PU_RD' in pstring:
numEvents = wfh.getRequestNumEvents()
eventsPerLumi = [getDatasetEventsPerLumi(prim) for prim in primary]
eventsPerLumi = sum(eventsPerLumi)/float(len(eventsPerLumi))
reqJobs = 500
if 'PU_RD2' in pstring:
reqJobs = 2000
eventsPerJob = int(numEvents/(reqJobs*1.4))
lumisPerJob = int(eventsPerJob/eventsPerLumi)
if lumisPerJob==0:
#sendEmail("issue with event splitting for run-dependent MC","%s needs to be split by event with %s per job"%(wfo.name, eventsPerJob))
sendLog('assignor', "%s needs to be split by event with %s per job"%(wfo.name, eventsPerJob), level='critical')
wfh.sendLog('assignor', "%s needs to be split by event with %s per job"%(wfo.name, eventsPerJob))
parameters['EventsPerJob'] = eventsPerJob
else:
spl = wfh.getSplittings()[0]
eventsPerJobEstimated = spl['events_per_job'] if 'events_per_job' in spl else None
eventsPerJobEstimated = spl['avg_events_per_job'] if 'avg_events_per_job' in spl else None
if eventsPerJobEstimated and eventsPerJobEstimated > eventsPerJob:
#sendEmail("setting lumi splitting for run-dependent MC","%s was assigned with %s lumis/job"%( wfo.name, lumisPerJob))
sendLog('assignor',"%s was assigned with %s lumis/job"%( wfo.name, lumisPerJob), level='critical')
wfh.sendLog('assignor',"%s was assigned with %s lumis/job"%( wfo.name, lumisPerJob))
parameters['LumisPerJob'] = lumisPerJob
else:
#sendEmail("leaving splitting untouched for PU_RD*","please check on "+wfo.name)
sendLog('assignor',"leaving splitting untouched for %s, please check on %s"%( pstring, wfo.name), level='critical')
wfh.sendLog('assignor',"leaving splitting untouched for PU_RD*, please check.")
result = reqMgrClient.assignWorkflow(url, wfo.name, team, parameters)
# set status
if not options.test:
if result:
wfo.status = 'away'
session.commit()
n_assigned+=1
try:
## refetch information and lock output
new_wfi = workflowInfo( url, wfo.name)
(_,prim,_,sec) = new_wfi.getIO()
for secure in list(prim)+list(sec)+new_wfi.request['OutputDatasets']:
## lock all outputs flat
NLI.lock( secure )
#for site in [SI.CE_to_SE(site) for site in sites_allowed]:
# for output in new_wfi.request['OutputDatasets']:
# LI.lock( output, site, 'dataset in production')
# for primary in prim:
# LI.lock( primary, site, 'dataset used in input')
# for secondary in sec:
# LI.lock( secondary, site, 'required for mixing' )
except Exception as e:
print "fail in locking output"
print str(e)
sendEmail("failed locking of output",str(e))
else:
print "ERROR could not assign",wfo.name
else:
pass
print "Assignment summary:"
sendLog('assignor',"Assigned %d Stalled %s"%(n_assigned, n_stalled))
def injector(url, options, specific):
mlock = moduleLock()
if mlock(): return
use_mcm = True
up = componentInfo(soft=['mcm','wtc','jira'] )
if not up.check(): return
use_mcm = up.status['mcm']
UC = unifiedConfiguration()
transform_keywords = UC.get('convert_to_stepchain')
workflows = getWorkflows(url, status=options.wmstatus, user=options.user)
for user in UC.get("user_rereco"):
workflows.extend( getWorkflows(url, status=options.wmstatus, user=user, rtype="ReReco"))
for user in (options.user_relval.split(',') if options.user_relval else UC.get("user_relval")) :
workflows.extend( getWorkflows(url, status=options.wmstatus, user=user, rtype="TaskChain"))
for user in (options.user_storeresults.split(',') if options.user_storeresults else UC.get("user_storeresults")) :
workflows.extend( getWorkflows(url, status=options.wmstatus, user=user, rtype="StoreResults"))
print len(workflows),"in line"
cannot_inject = set()
to_convert = set()
status_cache = defaultdict(str)
## browse for assignment-approved requests, browsed for ours, insert the diff
for wf in workflows:
if specific and not specific in wf: continue
exists = session.query(Workflow).filter(Workflow.name == wf ).first()
if not exists:
wfi = workflowInfo(url, wf)
## check first that there isn't related here with something valid
can_add = True
## first try at finding a match
familly = session.query(Workflow).filter(Workflow.name.contains(wfi.request['PrepID'])).all()
if not familly:
pids = wfi.getPrepIDs()
req_familly = []
for pid in pids:
req_familly.extend( getWorkflowById( url, pid, details=True) )
familly = []
print len(req_familly),"members"
for req_member in req_familly:
#print "member",req_member['RequestName']
owfi = workflowInfo(url, req_member['RequestName'], request=req_member)
other_pids = owfi.getPrepIDs()
if set(pids) == set(other_pids):
## this is a real match
familly.extend( session.query(Workflow).filter(Workflow.name == req_member['RequestName']).all() )
for lwfo in familly:
if lwfo:
## we have it already
if not lwfo.status in ['forget','trouble','forget-unlock','forget-out-unlock']:
wfi.sendLog('injector',"Should not put %s because of %s %s"%( wf, lwfo.name,lwfo.status ))
sendLog('injector',"Should not put %s because of %s %s"%( wf, lwfo.name,lwfo.status ), level='critical')
print "Should not put",wf,"because of",lwfo.name,lwfo.status
cannot_inject.add( wf )
can_add = False
## add a check on validity of input datasets
_,prim,par,sec = wfi.getIO()
for d in list(prim)+list(par)+list(sec):
if not d in status_cache:
status_cache[d] = getDatasetStatus(d)
if status_cache[d] != 'VALID':
wfi.sendLog('injector',"One of the input is not VALID. %s : %s"%( d, status_cache[d]))
sendLog('injector',"One of the input of %s is not VALID. %s : %s"%( wf, d, status_cache[d]), level='critical')
can_add = False
#else:
# ##make sure that all blocks get closed
# closeAllBlocks(url, d)
## check for any file in phedex, to verify existence
_,ph_files,_,_ = getDatasetFiles(url, d)
if not ph_files and not ( 'StoreResults' == wfi.request.setdefault('RequestType',None) ):
wfi.sendLog('injector',"One of the input has no file in phedex: %s" % d )
sendLog('injector',"One of the input has no file in phedex: %s"% d, level='critical')
can_add = False
### ban some workflow that you don't like anymore
#outputs = wfi.request['OutputDatasets']
if not can_add: continue
## temporary hack to transform specific taskchain into stepchains
good_for_stepchain = wfi.isGoodToConvertToStepChain( keywords = transform_keywords)
#good_for_stepchain = wfi.isGoodToConvertToStepChain( keywords = None)
## match keywords and technical constraints
if (not options.no_convert) and good_for_stepchain and not wfi.isRelval():
to_convert.add( wf )
wfi.sendLog('injector','Transforming %s TaskChain into StepChain'%wf)
sendEmail('convertion to stepchain','Transforming %s TaskChain into StepChain'%wf)
wfi.sendLog('injector',"considering %s"%wf)
new_wf = Workflow( name = wf , status = options.setstatus, wm_status = options.wmstatus)
session.add( new_wf )
session.commit()
time.sleep(0.5)
else:
#print "already have",wf
pass
if cannot_inject:
#sendEmail('workflow duplicates','These workflow cannot be added in because of duplicates \n\n %s'%( '\n'.join(cannot_inject)))
sendLog('injector','These workflow cannot be added in because of duplicates \n\n %s'%( '\n'.join(cannot_inject)), level='critical')
for wf in to_convert:
os.system('./Unified/rejector.py --clone --to_step --comments \"Transform to StepChain\" %s'% wf)
## passing a round of invalidation of what needs to be invalidated
if use_mcm and (options.invalidate or True):
invalidator(url)
no_replacement = set()
#print "getting all transfers"
#all_transfers=session.query(Transfer).all()
#print "go!"
## pick up replacements
for wf in session.query(Workflow).filter(Workflow.status == 'trouble').all():
print wf.name
if specific and not specific in wf.name: continue
print wf.name
wfi = workflowInfo(url, wf.name )
wl = wfi.request #getWorkLoad(url, wf.name)
familly = getWorkflowById( url, wl['PrepID'] )
true_familly = []
for member in familly:
if member == wf.name: continue
fwl = getWorkLoad(url , member)
if options.replace:
if member != options.replace: continue
else:
if fwl['RequestDate'] < wl['RequestDate']: continue
if fwl['RequestType']=='Resubmission': continue
if fwl['RequestStatus'] in ['None',None,'new']: continue
if fwl['RequestStatus'] in ['rejected','rejected-archived','aborted','aborted-archived']: continue
true_familly.append( fwl )
if len(true_familly)==0:
#sendLog('injector','%s had no replacement'%wf.name, level='critical')
if wfi.isRelval():
#wfi.sendLog('injector','the workflow was found in trouble with no replacement. As a relval, there is no clean way to handle this.')
wfi.sendLog('injector','the workflow was found in trouble with no replacement. As a relval, there is no clean way to handle this. Setting forget')
wf.status = 'forget'
session.commit()
else:
wfi.sendLog('injector','the workflow was found in trouble with no replacement')
no_replacement.add( wf.name )
continue
else:
wfi.sendLog('injector','the workflow was found in trouble and has a replacement')
print wf.name,"has",len(familly),"familly members"
print wf.name,"has",len(true_familly),"true familly members"
##we cannot have more than one of them !!! pick the last one
if len(true_familly)>1:
#sendEmail('multiple wf','please take a look at injector for %s'%wf.name)
sendLog('injector','Multiple wf in line, will take the last one for %s \n%s'%( wf.name, ', '.join(fwl['RequestName'] for fwl in true_familly)), level='critical')
for fwl in true_familly[-1:]:
member = fwl['RequestName']
new_wf = session.query(Workflow).filter(Workflow.name == member).first()
if not new_wf:
sendLog('injector',"putting %s as replacement of %s"%( member, wf.name))
status = 'away'
if fwl['RequestStatus'] in ['assignment-approved']:
status = 'considered'
new_wf = Workflow( name = member, status = status, wm_status = fwl['RequestStatus'])
wf.status = 'forget'
session.add( new_wf )
else:
if new_wf.status == 'forget': continue
sendLog('injector',"getting %s as replacement of %s"%( new_wf.name, wf.name ))
wf.status = 'forget'
for tr in session.query(TransferImp).filter( TransferImp.workflow_id == wf.id).all():
## get all transfer working for the old workflow
existing = session.query(TransferImp).filter( TransferImp.phedexid == tr.phedexid).filter( TransferImp.workflow_id == new_wf.id).all()
tr.active = False ## disable the old one
if not existing:
## create the transfer object for the new dependency
tri = TransferImp( phedexid = tr.phedexid,
workflow = new_wf)
session.add( tri )
session.commit()
## don't do that automatically
#wf.status = 'forget'
session.commit()
if no_replacement:
#sendEmail('workflow with no replacement','%s \n are dangling there'%( '\n'.join(no_replacement)))
sendLog('injector','workflow with no replacement\n%s \n are dangling there'% ( '\n'.join(no_replacement)), level='critical')
def recoveror(url,specific,options=None):
if userLock('recoveror'): return
up = componentInfo(mcm=False, soft=['mcm'])
if not up.check(): return
CI = campaignInfo()
SI = siteInfo()
UC = unifiedConfiguration()
def make_int_keys( d ):
for code in d:
d[int(code)] = d.pop(code)
error_codes_to_recover = UC.get('error_codes_to_recover')
error_codes_to_block = UC.get('error_codes_to_block')
error_codes_to_notify = UC.get('error_codes_to_notify')
make_int_keys( error_codes_to_recover )
make_int_keys( error_codes_to_block )
make_int_keys( error_codes_to_notify )
#wfs = session.query(Workflow).filter(Workflow.status == 'assistance-recovery').all()
wfs = session.query(Workflow).filter(Workflow.status.contains('recovery')).all()
if specific:
wfs.extend( session.query(Workflow).filter(Workflow.status == 'assistance-manual').all() )
for wfo in wfs:
if specific and not specific in wfo.name:continue
if not specific and 'manual' in wfo.status: continue
wfi = workflowInfo(url, wfo.name)
## need a way to verify that this is the first round of ACDC, since the second round will have to be on the ACDC themselves
all_errors = {}
try:
wfi.getSummary()
all_errors = wfi.summary['errors']
except:
pass
print '-'*100
print "Looking at",wfo.name,"for recovery options"
recover = True
if not 'MergedLFNBase' in wfi.request:
print "f****d up"
sendEmail('missing lfn','%s wl cache is screwed up'%wfo.name)
recover = False
if not len(all_errors):
print "\tno error for",wfo.name
recover = False
task_to_recover = defaultdict(list)
message_to_ops = ""
message_to_user = ""
if 'LheInputFilese' in wfi.request and wfi.request['LheInputFiles']:
## we do not try to recover pLHE
recover = False
if wfi.request['RequestType'] in ['MonteCarlo','ReReco']:
recover = False
if 'Campaign' in wfi.request:
c = wfi.request['Campaign']
if c in CI.campaigns and 'recover' in CI.campaigns[c]:
recover=CI.campaigns[c]['recover']
for task,errors in all_errors.items():
print "\tTask",task
## collect all error codes and #jobs regardless of step at which it occured
all_codes = []
for name, codes in errors.items():
if type(codes)==int: continue
all_codes.extend( [(int(code),info['jobs'],name,list(set([e['type'] for e in info['errors']])),list(set([e['details'] for e in info['errors']])) ) for code,info in codes.items()] )
all_codes.sort(key=lambda i:i[1], reverse=True)
sum_failed = sum([l[1] for l in all_codes])
for errorCode,njobs,name,types,details in all_codes:
rate = 100*njobs/float(sum_failed)
#print ("\t\t %10d (%6s%%) failures with error code %10d (%"+str(max_legend)+"s) at stage %s")%(njobs, "%4.2f"%rate, errorCode, legend, name)
print ("\t\t %10d (%6s%%) failures with error code %10d (%30s) at stage %s")%(njobs, "%4.2f"%rate, errorCode, ','.join(types), name)
added_in_recover=False
#if options.go:
# force the recovery of any task with error ?
if errorCode in error_codes_to_recover:
## the error code is registered
for case in error_codes_to_recover[errorCode]:
match = case['details']
matched= (match==None)
if not matched:
matched=False
for detail in details:
if match in detail:
print "[recover] Could find keyword",match,"in"
print 50*"#"
print detail
print 50*"#"
matched = True
break
if matched and rate > case['rate']:
print "\t\t => we should be able to recover that", case['legend']
task_to_recover[task].append( (code,case) )
added_in_recover=True
message_to_user = ""
else:
print "\t\t recoverable but not frequent enough, needs",case['rate']
if errorCode in error_codes_to_block:
for case in error_codes_to_block[errorCode]:
match = case['details']
matched= (match==None)
if not matched:
matched=False
for detail in details:
if match in detail:
print "[block] Could find keyword",match,"in"
print 50*"#"
print detail
print 50*"#"
matched = True
break
if matched and rate > case['rate']:
print "\t\t => that error means no ACDC on that workflow", case['legend']
if not options.go:
message_to_ops += "%s has an error %s blocking an ACDC.\n%s\n "%( wfo.name, errorCode, '#'*50 )
recover = False
added_in_recover=False
if errorCode in error_codes_to_notify and not added_in_recover:
print "\t\t => we should notify people on this"
message_to_user += "%s has an error %s in processing.\n%s\n" %( wfo.name, errorCode, '#'*50 )
if message_to_user:
print wfo.name,"to be notified to user(DUMMY)",message_to_user
if message_to_ops:
sendEmail( "notification in recoveror" , message_to_ops, destination=['*****@*****.**'])
if len(task_to_recover) != len(all_errors):
print "Should not be doing partial ACDC. skipping"
#sendEmail('recoveror','do not want to make partial acdc on %s'%wfo.name)
sendLog('recoveror','do not want to make partial acdc on %s'%wfo.name, level='warning')
recover = False
if task_to_recover and recover:
print "Initiating recovery"
print ', '.join(task_to_recover.keys()),"to be recovered"
recovering=set()
for task in task_to_recover:
print "Will be making a recovery workflow for",task
## from here you can fetch known solutions, to known error codes
actions = list(set([case['solution'] for code,case in task_to_recover[task] ]))
acdc = singleRecovery(url, task, wfi.request , actions, do = options.do)
if not acdc:
if options.do:
if recovering:
print wfo.name,"has been partially ACDCed. Needs manual attention"
#sendEmail( "failed ACDC partial recovery","%s has had %s/%s recoveries %s only"%( wfo.name, len(recovering), len(task_to_recover), list(recovering)), destination=['*****@*****.**'])
sendLog('recoveror', "%s has had %s/%s recoveries %s only"%( wfo.name, len(recovering), len(task_to_recover), list(recovering)), level='critical')
continue
else:
print wfo.name,"failed recovery once"
#break
continue
else:
print "no action to take further"
#sendEmail("an ACDC that can be done automatically","please check https://cmst2.web.cern.ch/cmst2/unified/logs/recoveror/last.log for details", destination=['*****@*****.**'])
sendLog('recoveror', "ACDC for %s can be done automatically"% wfo.name, level='critical')
continue
## and assign it ?
team = wfi.request['Teams'][0]
parameters={
#'SiteWhitelist' : wfi.request['SiteWhitelist'],
'SiteWhitelist' : SI.sites_ready,
'AcquisitionEra' : wfi.acquisitionEra(),
'ProcessingString' : wfi.processingString(),
'MergedLFNBase' : wfi.request['MergedLFNBase'],
'ProcessingVersion' : wfi.request['ProcessingVersion'],
}
## hackery for ACDC merge assignment
if wfi.request['RequestType'] == 'TaskChain' and 'Merge' in task.split('/')[-1]:
parameters['AcquisitionEra'] = None
parameters['ProcessingString'] = None
if options.ass:
print "really doing the assignment of the ACDC",acdc
parameters['execute']=True
#sendEmail("an ACDC was done and WAS assigned", "%s was assigned, please check https://cmst2.web.cern.ch/cmst2/unified/logs/recoveror/last.log for details"%( acdc ), destination=['*****@*****.**'])
wfi.sendLog('recoveror',"%s was assigned for recovery"% acdc)
else:
print "no assignment done with this ACDC",acdc
#sendEmail("an ACDC was done and need to be assigned", "%s needs to be assigned, please check https://cmst2.web.cern.ch/cmst2/unified/logs/recoveror/last.log for details"%( acdc ), destination=['*****@*****.**'])
sendLog('recoveror',"%s needs to be assigned"%(acdc), level='critical')
result = reqMgrClient.assignWorkflow(url, acdc, team, parameters)
if not result:
print acdc,"was not asigned"
#sendEmail("an ACDC was done and need to be assigned","%s needs to be assigned, please check https://cmst2.web.cern.ch/cmst2/unified/logs/recoveror/last.log for details"%( acdc ), destination=['*****@*****.**'])
sendLog('recoveror',"%s needs to be assigned"%(acdc), level='critical')
else:
recovering.add( acdc )
current = None
if recovering:
#if all went well, set the status to -recovering
current = wfo.status
if options.ass:
current = current.replace('recovery','recovering')
else:
current = 'assistance-manual'
print 'created ACDC: '+', '.join( recovering )
else:
## was set to be recovered, and no acdc was made
current = 'assistance-manual'
if current:
print wfo.name,"setting the status to",current
wfo.status = current
session.commit()
else:
## this workflow should be handled manually at that point
print wfo.name,"needs manual intervention"
wfo.status = 'assistance-manual'
session.commit()
def transferor(url, specific=None, talk=True, options=None):
if userLock(): return
mlock = moduleLock()
if mlock(): return
use_mcm = True
up = componentInfo(soft=['mcm', 'wtc', 'jira'])
if not up.check(): return
use_mcm = up.status['mcm']
if options and options.test:
execute = False
else:
execute = True
SI = siteInfo()
CI = campaignInfo()
#NLI = newLockInfo()
#if not NLI.free(): return
LI = lockInfo()
if not LI.free(): return
mcm = McMClient(dev=False)
dss = DSS()
#allowed_secondary = UC.get('')
print "counting all being handled..."
being_handled = len(
session.query(Workflow).filter(Workflow.status == 'away').all())
being_handled += len(
session.query(Workflow).filter(
Workflow.status.startswith('stag')).all())
being_transfered = len(
session.query(Workflow).filter(Workflow.status == 'staging').all())
#being_handled += len(session.query(Workflow).filter(Workflow.status.startswith('assistance-')).all())
being_handled += len(
session.query(Workflow).filter(
Workflow.status.startswith('assistance-')).filter(
~Workflow.status.contains('custodial')).all())
max_to_handle = options.maxworkflows
max_to_transfer = options.maxstaging
allowed_to_handle = max(0, max_to_handle - being_handled)
allowed_to_transfer = max(0, max_to_transfer - being_transfered)
wf_buffer = 5
if allowed_to_handle <= wf_buffer: ## buffer for having several wf per transfer
print "Not allowed to run more than", max_to_handle, "at a time. Currently", being_handled, "and", wf_buffer, "buffer"
else:
print being_handled, "already being handled", max_to_handle, "max allowed,", allowed_to_handle, "remaining", "and", wf_buffer, "buffer"
if allowed_to_transfer <= wf_buffer:
print "Not allowed to transfer more than", max_to_transfer, "at a time. Currently", being_transfered, "and", wf_buffer, "buffer"
else:
print being_transfered, "already being transfered", max_to_transfer, "max allowed,", allowed_to_transfer, "remaining", "and", wf_buffer, "buffer"
print "... done"
all_transfers = defaultdict(list)
workflow_dependencies = defaultdict(
set) ## list of wf.id per input dataset
wfs_and_wfh = []
max_per_round = UC.get('max_per_round').get('transferor', None)
print "getting all wf to consider ..."
cache = getWorkflows(url, 'assignment-approved', details=True)
all_to_include = session.query(Workflow).filter(
Workflow.status.startswith('considered')).all()
if len(cache) > 2000:
max_to_include = max_per_round
random.shuffle(cache) ## randomize first by wf name
cache = sorted(cache, key=lambda r: r['RequestPriority'],
reverse=True) ## order by prio
highest = [r['RequestName'] for r in cache[:max_to_include]]
all_to_include = [wfo for wfo in all_to_include if wfo.name in highest]
print "limiting what to consider to", max_to_include, "because there is too much stuff going on. Got", len(
all_to_include)
for wfo in all_to_include:
print "\t", wfo.name
if specific and not specific in wfo.name: continue
cache_r = filter(lambda d: d['RequestName'] == wfo.name, cache)
if len(cache_r):
wfs_and_wfh.append((wfo,
workflowInfo(url,
wfo.name,
spec=False,
request=cache_r[0])))
else:
wfs_and_wfh.append((wfo, workflowInfo(url, wfo.name, spec=False)))
print "... done"
transfers_per_sites = defaultdict(int)
input_sizes = defaultdict(float)
ignored_input_sizes = defaultdict(float)
input_cput = {}
input_st = {}
## list the size of those in transfer already
in_transfer_priority = None
min_transfer_priority = None
print "getting all wf in staging ..."
#stucks = json.loads(open('%s/stuck_transfers.json'%monitor_pub_dir).read())
stucks = json.loads(eosRead('%s/stuck_transfers.json' % monitor_pub_dir))
for wfo in session.query(Workflow).filter(
Workflow.status == 'staging').all():
wfh = workflowInfo(url, wfo.name, spec=False)
#(lheinput,primary,parent,secondary) = wfh.getIO()
#sites_allowed = getSiteWhiteList( (lheinput,primary,parent,secondary) )
print wfo.name, "staging"
(lheinput, primary, parent, secondary,
sites_allowed) = wfh.getSiteWhiteList()
for site in sites_allowed: ## we should get the actual transfer destination instead of the full white list
transfers_per_sites[site] += 1
#input_cput[wfo.name] = wfh.getComputingTime()
#input_st[wfo.name] = wfh.getSystemTime()
blocks = wfh.getBlocks()
for prim in primary:
ds_s = dss.get(prim, blocks=blocks)
if prim in stucks:
wfh.sendLog(
'transferor',
"%s appears stuck, so not counting it %s [GB]" %
(prim, ds_s))
ignored_input_sizes[prim] = max(ds_s,
ignored_input_sizes[prim])
else:
input_sizes[prim] = max(ds_s, input_sizes[prim])
wfh.sendLog('transferor',
"%s needs %s [GB]" % (wfo.name, ds_s))
if in_transfer_priority == None:
in_transfer_priority = int(wfh.request['RequestPriority'])
else:
in_transfer_priority = max(in_transfer_priority,
int(wfh.request['RequestPriority']))
if min_transfer_priority == None:
min_transfer_priority = int(wfh.request['RequestPriority'])
else:
min_transfer_priority = min(min_transfer_priority,
int(wfh.request['RequestPriority']))
try:
print "Ignored input sizes"
ignored_values = list(ignored_input_sizes.items())
ignored_values.sort(key=lambda i: i[1])
print "\n".join(map(str, ignored_values))
print "Considered input sizes"
considered_values = list(input_sizes.items())
considered_values.sort(key=lambda i: i[1])
print "\n".join(map(str, considered_values))
except Exception as e:
print "trying to print the summary of input size"
print str(e)
print "... done"
print "Max priority in transfer already", in_transfer_priority
print "Min priority in transfer already", min_transfer_priority
print "transfers per sites"
print json.dumps(transfers_per_sites, indent=2)
in_transfer_already = sum(input_sizes.values())
cput_in_transfer_already = sum(input_cput.values())
st_in_transfer_already = sum(input_st.values())
## list the size of all inputs
primary_input_per_workflow_gb = defaultdict(float)
print "getting all input sizes ..."
input_blocks = {}
for (wfo, wfh) in wfs_and_wfh:
(_, primary, _, _) = wfh.getIO()
blocks = wfh.getBlocks()
input_blocks[wfo.name] = blocks
for prim in primary:
## do not count it if it appears stalled !
prim_size = dss.get(prim, blocks=blocks)
input_sizes[prim] = max(prim_size, input_sizes[prim])
primary_input_per_workflow_gb[wfo.name] += prim_size
print "... done"
# shuffle first by name
random.shuffle(wfs_and_wfh)
# Sort smallest transfers first; allows us to transfer as many as possible workflows.
def prio_and_size(i, j):
if int(i[1].request['RequestPriority']) == int(
j[1].request['RequestPriority']):
return cmp(int(primary_input_per_workflow_gb.get(j[0].name, 0)),
int(primary_input_per_workflow_gb.get(i[0].name, 0)))
else:
return cmp(int(i[1].request['RequestPriority']),
int(j[1].request['RequestPriority']))
#wfs_and_wfh.sort(cmp = prio_and_size, reverse=True)
#wfs_and_wfh.sort(cmp = lambda i,j : cmp(int(primary_input_per_workflow_gb.get(i[0].name, 0)), int(primary_input_per_workflow_gb.get(j[0].name, 0)) ))
#sort by priority higher first
wfs_and_wfh.sort(cmp=lambda i, j: cmp(int(i[1].request[
'RequestPriority']), int(j[1].request['RequestPriority'])),
reverse=True)
if min_transfer_priority == None or in_transfer_priority == None:
print "nothing is lining up for transfer"
sendLog(
"transferor",
"No request in staging, using first request to set priority limit")
if len(wfs_and_wfh):
min_transfer_priority = wfs_and_wfh[0][1].request[
'RequestPriority']
in_transfer_priority = wfs_and_wfh[0][1].request['RequestPriority']
else:
return
cput_grand_total = sum(input_cput.values())
cput_to_transfer = cput_grand_total - cput_in_transfer_already
st_grand_total = sum(input_st.values())
st_to_transfer = st_grand_total - st_in_transfer_already
print "%15.4f [CPU h] worth already in transfer" % cput_in_transfer_already
print "%15.4f [CPU h] worth is the current requested transfer load" % cput_to_transfer
print "%15.4f [h] worth of absolute system time in transfer" % (
cput_in_transfer_already / SI.availableSlots())
print "%15.4f [h] worth of absolute system time is the current requested transfer load" % (
cput_to_transfer / SI.availableSlots())
print "%15.4f [h] worth of theoritical system time in transfer" % (
st_in_transfer_already)
print "%15.4f [h] worth of theoritical system time is the current requested transfer load" % (
st_to_transfer)
grand_total = sum(input_sizes.values())
to_transfer = grand_total - in_transfer_already
grand_transfer_limit = options.maxtransfer
#grand_transfer_limit = SI.total_disk()*0.25*1024## half of the free sapce in TB->GB
transfer_limit = grand_transfer_limit - in_transfer_already
print "%15.4f GB already being transfered" % in_transfer_already
print "%15.4f GB is the current requested transfer load" % to_transfer
print "%15.4f GB is the global transfer limit" % grand_transfer_limit
print "%15.4f GB is the available limit" % transfer_limit
max_staging_per_site = options.maxstagingpersite
# the max priority value per dataset.
max_priority = defaultdict(int)
needs_transfer = 0 ## so that we can count'em
passing_along = 0
transfer_sizes = defaultdict(float)
went_over_budget = False
destination_cache = {}
no_goes = set()
if max_per_round and not spec:
wfs_and_wfh = wfs_and_wfh[:max_per_round]
for (wfo, wfh) in wfs_and_wfh:
print wfo.name, "to be transfered with priority", wfh.request[
'RequestPriority']
if wfh.request['RequestStatus'] != 'assignment-approved':
if wfh.request['RequestStatus'] in [
'aborted', 'rejected', 'rejected-archived',
'aborted-archived'
]:
if wfh.isRelval():
wfo.status = 'forget'
else:
wfo.status = 'trouble' ## so that we look or a replacement
else:
wfo.status = 'away'
wfh.sendLog(
'transferor', '%s in status %s, setting %s' %
(wfo.name, wfh.request['RequestStatus'], wfo.status))
continue
(lheinput, primary, parent, secondary,
sites_allowed) = wfh.getSiteWhiteList()
blocks = input_blocks.get(wfo.name, wfh.getBlocks())
if blocks:
print "Reading only", len(blocks), "blocks in input"
this_load = sum([dss.get(prim, blocks=blocks) for prim in primary])
no_budget = False
if (this_load
and (sum(transfer_sizes.values()) + this_load > transfer_limit
or went_over_budget)):
if went_over_budget:
wfh.sendLog('transferor', "Transfer has gone over bubget.")
else:
wfh.sendLog('transferor', "Transfer will go over bubget.")
wfh.sendLog(
'transferor',
"%15.4f GB this load, %15.4f GB already this round, %15.4f GB is the available limit"
% (this_load, sum(transfer_sizes.values()), transfer_limit))
#if sum(transfer_sizes.values()) > transfer_limit:
went_over_budget = True
if in_transfer_priority != None and min_transfer_priority != None:
if int(
wfh.request['RequestPriority']
) >= in_transfer_priority and min_transfer_priority != in_transfer_priority:
wfh.sendLog(
'transferor',
"Higher priority sample %s >= %s go-on over budget" %
(wfh.request['RequestPriority'], in_transfer_priority))
else:
if not options.go:
wfh.sendLog(
'transferor',
"%s minimum priority %s < %s : stop" %
(min_transfer_priority,
wfh.request['RequestPriority'],
in_transfer_priority))
no_budget = True
## throtlle by campaign go
no_go = False
if not wfh.go(log=True) and not options.go:
no_go = True
no_goes.add(wfo.name)
allowed_secondary = {}
overide_parameters = {}
check_secondary = (not wfh.isRelval())
output_tiers = list(
set([o.split('/')[-1] for o in wfh.request['OutputDatasets']]))
for campaign in wfh.getCampaigns():
if campaign in CI.campaigns:
overide_parameters.update(CI.campaigns[campaign])
if campaign in CI.campaigns and 'secondaries' in CI.campaigns[
campaign]:
if CI.campaigns[campaign]['secondaries']:
allowed_secondary.update(
CI.campaigns[campaign]['secondaries'])
check_secondary = True
if campaign in CI.campaigns and 'banned_tier' in CI.campaigns[
campaign]:
banned_tier = list(
set(CI.campaigns[campaign]['banned_tier'])
& set(output_tiers))
if banned_tier:
no_go = True
wfh.sendLog(
'transferor', 'These data tiers %s are not allowed' %
(','.join(banned_tier)))
sendLog('transferor',
'These data tiers %s are not allowed in %s' %
(','.join(banned_tier), wfo.name),
level='critical')
if secondary and check_secondary:
if (set(secondary) & set(allowed_secondary.keys()) !=
set(secondary)):
msg = '%s is not an allowed secondary' % (
', '.join(set(secondary) - set(allowed_secondary.keys())))
wfh.sendLog('transferor', msg)
critical_msg = msg + '\nWorkflow URL: https://dmytro.web.cern.ch/dmytro/cmsprodmon/workflows.php?prep_id=task_{}'.format(
wfh.getPrepIDs()[0])
sendLog('transferor', critical_msg, level='critical')
if not options.go:
no_go = True
for sec in secondary:
if sec in allowed_secondary:
overide_parameters.update(allowed_secondary[sec])
if 'SiteWhitelist' in overide_parameters:
sites_allowed = list(
set(sites_allowed) & set(overide_parameters['SiteWhitelist']))
wfh.sendLog(
'transferor',
'Intersecting with the overriding whitelist parameters, allowed sites become {}'
.format(sites_allowed))
if no_go:
continue
if passing_along >= allowed_to_handle:
#if int(wfh.request['RequestPriority']) >= in_transfer_priority and min_transfer_priority!=in_transfer_priority:
if in_transfer_priority != None and min_transfer_priority != None:
if int(wfh.request['RequestPriority']
) >= in_transfer_priority and int(
wfh.request['RequestPriority']
) != min_transfer_priority:
## higher priority, and not only this priority being transfered
wfh.sendLog(
'transferor',
"Higher priority sample %s >= %s go-on over %s" %
(wfh.request['RequestPriority'], in_transfer_priority,
max_to_handle))
else:
wfh.sendLog(
'transferor',
" Not allowed to pass more than %s at a time. Currently %s handled, and adding %s"
% (max_to_handle, being_handled, passing_along))
if not options.go:
## should not allow to jump that fence
break
if this_load and needs_transfer >= allowed_to_transfer:
if in_transfer_priority != None and min_transfer_priority != None:
if int(wfh.request['RequestPriority']
) >= in_transfer_priority and int(
wfh.request['RequestPriority']
) != min_transfer_priority:
## higher priority, and not only this priority being transfered
wfh.sendLog(
'transferor',
"Higher priority sample %s >= %s go-on over %s" %
(wfh.request['RequestPriority'], in_transfer_priority,
max_to_transfer))
else:
wfh.sendLog(
'transferor',
"Not allowed to transfer more than %s at a time. Currently %s transfering, and adding %s"
% (max_to_transfer, being_transfered, needs_transfer))
if not options.go:
no_budget = True
if no_budget:
continue
# break ## try this for a while to make things faster
## the site white list considers site, campaign, memory and core information
if options and options.tosites:
sites_allowed = options.tosites.split(',')
for dataset in list(primary) + list(parent) + list(secondary):
LI.lock(dataset, reason='staging')
if not sites_allowed:
wfh.sendLog('transferor', "not possible site to run at")
#sendEmail("no possible sites","%s has no possible sites to run at"%( wfo.name ))
sendLog('transferor',
"%s has no possible sites to run at" % (wfo.name),
level='critical')
continue
can_go = True
staging = False
allowed = True
primary_destinations = set()
if primary:
copies_needed_from_CPUh, CPUh = wfh.getNCopies()
if talk:
print wfo.name, 'reads', ', '.join(primary), 'in primary'
## chope the primary dataset
for prim in primary:
## keep track of what needs what
workflow_dependencies[prim].add(wfo.id)
max_priority[prim] = max(max_priority[prim],
int(wfh.request['RequestPriority']))
wfh.sendLog(
'transferor', "Would make %s from cpu requirement %s" %
(copies_needed_from_CPUh, CPUh))
copies_needed = copies_needed_from_CPUh
if 'Campaign' in wfh.request and wfh.request[
'Campaign'] in CI.campaigns and 'maxcopies' in CI.campaigns[
wfh.request['Campaign']]:
copies_needed_from_campaign = CI.campaigns[
wfh.request['Campaign']]['maxcopies']
copies_needed = min(copies_needed_from_campaign,
copies_needed)
wfh.sendLog(
'transferor',
"Maxed to %s by campaign configuration %s" %
(copies_needed, wfh.request['Campaign']))
if blocks:
print "limiting to blocks", "\n".join(sorted(blocks))
### new ways of making the whole thing
destinations, all_block_names = getDatasetDestinations(
url,
prim,
within_sites=[SI.CE_to_SE(site) for site in sites_allowed],
only_blocks=blocks)
print json.dumps(destinations, indent=2)
## get where the dataset is in full and completed
prim_location = [
site for (site, info) in destinations.items()
if info['completion'] == 100 and info['data_fraction'] == 1
]
## the rest is places it is going to be
#prim_destination = [site for site in destinations.keys() if not site in prim_location]
prim_destination = [
site for (site, info) in destinations.items()
if info['data_fraction'] == 1 and info['completion'] != 100
]
## veto the site with no current disk space, for things that are not relval
prim_destination = [
site for site in prim_destination
if (SI.disk[site] or wfh.isRelval())
]
if len(prim_location) >= copies_needed:
wfh.sendLog(
'transferor',
"The input is all fully in place at %s sites %s" %
(len(prim_location), sorted(prim_location)))
continue
copies_needed = max(0, copies_needed - len(prim_location))
wfh.sendLog(
'transferor',
"Counting existing copies ; now need %s" % copies_needed)
copies_being_made = [
sum([
info['blocks'].keys().count(block)
for site, info in destinations.items()
if site in prim_destination
]) for block in all_block_names
]
latching_on_transfers = set()
[
latching_on_transfers.update(info['blocks'].values())
for site, info in destinations.items()
if site in prim_destination
]
latching_on_transfers = list(latching_on_transfers)
#print latching_on_transfers
## figure out where all this is going to go
prim_to_distribute = [
site for site in sites_allowed
if not SI.CE_to_SE(site) in prim_location
]
prim_to_distribute = [
site for site in prim_to_distribute
if not SI.CE_to_SE(site) in prim_destination
]
## take out the ones that cannot receive transfers
potential_destinations = len(prim_to_distribute)
#prim_to_distribute = [site for site in prim_to_distribute if not SI.CE_to_SE(site) in SI.sites_veto_transfer]
prim_to_distribute = [
site for site in prim_to_distribute
if (SI.disk[SI.CE_to_SE(site)] or wfh.isRelval())
]
## do we want to restrict transfers if the amount of site in vetoe are too large ?
wfh.sendLog(
'transferor',
"Could be going to: %s" % sorted(prim_to_distribute))
if not prim_to_distribute or any([
transfers_per_sites[site] < max_staging_per_site
for site in prim_to_distribute
]):
## means there is openings let me go
print "There are transfer slots available:", [
(site, transfers_per_sites[site])
for site in prim_to_distribute
]
else:
if int(
wfh.request['RequestPriority']
) >= in_transfer_priority and min_transfer_priority != in_transfer_priority:
wfh.sendLog(
'transferor',
"Higher priority sample %s >= %s go-on over transfer slots available"
% (wfh.request['RequestPriority'],
in_transfer_priority))
else:
wfh.sendLog(
'transferor',
"Not allowed to transfer more than %s per site at a time. Going overboard for %s"
% (max_staging_per_site,
sorted([
site for site in prim_to_distribute
if transfers_per_sites[site] >=
max_staging_per_site
])))
if not options.go:
allowed = False
break
for latching in latching_on_transfers:
existings = session.query(TransferImp).filter(
TransferImp.phedexid == int(latching)).filter(
TransferImp.workflow_id == wfo.id).all()
if not existings:
tri = TransferImp(phedexid=int(latching), workflow=wfo)
print "adding", wfo.id, "with phedexid", latching
session.add(tri)
else:
for existing in existings:
existing.active = True
session.flush()
can_go = False
transfer_sizes[prim] = max(this_load, transfer_sizes[prim])
staging = True
# reduce the number of copies required by the on-going full transfer : how do we bootstrap on waiting for them ??
#copies_needed = max(0,copies_needed - len(prim_destination))
copies_needed = max(0, copies_needed - min(copies_being_made))
wfh.sendLog(
'transferor',
"Counting the copies being made ; then need %s" %
copies_needed)
if copies_needed == 0:
wfh.sendLog(
'transferor',
"The input is either fully in place or getting in full somewhere with %s"
% latching_on_transfers)
can_go = True
continue
elif len(prim_to_distribute) == 0:
wfh.sendLog(
'transferor',
"We are going to need extra copies of %s, but no destinations seems available"
% (prim))
sendLog(
'transferor',
"We are going to need extra copies of %s, but no destinations seems available"
% (prim),
level='critical')
print json.dumps(prim_to_distribute, indent=2)
print json.dumps(prim_location, indent=2)
print json.dumps(prim_destination, indent=2)
prim_to_distribute = [
site for site in sites_allowed
if not SI.CE_to_SE(site) in prim_location
]
#prim_to_distribute = [site for site in prim_to_distribute if not SI.CE_to_SE(site) in SI.sites_veto_transfer ]
prim_to_distribute = [
site for site in prim_to_distribute
if (SI.disk[SI.CE_to_SE(site)] or wfh.isRelval())
]
print "changed to"
print json.dumps(prim_to_distribute, indent=2)
if len(
prim_to_distribute
) > 0: ## maybe that a parameter we can play with to limit the
if not options or options.chop:
### hard include the tape disk andpoint ?
#tapes = [site for site in getDatasetPresence( url, prim, vetos=['T0','T2','T3','Disk']) if site.endswith('MSS')]
chops, sizes = getDatasetChops(
prim,
chop_threshold=options.chopsize,
only_blocks=blocks)
spreading = distributeToSites(chops,
prim_to_distribute,
n_copies=copies_needed,
weights=SI.cpu_pledges,
sizes=sizes)
## prune the blocks/destination that are already in the making, so that subscription don't overlap
for site in spreading:
for block in list(spreading[site]):
if site in destinations and block in destinations[
site]['blocks'].keys():
## prune it
spreading[site].remove(block)
transfer_sizes[prim] = sum(sizes)
if not spreading:
sendLog(
'transferor',
'cannot send %s to any site, it cannot fit anywhere'
% prim,
level='critical')
wfh.sendLog(
'transferor',
"cannot send to any site. %s cannot seem to fit anywhere"
% (prim))
staging = False
can_go = False
else:
spreading = {}
for site in prim_to_distribute:
if blocks:
spreading[site] = blocks
else:
spreading[site] = [prim]
transfer_sizes[prim] = max(this_load,
transfer_sizes[prim])
can_go = False
wfh.sendLog(
'transferor', "selected CE destinations %s" %
(sorted(spreading.keys())))
for (site, items) in spreading.items():
all_transfers[site].extend(items)
transfers_per_sites[site] += 1
primary_destinations.add(site)
else:
can_go = False
allowed = False
if not allowed:
wfh.sendLog('transferor', "Not allowed to move on with")
continue
if secondary:
override_sec_destination = []
if 'SecondaryLocation' in CI.campaigns[wfh.request['Campaign']]:
override_sec_destination = CI.campaigns[
wfh.request['Campaign']]['SecondaryLocation']
if 'SecondaryLocation' in overide_parameters:
override_sec_destination = overide_parameters[
'SecondaryLocation']
print wfo.name, 'reads', ', '.join(secondary), 'in secondary'
for sec in secondary:
workflow_dependencies[sec].add(wfo.id)
if True:
## new style, failing on minbias
if not sec in destination_cache:
## this is barbbaric, and does not show the correct picture on workflow by workflow with different whitelist
destination_cache[sec], _ = getDatasetDestinations(
url, sec) ## NO SITE WHITE LIST ADDED
#destination_cache[sec],_ = getDatasetDestinations(url, sec, within_sites = [SI.CE_to_SE(site) for site in sites_allowed])
## limit to the site whitelist NOW
se_allowed = set(
[SI.CE_to_SE(site) for site in sites_allowed])
destinations = dict([
(k, v) for (k, v) in destination_cache[sec].items()
if k in se_allowed
])
## truncate location/destination to those making up for >90% of the dataset
bad_destinations = [
destinations.pop(site)
for (site, info) in destinations.items()
if info['data_fraction'] < 0.9
]
print sec, json.dumps(destinations, indent=2)
sec_location = [
site for (site, info) in destinations.items()
if info['completion'] >= 95
]
sec_destination = [
site for site in destinations.keys()
if not site in sec_location
] ## this is in SE
else:
## old style
presence = getDatasetPresence(url, sec)
sec_location = [
site for site, pres in presence.items()
if pres[1] > 90.
] ## more than 90% of the minbias at sites
subscriptions = listSubscriptions(url, sec)
sec_destination = [site for site in subscriptions]
## how to make unified understand that it has to wait for the secondary if the sec_destination and
#sec_to_distribute = [site for site in sites_allowed if not any([osite.startswith(site) for osite in sec_location])]
sec_to_distribute = [
site for site in sites_allowed
if not SI.CE_to_SE(site) in sec_location
]
#sec_to_distribute = [site for site in sec_to_distribute if not any([osite.startswith(site) for osite in sec_destination])]
sec_to_distribute = [
site for site in sec_to_distribute
if not SI.CE_to_SE(site) in sec_destination
]
presitespace_sec_to_distribute = copy.deepcopy(
sec_to_distribute)
#sec_to_distribute = [site for site in sec_to_distribute if not any([osite.startswith(site) for osite in SI.sites_veto_transfer])]
#sec_to_distribute = [site for site in sec_to_distribute if not SI.CE_to_SE(site) in SI.sites_veto_transfer]
sec_to_distribute = [
site for site in sec_to_distribute
if (SI.disk[SI.CE_to_SE(site)] or wfh.isRelval())
]
## at this point you have a problem
if len(sec_to_distribute) == 0 and len(
presitespace_sec_to_distribute):
sendLog(
'transferor',
'%s is getting no possible destinations because of lack of space. To be decided what to do in general'
% (sec),
level='critical')
if override_sec_destination:
## intersect with where we want the PU to be
not_needed_anymore = list(
set(sec_to_distribute) - set(override_sec_destination))
#sendEmail("secondary superfluous","the dataset %s could be removed from %s"%( sec, not_needed_anymore ))
sendLog(
'transferor',
"the dataset %s could be removed from %s" %
(sec, not_needed_anymore))
sec_to_distribute = list(
set(sec_to_distribute) & set(override_sec_destination))
if len(sec_to_distribute) > 0:
print "secondary could go to", sorted(sec_to_distribute)
sec_size = dss.get(sec)
for site in sec_to_distribute:
site_se = SI.CE_to_SE(site)
if (SI.disk[site_se] *
1024.) > sec_size or wfh.isRelval():
wfh.sendLog('transferor',
'Sending %s to %s' % (sec, site))
all_transfers[site].append(sec)
can_go = False
else:
print "could not send the secondary input to", site_se, "because it is too big for the available disk", SI.disk[
site_se] * 1024, "GB need", sec_size
if primary_destinations and site in primary_destinations:
#sendEmail('secondary input too big','%s is too big (%s) for %s (%s)'%( sec, sec_size, site_se, SI.disk[site_se]*1024))
sendLog(
'transferor',
'%s is too big (%s) for %s (%s). %s will not be able to run there.'
% (sec, sec_size, site_se,
SI.disk[site_se] * 1024, wfo.name),
level='critical')
wfh.sendLog(
'transferor',
'%s is too big (%s) for %s (%s). will not be able to run there.'
% (sec, sec_size, site_se,
SI.disk[site_se] * 1024))
else:
## this is bas overall
print "the secondary input does not have to be send to site"
## is that possible to do something more
if can_go:
## no explicit transfer required this time
if staging:
## but using existing ones
wfh.sendLog(
'transferor',
"latches on existing transfers, and nothing else, settin staging"
)
wfo.status = 'staging'
needs_transfer += 1
else:
wfh.sendLog(
'transferor', "should just be assigned now to %s" %
sorted(sites_allowed))
wfo.status = 'staged'
passing_along += 1
wfh.sendLog('transferor',
"setting %s status to %s" % (wfo.name, wfo.status))
#session.commit()
continue
else:
## there is an explicit transfer required
if staging:
## and also using an existing one
wfh.sendLog('transferor', "latches on existing transfers")
if not options.test:
wfo.status = 'staging'
wfh.sendLog(
'transferor',
"setting %s status to %s" % (wfo.name, wfo.status))
#session.commit()
wfh.sendLog('transferor', "needs a transfer")
needs_transfer += 1
passing_along += 1
if no_goes:
#sendEmail("no go for managing","No go for \n"+"\n".join( no_goes ))
sendLog('transferor',
"No go for \n" + "\n".join(sorted(no_goes)),
level='critical')
print "accumulated transfers"
print json.dumps(all_transfers, indent=2)
fake_id = -1
wf_id_in_prestaging = set()
for (site, items_to_transfer) in all_transfers.iteritems():
items_to_transfer = list(set(items_to_transfer))
## convert to storage element
site_se = SI.CE_to_SE(site)
## site that do not want input datasets
#if site in SI.sites_veto_transfer:
# print site,"does not want transfers"
# continue
## throttle the transfer size to T2s ? we'd be screwed by a noPU sample properly configured.
## massage a bit the items
blocks = [it for it in items_to_transfer if '#' in it]
block_datasets = list(set([it.split('#')[0] for it in blocks]))
datasets = [it for it in items_to_transfer if not '#' in it]
details_text = "Making a replica to %s (CE) %s (SE) for" % (site,
site_se)
#print "\t",len(blocks),"blocks"
## remove blocks if full dataset is send out
blocks = [
block for block in blocks if not block.split('#')[0] in datasets
]
#print "\t",len(blocks),"needed blocks for",list(set([block.split('#')[0] for block in blocks]))
#print "\t",len(datasets),"datasets"
#print "\t",datasets
details_text += '\n\t%d blocks' % len(blocks)
details_text += '\n\t%d needed blocks for %s' % (
len(blocks),
sorted(list(set([block.split('#')[0] for block in blocks]))))
details_text += '\n\t%d datasets' % len(datasets)
details_text += '\n\t%s' % sorted(datasets)
items_to_transfer = blocks + datasets
if execute:
sendLog('transferor', details_text)
else:
print "Would make a replica to", site, "(CE)", site_se, "(SE) for"
print details_text
## operate the transfer
if options and options.stop:
## ask to move-on
answer = raw_input('Continue with that ?')
if not answer.lower() in ['y', 'yes', 'go']:
continue
transfered_items = defaultdict(set)
if execute:
priority = 'normal'
cds = [
ds for ds in set(datasets + block_datasets)
if ds in max_priority
]
## bucketize the transfers by priority of workflows
prioritized_items = defaultdict(set)
for item in items_to_transfer:
d = item.split('#')[0]
p = max_priority.get(d, 80000)
q = 'normal'
if p > 100000:
q = 'reserved'
elif p < 70000:
q = 'low'
prioritized_items[q].add(item)
for priority, items in prioritized_items.items():
result = makeReplicaRequest(url,
site_se,
list(items),
'prestaging',
priority=priority,
approve=True)
if result:
these_transfers = [
o['id'] for o in result['phedex']['request_created']
]
#phedexids.extend( these_transfers )
for ph in these_transfers:
transfered_items[ph].update(items)
else:
sendLog(
'transferor',
'Could not make a replica request for items %s to site %s'
% (items, site_se),
level='critical')
#result = makeReplicaRequest(url, site_se, items_to_transfer, 'prestaging', priority=priority, approve=True)
#phedexids = [o['id'] for o in result['phedex']['request_created']]:
#else:
# #result= {'phedex':{'request_created' : []}}
# phedexids = []
# fake_id-=1
if not transfered_items:
sendLog(
'transferor',
'Could not make a replica request for items %s to site %s' %
(items_to_transfer, site),
level='critical')
continue
for phedexid, items in transfered_items.items():
print phedexid, "transfer created"
for transfering in list(
set(map(lambda it: it.split('#')[0], items))):
for wfid in workflow_dependencies[transfering]:
new_transfer = session.query(TransferImp).filter(
TransferImp.phedexid == int(phedexid)).filter(
TransferImp.workflow_id == wfid).first()
if not new_transfer:
new_transfer = TransferImp(
phedexid=phedexid,
workflow=session.query(Workflow).get(wfid))
session.add(new_transfer)
else:
new_transfer.active = True
wf_id_in_prestaging.add(wfid)
#session.commit()
for wfid in wf_id_in_prestaging:
tr_wf = session.query(Workflow).get(wfid)
if tr_wf and tr_wf.status != 'staging':
if execute:
tr_wf.status = 'staging'
if talk:
print "setting", tr_wf.name, "to staging"
#session.commit()
## one big session commit at the end that everything went fine
session.commit()
#session.commit()
wfs = getWorkflows(url, 'assigned', details=True)
now = time.mktime(time.gmtime())
for wf in wfs:
assigned_log = filter(lambda change: change["Status"] in ["assigned"],
wf['RequestTransition'])
if assigned_log:
then = assigned_log[-1]['UpdateTime']
since = (now - then) / float(1 * 24 * 60 * 60.)
if since > 1.:
print "workflow", wf[
'RequestName'], "is assigned since", then, " that is", since, "days"
sendLog('GQ',
'The workflow %s has been assigned for %.2f days' %
(wf['RequestName'], since),
level='critical')
overall_timeout = 14 #days
may_have_one = set()
may_have_one.update([
wfo.name for wfo in session.query(Workflow).filter(
Workflow.status.startswith('away')).all()
])
may_have_one.update([
wfo.name for wfo in session.query(Workflow).filter(
Workflow.status.startswith('assistance')).all()
])
wfs = []
wfs.extend(getWorkflows(url, 'running-open', details=True))
def recoveror(url, specific, options=None):
if userLock('recoveror'): return
up = componentInfo(soft=['mcm', 'wtc'])
if not up.check(): return
CI = campaignInfo()
SI = siteInfo()
UC = unifiedConfiguration()
use_recoveror = UC.get('use_recoveror')
if not use_recoveror and not options.go:
print "We are told not to run recoveror"
return
def make_int_keys(d):
for code in d:
d[int(code)] = d.pop(code)
error_codes_to_recover = UC.get('error_codes_to_recover')
error_codes_to_block = UC.get('error_codes_to_block')
error_codes_to_notify = UC.get('error_codes_to_notify')
make_int_keys(error_codes_to_recover)
make_int_keys(error_codes_to_block)
make_int_keys(error_codes_to_notify)
#wfs = session.query(Workflow).filter(Workflow.status == 'assistance-recovery').all()
wfs = session.query(Workflow).filter(
Workflow.status.contains('recovery')).all()
if specific:
wfs.extend(
session.query(Workflow).filter(
Workflow.status == 'assistance-manual').all())
for wfo in wfs:
if specific and not specific in wfo.name: continue
if not specific and 'manual' in wfo.status: continue
wfi = workflowInfo(url, wfo.name)
## need a way to verify that this is the first round of ACDC, since the second round will have to be on the ACDC themselves
all_errors = {}
try:
## this is clearly very truncated and should be changed completely
wfi.getSummary()
all_errors = wfi.summary['errors']
except:
pass
print '-' * 100
print "Looking at", wfo.name, "for recovery options"
recover = True
if not 'MergedLFNBase' in wfi.request:
print "f****d up"
sendEmail('missing lfn', '%s wl cache is screwed up' % wfo.name)
recover = False
if not len(all_errors):
print "\tno error for", wfo.name
recover = False
task_to_recover = defaultdict(list)
message_to_ops = ""
message_to_user = ""
if 'LheInputFilese' in wfi.request and wfi.request['LheInputFiles']:
## we do not try to recover pLHE
recover = False
if wfi.request['RequestType'] in ['MonteCarlo', 'ReReco']:
recover = False
if 'Campaign' in wfi.request:
c = wfi.request['Campaign']
if c in CI.campaigns and 'recover' in CI.campaigns[c]:
recover = CI.campaigns[c]['recover']
for task, errors in all_errors.items():
print "\tTask", task
## collect all error codes and #jobs regardless of step at which it occured
all_codes = []
for name, codes in errors.items():
if type(codes) == int: continue
all_codes.extend([
(int(code), info['jobs'], name,
list(set([e['type'] for e in info['errors']])),
list(set([e['details'] for e in info['errors']])))
for code, info in codes.items()
])
all_codes.sort(key=lambda i: i[1], reverse=True)
sum_failed = sum([l[1] for l in all_codes])
for errorCode, njobs, name, types, details in all_codes:
rate = 100 * njobs / float(sum_failed)
#print ("\t\t %10d (%6s%%) failures with error code %10d (%"+str(max_legend)+"s) at stage %s")%(njobs, "%4.2f"%rate, errorCode, legend, name)
print(
"\t\t %10d (%6s%%) failures with error code %10d (%30s) at stage %s"
) % (njobs, "%4.2f" % rate, errorCode, ','.join(types), name)
added_in_recover = False
#if options.go:
# force the recovery of any task with error ?
if errorCode in error_codes_to_recover:
## the error code is registered
for case in error_codes_to_recover[errorCode]:
match = case['details']
matched = (match == None)
if not matched:
matched = False
for detail in details:
if match in detail:
print "[recover] Could find keyword", match, "in"
print 50 * "#"
print detail
print 50 * "#"
matched = True
break
if matched and rate > case['rate']:
print "\t\t => we should be able to recover that", case[
'legend']
task_to_recover[task].append((code, case))
added_in_recover = True
message_to_user = ""
else:
print "\t\t recoverable but not frequent enough, needs", case[
'rate']
if errorCode in error_codes_to_block:
for case in error_codes_to_block[errorCode]:
match = case['details']
matched = (match == None)
if not matched:
matched = False
for detail in details:
if match in detail:
print "[block] Could find keyword", match, "in"
print 50 * "#"
print detail
print 50 * "#"
matched = True
break
if matched and rate > case['rate']:
print "\t\t => that error means no ACDC on that workflow", case[
'legend']
if not options.go:
message_to_ops += "%s has an error %s blocking an ACDC.\n%s\n " % (
wfo.name, errorCode, '#' * 50)
recover = False
added_in_recover = False
if errorCode in error_codes_to_notify and not added_in_recover:
print "\t\t => we should notify people on this"
message_to_user += "%s has an error %s in processing.\n%s\n" % (
wfo.name, errorCode, '#' * 50)
if message_to_user:
print wfo.name, "to be notified to user(DUMMY)", message_to_user
if message_to_ops:
#sendEmail( "notification in recoveror" , message_to_ops, destination=['*****@*****.**'])
sendLog('recoveror', message_to_ops, level='warning')
if len(task_to_recover) != len(all_errors):
print "Should not be doing partial ACDC. skipping"
#sendEmail('recoveror','do not want to make partial acdc on %s'%wfo.name)
sendLog('recoveror',
'do not want to make partial acdc on %s' % wfo.name,
level='warning')
recover = False
if task_to_recover and recover:
print "Initiating recovery"
print ', '.join(task_to_recover.keys()), "to be recovered"
recovering = set()
for task in task_to_recover:
print "Will be making a recovery workflow for", task
## from here you can fetch known solutions, to known error codes
actions = list(
set([
case['solution']
for code, case in task_to_recover[task]
]))
acdc = singleRecovery(url,
task,
wfi.request,
actions,
do=options.do)
if not acdc:
if options.do:
if recovering:
print wfo.name, "has been partially ACDCed. Needs manual attention"
#sendEmail( "failed ACDC partial recovery","%s has had %s/%s recoveries %s only"%( wfo.name, len(recovering), len(task_to_recover), list(recovering)), destination=['*****@*****.**'])
sendLog('recoveror',
"%s has had %s/%s recoveries %s only" %
(wfo.name, len(recovering),
len(task_to_recover), list(recovering)),
level='critical')
continue
else:
print wfo.name, "failed recovery once"
#break
continue
else:
print "no action to take further"
sendLog('recoveror',
"ACDC for %s can be done automatically" %
wfo.name,
level='critical')
continue
## and assign it ?
team = wfi.request['Team']
#assign_to_sites = set(SI.sites_ready) ## that needs to be massaged to prevent assigning to something out.
assign_to_sites = set(SI.all_sites)
parameters = {
#'SiteWhitelist' : wfi.request['SiteWhitelist'],
'SiteWhitelist': sorted(assign_to_sites),
'AcquisitionEra': wfi.acquisitionEra(),
'ProcessingString': wfi.processingString(),
'MergedLFNBase': wfi.request['MergedLFNBase'],
'ProcessingVersion': wfi.request['ProcessingVersion'],
}
## hackery for ACDC merge assignment
if wfi.request[
'RequestType'] == 'TaskChain' and 'Merge' in task.split(
'/')[-1]:
parameters['AcquisitionEra'] = None
parameters['ProcessingString'] = None
## xrootd setttings on primary and secondary
if 'TrustSitelists' in wfi.request and wfi.request[
'TrustSitelists']:
parameters['TrustSitelists'] = True
if 'TrustPUSitelists' in wfi.request and wfi.request[
'TrustPUSitelists']:
parameters['TrustPUSitelists'] = True
if options.ass:
print "really doing the assignment of the ACDC", acdc
parameters['execute'] = True
wfi.sendLog('recoveror',
"%s was assigned for recovery" % acdc)
else:
print "no assignment done with this ACDC", acdc
sendLog('recoveror',
"%s needs to be assigned" % (acdc),
level='critical')
result = reqMgrClient.assignWorkflow(url, acdc, team,
parameters)
if not result:
print acdc, "was not asigned"
sendLog('recoveror',
"%s needs to be assigned" % (acdc),
level='critical')
else:
recovering.add(acdc)
current = None
if recovering:
#if all went well, set the status to -recovering
current = wfo.status
if options.ass:
current = current.replace('recovery', 'recovering')
else:
current = 'assistance-manual'
print 'created ACDC: ' + ', '.join(recovering)
else:
## was set to be recovered, and no acdc was made
current = 'assistance-manual'
if current:
print wfo.name, "setting the status to", current
wfo.status = current
session.commit()
else:
## this workflow should be handled manually at that point
print wfo.name, "needs manual intervention"
wfo.status = 'assistance-manual'
session.commit()
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):
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 stagor(url,specific =None, options=None):
if not componentInfo().check(): return
SI = siteInfo()
CI = campaignInfo()
UC = unifiedConfiguration()
done_by_wf_id = {}
done_by_input = {}
completion_by_input = {}
good_enough = 100.0
lost_blocks = json.loads(open('%s/lost_blocks_datasets.json'%monitor_dir).read())
lost_files = json.loads(open('%s/lost_files_datasets.json'%monitor_dir).read())
known_lost_blocks = {}
known_lost_files = {}
for dataset in set(lost_blocks.keys()+lost_files.keys()):
b,f = findLostBlocksFiles(url, dataset)
if dataset in lost_blocks and not b:
print dataset,"has no really lost blocks"
else:
known_lost_blocks[dataset] = [i['name'] for i in b]
if dataset in lost_files and not f:
print dataset,"has no really lost files"
else:
known_lost_files[dataset] = [i['name'] for i in f]
try:
cached_transfer_statuses = json.loads(open('cached_transfer_statuses.json').read())
except:
print "inexisting transfer statuses. starting fresh"
cached_transfer_statuses = {}
transfer_statuses = {}
## pop all that are now in negative values
for phedexid in cached_transfer_statuses.keys():
transfers = session.query(Transfer).filter(Transfer.phedexid==int(phedexid)).all()
if not transfers:
print phedexid,"does not look relevant to be in cache anymore. poping"
print cached_transfer_statuses.pop( phedexid )
## collect all datasets that are needed for wf in staging, correcting the status of those that are not really in staging
wfois = []
needs = defaultdict(list)
for wfo in session.query(Workflow).filter(Workflow.status == 'staging').all():
wfi = workflowInfo(url, wfo.name)
if wfi.request['RequestStatus'] in ['running-open','running-closed','completed','assigned','acquired']:
wfi.sendLog('stagor', "is in status %s"%wfi.request['RequestStatus'])
wfi.status='away'
session.commit()
continue
if not wfi.request['RequestStatus'] in ['assignment-approved']:
## should be setting 'away' too
print wfo.name,"is",wfi.request['RequestStatus']
sendEmail("wrong status in staging. debug","%s is in %s, should set away."%(wfo.name,wfi.request['RequestStatus']))
wfois.append( (wfo,wfi) )
_,primaries,_,secondaries = wfi.getIO()
for dataset in list(primaries)+list(secondaries):
needs[wfo.name].append( dataset)
done_by_input[dataset] = {}
completion_by_input[dataset] = {}
wfi.sendLog('stagor', '%s needs %s'%( wfo.name, dataset))
open('%s/dataset_requirements.json'%monitor_dir,'w').write( json.dumps( needs, indent=2))
dataset_endpoints = defaultdict(set)
endpoint_in_downtime = defaultdict(set)
#endpoint_completed = defaultdict(set)
endpoint_incompleted = defaultdict(set)
#endpoint = defaultdict(set)
send_back_to_considered = set()
## phedexid are set negative when not relevant anymore
# probably there is a db schema that would allow much faster and simpler query
for transfer in session.query(Transfer).filter(Transfer.phedexid>0).all():
if specific and str(transfer.phedexid)!=str(specific): continue
skip=True
for wfid in transfer.workflows_id:
tr_wf = session.query(Workflow).get(wfid)
if tr_wf:
if tr_wf.status == 'staging':
sendLog('stagor',"\t%s is staging for %s"%(transfer.phedexid, tr_wf.name))
skip=False
if skip:
sendLog('stagor',"setting %s to negative value"%transfer.phedexid)
transfer.phedexid = -transfer.phedexid
session.commit()
continue
if transfer.phedexid<0: continue
## check the status of transfers
checks = checkTransferApproval(url, transfer.phedexid)
approved = all(checks.values())
if not approved:
sendLog('stagor', "%s is not yet approved"%transfer.phedexid)
approveSubscription(url, transfer.phedexid)
continue
## check on transfer completion
if str(transfer.phedexid) in cached_transfer_statuses:
### use a cache for transfer that already looked done
sendLog('stagor',"read %s from cache"%transfer.phedexid)
checks = cached_transfer_statuses[str(transfer.phedexid)]
else:
checks = checkTransferStatus(url, transfer.phedexid, nocollapse=True)
## just write this out
transfer_statuses[str(transfer.phedexid)] = copy.deepcopy(checks)
if not specific:
for dsname in checks:
if not dsname in done_by_input: done_by_input[dsname]={}
if not dsname in completion_by_input: completion_by_input[dsname] = {}
done_by_input[dsname][transfer.phedexid]=all(map(lambda i:i>=good_enough, checks[dsname].values()))
completion_by_input[dsname][transfer.phedexid]=checks[dsname].values()
if checks:
sendLog('stagor',"Checks for %s are %s"%( transfer.phedexid, [node.values() for node in checks.values()]))
done = all(map(lambda i:i>=good_enough,list(itertools.chain.from_iterable([node.values() for node in checks.values()]))))
else:
## it is empty, is that a sign that all is done and away ?
print "ERROR with the scubscriptions API of ",transfer.phedexid
print "Most likely something else is overiding the transfer request. Need to work on finding the replacement automatically, if the replacement exists"
done = False
## the thing above is NOT giving the right number
#done = False
for wfid in transfer.workflows_id:
tr_wf = session.query(Workflow).get(wfid)
if tr_wf:# and tr_wf.status == 'staging':
if not tr_wf.id in done_by_wf_id: done_by_wf_id[tr_wf.id]={}
done_by_wf_id[tr_wf.id][transfer.phedexid]=done
## for those that are in staging, and the destination site is in drain
#if not done and tr_wf.status == 'staging':
for ds in checks:
for s,v in checks[ds].items():
dataset_endpoints[ds].add( s )
if done:
## transfer.status = 'done'
sendLog('stagor',"%s is done"%transfer.phedexid)
cached_transfer_statuses[str(transfer.phedexid)] = copy.deepcopy(checks)
else:
sendLog('stagor',"%s is not finished %s"%(transfer.phedexid, pprint.pformat( checks )))
pprint.pprint( checks )
## check if the destination is in down-time
for ds in checks:
sites_incomplete = [SI.SE_to_CE(s) for s,v in checks[ds].items() if v<good_enough]
sites_incomplete_down = [s for s in sites_incomplete if not s in SI.sites_ready]
if sites_incomplete_down:
sendLog('stagor',"%s are in downtime, while waiting for %s to get there"%( ",".join(sites_incomplete_down), ds))
#sites_complete = [SI.SE_to_CE(s) for s,v in checks[ds].items() if v>=good_enough]
#endpoint[ds].update( sites_complete )
#endpoint[ds].update( sites_incomplete )
#endpoint_completed[ds].update( sites_complete )
endpoint_incompleted[ds].update( sites_incomplete )
endpoint_in_downtime[ds].update( sites_incomplete_down )
print "End point in down time"
for k in endpoint_in_downtime: endpoint_in_downtime[k] = list(endpoint_in_downtime[k])
for k in dataset_endpoints: dataset_endpoints[k] = list(dataset_endpoints[k])
print json.dumps( endpoint_in_downtime , indent=2)
open('cached_transfer_statuses.json','w').write( json.dumps( cached_transfer_statuses, indent=2))
open('%s/transfer_statuses.json'%monitor_dir,'w').write( json.dumps( transfer_statuses, indent=2))
open('%s/dataset_endpoints.json'%monitor_dir,'w').write( json.dumps(dataset_endpoints, indent=2))
already_stuck = json.loads( open('%s/stuck_transfers.json'%monitor_dir).read() )
missing_in_action = defaultdict(list)
print "-"*10,"Checking on workflows in staging","-"*10
#forget_about = ['/MinBias_TuneCUETP8M1_13TeV-pythia8/RunIISummer15GS-MCRUN2_71_V1-v2/GEN-SIM']
#for what in forget_about:
# if not done_by_input[what]:
# done_by_input[what] = {'fake':True}
## come back to workflows and check if they can go
available_cache = defaultdict(lambda : defaultdict(float))
presence_cache = defaultdict(dict)
for wfo,wfi in wfois:
print "#"*30
## the site white list takes site, campaign, memory and core information
(_,primaries,_,secondaries,sites_allowed) = wfi.getSiteWhiteList(verbose=False)
se_allowed = [SI.CE_to_SE(site) for site in sites_allowed]
se_allowed.sort()
se_allowed_key = ','.join(se_allowed)
readys={}
for need in list(primaries)+list(secondaries):
if not need in done_by_input:
wfi.sendLog('stagor',"missing transfer report for %s"%need)
readys[need] = False
## should warn someone about this !!!
## it cannot happen, by construction
sendEmail('missing transfer report','%s does not have a transfer report'%(need))
continue
if not done_by_input[need] and need in list(secondaries):
wfi.sendLog('stagor',"assuming it is OK for secondary %s to have no attached transfers"% need)
readys[need] = True
done_by_input[need] = { "fake" : True }
continue
if len(done_by_input[need]) and all(done_by_input[need].values()):
wfi.sendLog('stagor',"%s is ready"%need)
print json.dumps( done_by_input[need] , indent=2)
readys[need] = True
else:
wfi.sendLog('stagor',"%s is not ready"%need)
print json.dumps( done_by_input[need] , indent=2)
readys[need] = False
if readys and all(readys.values()):
if wfo.status == 'staging':
wfi.sendLog('stagor',"all needs are fullfilled, setting staged")
wfo.status = 'staged'
session.commit()
else:
wfi.sendLog('stagor',"all needs are fullfilled, already")
print json.dumps( readys, indent=2 )
else:
wfi.sendLog('stagor',"missing requirements")
copies_needed,_ = wfi.getNCopies()
jump_ahead = False
re_transfer = False
## there is missing input let's do something more elaborated
for need in list(primaries):#+list(secondaries):
if endpoint_in_downtime[need] == endpoint_incompleted[need]:
#print need,"is going to an end point in downtime"
wfi.sendLog('stagor',"%s has only incomplete endpoint in downtime"%need)
re_transfer=True
if not se_allowed_key in available_cache[need]:
available_cache[need][se_allowed_key] = getDatasetBlocksFraction( url , need, sites=se_allowed )
if available_cache[need][se_allowed_key] >= copies_needed:
wfi.sendLog('stagor',"assuming it is OK to move on like this already for %s"%need)
jump_ahead = True
## compute a time since staging to filter jump starting ?
# check whether the inputs is already in the stuck list ...
for need in list(primaries)+list(secondaries):
if need in already_stuck:
wfi.sendLog('stagor',"%s is stuck, so try to jump ahead"%need)
jump_ahead = True
if jump_ahead or re_transfer:
details_text = "checking on availability for %s to jump ahead"%wfo.name
details_text += '\n%s wants %s copies'%(wfo.name,copies_needed)
copies_needed = max(1,copies_needed-1)
details_text += '\nlowering by one unit to %s'%copies_needed
wfi.sendLog('stagor', details_text)
all_check = True
prim_where = set()
for need in list(primaries):
if not se_allowed_key in presence_cache[need]:
presence_cache[need][se_allowed_key] = getDatasetPresence( url, need , within_sites=se_allowed)
presence = presence_cache[need][se_allowed_key]
prim_where.update( presence.keys() )
available = available_cache[need][se_allowed_key]
this_check = (available >= copies_needed)
wfi.sendLog('stagor', "%s is available %s times %s"%( need, available, this_check))
all_check &= this_check
if not all_check: break
for need in list(secondaries):
## I do not want to check on the secon
this_check = all(done_by_input[need].values())
wfi.sendLog('stagor',"%s is all transfered %s"%(need, json.dumps(done_by_input[need], indent=2)))
all_check&= this_check
#if not se_allowed_key in presence_cache[need]:
# presence_cache[need][se_allowed_key] = getDatasetPresence( url, need , within_sites=se_allowed)
## restrict to where the primary is
#presence = dict([(k,v) for (k,v) in presence_cache[need][se_allowed_key].items() if k in prim_where])
#this_check = all([there for (there,frac) in presence.values()])
#print need,"is present at all sites:",this_check
#all_check&= this_check
if all_check:
wfi.sendLog('stagor',"needs are sufficiently fullfilled, setting staged")
wfo.status = 'staged'
session.commit()
else:
print wfo.name,"has to wait a bit more"
wfi.sendLog('stagor',"needs to wait a bit more")
else:
wfi.sendLog('stagor',"not checking availability")
if re_transfer:
wfi.sendLog('stagor',"Sending back to considered because of endpoint in downtime")
if wfo.status == 'staging':
wfo.status = 'considered'
session.commit()
send_back_to_considered.add( wfo.name )
if send_back_to_considered:
#sendEmail("transfer to endpoint in downtime","sending back to considered the following workflows \n%s"%('\n'.join( send_back_to_considered)))
sendLog('stagor', "sending back to considered the following workflows \n%s"%('\n'.join( send_back_to_considered)), level='critical')
print "-"*10,"Checking on non-available datasets","-"*10
## now check on those that are not fully available
for dsname in available_cache.keys():
## squash the se_allowed_key key
available_cache[dsname] = min( available_cache[dsname].values() )
for dsname,available in available_cache.items():
using_its = getWorkflowByInput(url, dsname)
#print using_its
using_wfos = []
for using_it in using_its:
wf = session.query(Workflow).filter(Workflow.name == using_it).first()
if wf:
using_wfos.append( wf )
if not len(done_by_input[dsname]):
print "For dataset",dsname,"there are no transfer report. That's an issue."
for wf in using_wfos:
if wf.status == 'staging':
if UC.get("stagor_sends_back"):
print "sending",wf.name,"back to considered"
wf.status = 'considered'
session.commit()
#sendEmail( "send back to considered","%s was send back and might be trouble"% wf.name)
sendLog('stagor', "%s was send back and might be trouble"% wf.name, level='critical')
else:
print "would send",wf.name,"back to considered"
#sendEmail( "subscription lagging behind","susbscriptions to get %s running are not appearing in phedex. I would have send it back to considered but that's not good."% wf.name)
sendLog('stagor', "susbscriptions to get %s running are not appearing in phedex. I would have send it back to considered but that's not good."% wf.name, level='critical')
continue
## not compatible with checking on secondary availability
#if all([wf.status != 'staging' for wf in using_wfos]):
# ## means despite all checks that input is not needed
# continue
if available < 1.:
print "incomplete",dsname
## there is a problem in the method below that it does not account for files stuck in T1*Buffer only
lost_blocks,lost_files = findLostBlocksFiles( url, dsname )
lost_block_names = [item['name'] for item in lost_blocks]
lost_file_names = [item['name'] for item in lost_files]
if lost_blocks:
#print json.dumps( lost , indent=2 )
## estimate for how much !
fraction_loss,_,n_missing = getDatasetBlockFraction(dsname, lost_block_names)
print "We have lost",len(lost_block_names),"blocks",lost_block_names,"for %f%%"%(100.*fraction_loss)
if fraction_loss > 0.05: ## 95% completion mark
#sendEmail('we have lost too many blocks','%s is missing %d blocks, for %d events, %f %% loss'%(dsname, len(lost_block_names), n_missing, fraction_loss))
sendLog('stagor', '%s is missing %d blocks, for %d events, %3.2f %% loss'%(dsname, len(lost_block_names), n_missing, 100*fraction_loss), level='warning')
## the workflow should be rejected !
for wf in using_wfos:
if wf.status == 'staging':
print wf.name,"is doomed. setting to trouble"
wf.status = 'trouble'
session.commit()
#sendEmail('doomed workflow','%s has too much loss on the input dataset %s. please check on stagor logs https://cmst2.web.cern.ch/cmst2/unified/logs/stagor/last.log'%(wf.name, dsname))
sendLog('stagor', '%s has too much loss on the input dataset %s. Missing %d blocks, for %d events, %3.2f %% loss'%(wf.name, dsname, len(lost_block_names), n_missing, 100*fraction_loss), level='critical')
else:
## probably enough to make a ggus and remove
if not dsname in known_lost_blocks:
#sendEmail('we have lost a few blocks', '%s is missing %d blocks, for %d events, %f %% loss\n\n%s'%(dsname, len(lost_block_names), n_missing, fraction_loss, '\n'.join( lost_block_names ) ))
sendLog('stagor', '%s is missing %d blocks, for %d events, %f %% loss\n\n%s'%(dsname, len(lost_block_names), n_missing, fraction_loss, '\n'.join( lost_block_names ) ), level='critical')
known_lost_blocks[dsname] = [i['name'] for i in lost_blocks]
if lost_files:
fraction_loss,_,n_missing = getDatasetFileFraction(dsname, lost_file_names)
print "We have lost",len(lost_file_names),"files",lost_file_names,"for %f%%"%fraction_loss
if fraction_loss > 0.05:
#sendEmail('we have lost too many files','%s is missing %d files, for %d events, %f %% loss'%(dsname, len(lost_file_names),n_missing, fraction_loss))
sendLog('stagor', '%s is missing %d files, for %d events, %f %% loss'%(dsname, len(lost_file_names),n_missing, fraction_loss), level='critical')
for wf in using_wfos:
if wf.status == 'staging':
print wf.name,"is doomed. setting to trouble"
wf.status = 'trouble'
session.commit()
else:
## probably enough to make a ggus and remove
if not dsname in known_lost_files:
#sendEmail('we have lost a few files','%s is missing %d files, for %d events, %f %% loss\n\n%s'%(dsname, len(lost_file_names),n_missing, fraction_loss, '\n'.join(lost_file_names)))
sendLog('stagor', '%s is missing %d files, for %d events, %f %% loss\n\n%s'%(dsname, len(lost_file_names),n_missing, fraction_loss, '\n'.join(lost_file_names)), level='critical')
known_lost_files[dsname] = [i['name'] for i in lost_files]
## should the status be change to held-staging and pending on a ticket
missings = [pid for (pid,d) in done_by_input[dsname].items() if d==False]
print "\t",done_by_input[dsname]
print "\tneeds",len(done_by_input[dsname])
print "\tgot",done_by_input[dsname].values().count(True)
print "\tmissing",missings
missing_in_action[dsname].extend( missings )
rr= open('%s/lost_blocks_datasets.json'%monitor_dir,'w')
rr.write( json.dumps( known_lost_blocks, indent=2))
rr.close()
rr= open('%s/lost_files_datasets.json'%monitor_dir,'w')
rr.write( json.dumps( known_lost_files, indent=2))
rr.close()
open('%s/incomplete_transfers.json'%monitor_dir,'w').write( json.dumps(missing_in_action, indent=2) )
print "Stuck transfers and datasets"
print json.dumps( missing_in_action, indent=2 )
print "Going further and make a report of stuck transfers"
datasets_by_phid = defaultdict(set)
for dataset in missing_in_action:
for phid in missing_in_action[dataset]:
#print dataset,"stuck through",phid
datasets_by_phid[phid].add( dataset )
bad_destinations = defaultdict(set)
bad_sources = defaultdict(set)
report = ""
really_stuck_dataset = set()
transfer_timeout = UC.get("transfer_timeout")
transfer_lowrate = UC.get("transfer_lowrate")
for phid,datasets in datasets_by_phid.items():
issues = checkTransferLag( url, phid , datasets=list(datasets) )
for dataset in issues:
for block in issues[dataset]:
for destination in issues[dataset][block]:
(block_size,destination_size,delay,rate,dones) = issues[dataset][block][destination]
## count x_Buffer and x_MSS as one source
redones=[]
for d in dones:
if d.endswith('Buffer') or d.endswith('Export'):
if d.replace('Buffer','MSS').replace('Export','MSS') in dones:
continue
else:
redones.append( d )
else:
redones.append( d )
dones = list(set( redones ))
#dones = filter(lambda s : (s.endswith('Buffer') and not s.replace('Buffer','MSS') in dones) or (not s.endswith('Buffer')) , dones)
if delay>transfer_timeout and rate<transfer_lowrate:
if len(dones)>1:
## its the destination that sucks
bad_destinations[destination].add( block )
else:
dum=[bad_sources[d].add( block ) for d in dones]
really_stuck_dataset.add( dataset )
print "add",dataset,"to really stuck"
report += "%s is not getting to %s, out of %s faster than %f [GB/s] since %f [d]\n"%(block,destination,", ".join(dones), rate, delay)
print "\n"*2
## create tickets right away ?
report+="\nbad sources "+",".join(bad_sources.keys())+"\n"
for site,blocks in bad_sources.items():
report+="\n\n%s:"%site+"\n\t".join(['']+list(blocks))
report+="\nbad destinations "+",".join(bad_destinations.keys())+"\n"
for site,blocks in bad_destinations.items():
report+="\n\n%s:"%site+"\n\t".join(['']+list(blocks))
print '\n'*2,"Datasets really stuck"
print '\n'.join( really_stuck_dataset )
print '\n'*2,"report written at https://cmst2.web.cern.ch/cmst2/unified/logs/incomplete_transfers.log"
print report
stuck_transfers = dict([(k,v) for (k,v) in missing_in_action.items() if k in really_stuck_dataset])
print '\n'*2,'Stuck dataset transfers'
print json.dumps(stuck_transfers , indent=2)
open('%s/stuck_transfers.json'%monitor_dir,'w').write( json.dumps(stuck_transfers , indent=2) )
open('%s/logs/incomplete_transfers.log'%monitor_dir,'w').write( report )
def transferor(url ,specific = None, talk=True, options=None):
if userLock(): return
if duplicateLock(): return
use_mcm = True
up = componentInfo(mcm=use_mcm, soft=['mcm'])
if not up.check(): return
use_mcm = up.status['mcm']
if options and options.test:
execute = False
else:
execute = True
SI = siteInfo()
CI = campaignInfo()
NLI = newLockInfo()
mcm = McMClient(dev=False)
dss = DSS()
#allowed_secondary = UC.get('')
print "counting all being handled..."
being_handled = len(session.query(Workflow).filter(Workflow.status == 'away').all())
being_handled += len(session.query(Workflow).filter(Workflow.status.startswith('stag')).all())
being_transfered = len(session.query(Workflow).filter(Workflow.status == 'staging').all())
being_handled += len(session.query(Workflow).filter(Workflow.status.startswith('assistance-')).all())
max_to_handle = options.maxworkflows
max_to_transfer = options.maxstaging
allowed_to_handle = max(0,max_to_handle - being_handled)
allowed_to_transfer = max(0,max_to_transfer - being_transfered)
wf_buffer = 5
if allowed_to_handle<=wf_buffer: ## buffer for having several wf per transfer
print "Not allowed to run more than",max_to_handle,"at a time. Currently",being_handled,"and",wf_buffer,"buffer"
else:
print being_handled,"already being handled",max_to_handle,"max allowed,",allowed_to_handle,"remaining","and",wf_buffer,"buffer"
if allowed_to_transfer <= wf_buffer:
print "Not allowed to transfer more than",max_to_transfer,"at a time. Currently",being_transfered,"and",wf_buffer,"buffer"
else:
print being_transfered,"already being transfered",max_to_transfer,"max allowed,",allowed_to_transfer,"remaining","and",wf_buffer,"buffer"
print "... done"
all_transfers=defaultdict(list)
workflow_dependencies = defaultdict(set) ## list of wf.id per input dataset
wfs_and_wfh=[]
print "getting all wf to consider ..."
cache = getWorkflows(url, 'assignment-approved', details=True)
for wfo in session.query(Workflow).filter(Workflow.status.startswith('considered')).all():
print "\t",wfo.name
if specific and not specific in wfo.name: continue
cache_r =filter(lambda d:d['RequestName']==wfo.name, cache)
if len(cache_r):
wfs_and_wfh.append( (wfo, workflowInfo( url, wfo.name, spec=False, request = cache_r[0]) ) )
else:
wfs_and_wfh.append( (wfo, workflowInfo( url, wfo.name, spec=False) ) )
print "... done"
transfers_per_sites = defaultdict(int)
input_sizes = {}
ignored_input_sizes = {}
input_cput = {}
input_st = {}
## list the size of those in transfer already
in_transfer_priority=None
min_transfer_priority=None
print "getting all wf in staging ..."
stucks = json.loads(open('%s/stuck_transfers.json'%monitor_dir).read())
for wfo in session.query(Workflow).filter(Workflow.status=='staging').all():
wfh = workflowInfo( url, wfo.name, spec=False)
#(lheinput,primary,parent,secondary) = wfh.getIO()
#sites_allowed = getSiteWhiteList( (lheinput,primary,parent,secondary) )
(lheinput,primary,parent,secondary,sites_allowed) = wfh.getSiteWhiteList()
for site in sites_allowed: ## we should get the actual transfer destination instead of the full white list
transfers_per_sites[site] += 1
#input_cput[wfo.name] = wfh.getComputingTime()
#input_st[wfo.name] = wfh.getSystemTime()
for prim in primary:
ds_s = dss.get( prim )
if prim in stucks:
sendLog('transferor', "%s appears stuck, so not counting it %s [GB]"%( prim, ds_s), wfi=wfh)
ignored_input_sizes[prim] = ds_s
else:
input_sizes[prim] = ds_s
sendLog('transferor', "%s needs %s [GB]"%( wfo.name, ds_s), wfi=wfh)
if in_transfer_priority==None:
in_transfer_priority = int(wfh.request['RequestPriority'])
else:
in_transfer_priority = max(in_transfer_priority, int(wfh.request['RequestPriority']))
if min_transfer_priority==None:
min_transfer_priority = int(wfh.request['RequestPriority'])
else:
min_transfer_priority = min(min_transfer_priority, int(wfh.request['RequestPriority']))
if min_transfer_priority==None or in_transfer_priority ==None:
print "nothing is lining up for transfer"
sendEmail("no request in staging","no request in staging")
return
pass
try:
print "Ignored input sizes"
ignored_values = list(ignored_input_sizes.items())
ignored_values.sort( key = lambda i : i[1] )
print "\n".join( map(str, ignored_values ) )
print "Considered input sizes"
considered_values = list(input_sizes.items())
considered_values.sort( key = lambda i : i[1] )
print "\n".join( map(str, considered_values) )
except Exception as e:
print "trying to print the summary of input size"
print str(e)
print "... done"
print "Max priority in transfer already",in_transfer_priority
print "Min priority in transfer already",min_transfer_priority
print "transfers per sites"
print json.dumps( transfers_per_sites, indent=2)
in_transfer_already = sum(input_sizes.values())
cput_in_transfer_already = sum(input_cput.values())
st_in_transfer_already = sum(input_st.values())
## list the size of all inputs
primary_input_per_workflow_gb = defaultdict(float)
print "getting all input sizes ..."
for (wfo,wfh) in wfs_and_wfh:
(_,primary,_,_) = wfh.getIO()
#input_cput[wfo.name] = wfh.getComputingTime()
#input_st[wfo.name] = wfh.getSystemTime()
for prim in primary:
## do not count it if it appears stalled !
prim_size = dss.get( prim )
input_sizes[prim] = prim_size
primary_input_per_workflow_gb[wfo.name] += prim_size
print "... done"
# shuffle first by name
random.shuffle( wfs_and_wfh )
# Sort smallest transfers first; allows us to transfer as many as possible workflows.
def prio_and_size( i, j):
if int(i[1].request['RequestPriority']) == int(j[1].request['RequestPriority']):
return cmp(int(primary_input_per_workflow_gb.get(j[0].name, 0)), int(primary_input_per_workflow_gb.get(i[0].name, 0)) )
else:
return cmp(int(i[1].request['RequestPriority']),int(j[1].request['RequestPriority']))
#wfs_and_wfh.sort(cmp = prio_and_size, reverse=True)
#wfs_and_wfh.sort(cmp = lambda i,j : cmp(int(primary_input_per_workflow_gb.get(i[0].name, 0)), int(primary_input_per_workflow_gb.get(j[0].name, 0)) ))
#sort by priority higher first
wfs_and_wfh.sort(cmp = lambda i,j : cmp(int(i[1].request['RequestPriority']),int(j[1].request['RequestPriority']) ), reverse=True)
cput_grand_total = sum(input_cput.values())
cput_to_transfer = cput_grand_total - cput_in_transfer_already
st_grand_total = sum(input_st.values())
st_to_transfer = st_grand_total - st_in_transfer_already
print "%15.4f [CPU h] worth already in transfer"%cput_in_transfer_already
print "%15.4f [CPU h] worth is the current requested transfer load"%cput_to_transfer
print "%15.4f [h] worth of absolute system time in transfer"%( cput_in_transfer_already / SI.availableSlots())
print "%15.4f [h] worth of absolute system time is the current requested transfer load"%( cput_to_transfer / SI.availableSlots())
print "%15.4f [h] worth of theoritical system time in transfer"%( st_in_transfer_already )
print "%15.4f [h] worth of theoritical system time is the current requested transfer load"%( st_to_transfer )
grand_total = sum(input_sizes.values())
to_transfer = grand_total - in_transfer_already
grand_transfer_limit = options.maxtransfer
#grand_transfer_limit = SI.total_disk()*0.25*1024## half of the free sapce in TB->GB
transfer_limit = grand_transfer_limit - in_transfer_already
print "%15.4f GB already being transfered"%in_transfer_already
print "%15.4f GB is the current requested transfer load"%to_transfer
print "%15.4f GB is the global transfer limit"%grand_transfer_limit
print "%15.4f GB is the available limit"%transfer_limit
max_staging_per_site = options.maxstagingpersite
# the max priority value per dataset.
max_priority = defaultdict(int)
needs_transfer=0 ## so that we can count'em
passing_along = 0
transfer_sizes={}
went_over_budget=False
destination_cache = {}
no_goes = set()
max_per_round = UC.get('max_per_round').get('transferor',None)
if max_per_round and not spec:
wfs_and_wfh = wfs_and_wfh[:max_per_round]
for (wfo,wfh) in wfs_and_wfh:
print wfo.name,"to be transfered with priority",wfh.request['RequestPriority']
if wfh.request['RequestStatus']!='assignment-approved':
if wfh.request['RequestStatus'] in ['aborted','rejected','rejected-archived','aborted-archived']:
wfo.status = 'trouble' ## so that we look or a replacement
else:
wfo.status = 'away'
wfh.sendLog('transferor', '%s in status %s, setting %s'%( wfo.name,wfh.request['RequestStatus'],wfo.status))
continue
(_,primary,_,_) = wfh.getIO()
this_load=sum([input_sizes[prim] for prim in primary])
no_budget = False
if ( this_load and (sum(transfer_sizes.values())+this_load > transfer_limit or went_over_budget ) ):
if went_over_budget:
wfh.sendLog('transferor', "Transfer has gone over bubget.")
else:
wfh.sendLog('transferor', "Transfer will go over bubget.")
wfh.sendLog('transferor', "%15.4f GB this load, %15.4f GB already this round, %15.4f GB is the available limit"%(this_load, sum(transfer_sizes.values()), transfer_limit))
#if sum(transfer_sizes.values()) > transfer_limit:
went_over_budget = True
if in_transfer_priority!=None and min_transfer_priority!=None:
if int(wfh.request['RequestPriority']) >= in_transfer_priority and min_transfer_priority!=in_transfer_priority:
wfh.sendLog('transferor',"Higher priority sample %s >= %s go-on over budget"%( wfh.request['RequestPriority'], in_transfer_priority))
else:
if not options.go:
wfh.sendLog('transferor',"%s minimum priority %s < %s : stop"%( min_transfer_priority,wfh.request['RequestPriority'],in_transfer_priority))
no_budget = True
## throtlle by campaign go
no_go = False
if not wfh.go(log=True) and not options.go:
no_go = True
no_goes.add( wfo.name )
allowed_secondary = set()
for campaign in wfh.getCampaigns():
if campaign in CI.campaigns and 'secondaries' in CI.campaigns[campaign]:
allowed_secondary.update( CI.campaigns[campaign]['secondaries'] )
if secondary:
if (secondary and allowed_secondary) and (set(secondary)&allowed_secondary!=set(secondary)):
wfh.sendLog('assignor','%s is not an allowed secondary'%(', '.join(set(secondary)-allowed_secondary)))
no_go = True
if no_go:
continue
## check if the batch is announced
def check_mcm(wfn):
announced=False
is_real=False
if not wfn.startswith('pdmvserv'):
is_real = True
try:
for b in mcm.getA('batches',query='contains=%s'% wfo.name):
is_real = True
if b['status']=='announced':
announced=True
break
except:
try:
for b in mcm.getA('batches',query='contains=%s'% wfo.name):
is_real = True
if b['status']=='announced':
announced=True
break
except:
print "could not get mcm batch announcement, assuming not real"
return announced,is_real
if not use_mcm:
announced,is_real = False,True
else:
if wfh.request['RequestType'] in ['ReReco']:
announced,is_real = True,True
else:
announced,is_real = check_mcm( wfo.name )
if not announced:
wfh.sendLog('transferor', "does not look announced.")
if not is_real:
wfh.sendLog('transferor', "does not appear to be genuine.")
## prevent any duplication. if the wf is not mentioned in any batch, regardless of status
continue
## check on a grace period
injection_time = time.mktime(time.strptime('.'.join(map(str,wfh.request['RequestDate'])),"%Y.%m.%d.%H.%M.%S")) / (60.*60.)
now = time.mktime(time.gmtime()) / (60.*60.)
if float(now - injection_time) < 4.:
if not options.go and not announced:
wfh.sendLog('transferor', "It is too soon to start transfer: %3.2fH remaining"%(now - injection_time))
continue
if passing_along >= allowed_to_handle:
#if int(wfh.request['RequestPriority']) >= in_transfer_priority and min_transfer_priority!=in_transfer_priority:
if in_transfer_priority!=None and min_transfer_priority!=None:
if int(wfh.request['RequestPriority']) >= in_transfer_priority and int(wfh.request['RequestPriority']) !=min_transfer_priority:
## higher priority, and not only this priority being transfered
wfh.sendLog('transferor',"Higher priority sample %s >= %s go-on over %s"%( wfh.request['RequestPriority'], in_transfer_priority, max_to_handle))
else:
wfh.sendLog('transferor'," Not allowed to pass more than %s at a time. Currently %s handled, and adding %s"%( max_to_handle, being_handled, passing_along))
if not options.go:
## should not allow to jump that fence
break
if this_load and needs_transfer >= allowed_to_transfer:
if in_transfer_priority!=None and min_transfer_priority!=None:
if int(wfh.request['RequestPriority']) >= in_transfer_priority and int(wfh.request['RequestPriority']) !=min_transfer_priority:
## higher priority, and not only this priority being transfered
wfh.sendLog('transferor',"Higher priority sample %s >= %s go-on over %s"%(wfh.request['RequestPriority'], in_transfer_priority,max_to_transfer))
else:
wfh.sendLog('transferor',"Not allowed to transfer more than %s at a time. Currently %s transfering, and adding %s"%( max_to_transfer, being_transfered, needs_transfer))
if not options.go:
no_budget = True
if no_budget:
continue
## the site white list considers site, campaign, memory and core information
(lheinput,primary,parent,secondary,sites_allowed) = wfh.getSiteWhiteList()
if options and options.tosites:
sites_allowed = options.tosites.split(',')
for dataset in list(primary)+list(parent)+list(secondary):
## lock everything flat
NLI.lock( dataset )
if not sites_allowed:
wfh.sendLog('transferor',"not possible site to run at")
#sendEmail("no possible sites","%s has no possible sites to run at"%( wfo.name ))
sendLog('transferor',"%s has no possible sites to run at"%( wfo.name ),level='critical')
continue
blocks = []
if 'BlockWhitelist' in wfh.request and wfh.request['BlockWhitelist']:
blocks = wfh.request['BlockWhitelist']
if 'RunWhitelist' in wfh.request and wfh.request['RunWhitelist']:
## augment with run white list
for dataset in primary:
blocks = list(set( blocks + getDatasetBlocks( dataset, runs=wfh.request['RunWhitelist'] ) ))
if 'LumiList' in wfh.request and wfh.request['LumiList']:
## augment with the lumi white list
blocks = list(set( blocks + getDatasetBlocks( dataset, lumis= wfh.request['LumiList'] ) ))
if blocks:
print "Reading",len(blocks),"in block whitelist"
can_go = True
staging=False
allowed=True
primary_destinations = set()
if primary:
copies_needed_from_CPUh,CPUh = wfh.getNCopies()
if talk:
print wfo.name,'reads',', '.join(primary),'in primary'
## chope the primary dataset
for prim in primary:
## keep track of what needs what
workflow_dependencies[prim].add( wfo.id )
max_priority[prim] = max(max_priority[prim],int(wfh.request['RequestPriority']))
wfh.sendLog('transferor',"Would make %s from cpu requirement %s"%( copies_needed_from_CPUh, CPUh))
copies_needed = copies_needed_from_CPUh
if 'Campaign' in wfh.request and wfh.request['Campaign'] in CI.campaigns and 'maxcopies' in CI.campaigns[wfh.request['Campaign']]:
copies_needed_from_campaign = CI.campaigns[wfh.request['Campaign']]['maxcopies']
copies_needed = min(copies_needed_from_campaign, copies_needed)
wfh.sendLog('transferor',"Maxed to %s by campaign configuration %s"%( copies_needed, wfh.request['Campaign']))
### new ways of making the whole thing
destinations,all_block_names = getDatasetDestinations(url, prim, within_sites = [SI.CE_to_SE(site) for site in sites_allowed], only_blocks=blocks )
print json.dumps(destinations, indent=2)
## get where the dataset is in full and completed
prim_location = [site for (site,info) in destinations.items() if info['completion']==100 and info['data_fraction']==1]
## the rest is places it is going to be
prim_destination = [site for site in destinations.keys() if not site in prim_location]
if len(prim_location) >= copies_needed:
wfh.sendLog('transferor',"The input is all fully in place at %s sites %s"%( len(prim_location), sorted(prim_location)))
continue
copies_needed = max(0,copies_needed - len(prim_location))
wfh.sendLog('transferor',"not counting existing copies ; now need %s"% copies_needed)
copies_being_made = [ sum([info['blocks'].keys().count(block) for site,info in destinations.items() if site in prim_destination]) for block in all_block_names]
latching_on_transfers = set()
[latching_on_transfers.update(info['blocks'].values()) for site,info in destinations.items() if site in prim_destination]
latching_on_transfers = list(latching_on_transfers)
#print latching_on_transfers
## figure out where all this is going to go
prim_to_distribute = [site for site in sites_allowed if not SI.CE_to_SE(site) in prim_location]
prim_to_distribute = [site for site in prim_to_distribute if not SI.CE_to_SE(site) in prim_destination]
## take out the ones that cannot receive transfers
prim_to_distribute = [site for site in prim_to_distribute if not any([osite.startswith(site) for osite in SI.sites_veto_transfer])]
wfh.sendLog('transferor',"Could be going to: %s"% sorted( prim_to_distribute))
if not prim_to_distribute or any([transfers_per_sites[site] < max_staging_per_site for site in prim_to_distribute]):
## means there is openings let me go
print "There are transfer slots available:",[(site,transfers_per_sites[site]) for site in prim_to_distribute]
#for site in sites_allowed:
# #increment accross the board, regardless of real destination: could be changed
# transfers_per_sites[site] += 1
else:
if int(wfh.request['RequestPriority']) >= in_transfer_priority and min_transfer_priority!=in_transfer_priority:
wfh.sendLog('transferor', "Higher priority sample %s >= %s go-on over transfer slots available"%(wfh.request['RequestPriority'], in_transfer_priority))
else:
wfh.sendLog('transferor',"Not allowed to transfer more than %s per site at a time. Going overboard for %s"%( max_staging_per_site, sorted([site for site in prim_to_distribute if transfers_per_sites[site]>=max_staging_per_site])))
if not options.go:
allowed = False
break
for latching in latching_on_transfers:
tfo = session.query(Transfer).filter(Transfer.phedexid == int(latching)).first()
if not tfo:
tfo = session.query(Transfer).filter(Transfer.phedexid == -int(latching)).first()
if not tfo:
tfo = Transfer( phedexid = latching)
tfo.workflows_id = []
session.add(tfo)
else:
tfo.phedexid = latching ## make it positive ever
if not wfo.id in tfo.workflows_id:
print "adding",wfo.id,"to",tfo.id,"with phedexid",latching
l = copy.deepcopy( tfo.workflows_id )
l.append( wfo.id )
tfo.workflows_id = l
if not options.test:
session.commit()
else:
session.flush() ## regardless of commit later on, we need to let the next wf feeding on this transfer to see it in query
can_go = False
transfer_sizes[prim] = input_sizes[prim]
staging = True
# reduce the number of copies required by the on-going full transfer : how do we bootstrap on waiting for them ??
#copies_needed = max(0,copies_needed - len(prim_destination))
copies_needed = max(0,copies_needed - min(copies_being_made))
wfh.sendLog('transferor', "Not counting the copies being made ; then need %s"% copies_needed)
if copies_needed == 0:
wfh.sendLog('transferor', "The output is either fully in place or getting in full somewhere with %s"% latching_on_transfers)
can_go = True
continue
elif len(prim_to_distribute)==0:
wfh.sendLog('transferor', "We are going to need extra copies, but no destinations seems available")
prim_to_distribute = [site for site in sites_allowed if not SI.CE_to_SE(site) in prim_location]
prim_to_distribute = [site for site in prim_to_distribute if not any([osite.startswith(site) for osite in SI.sites_veto_transfer])]
if len(prim_to_distribute)>0: ## maybe that a parameter we can play with to limit the
if not options or options.chop:
### hard include the tape disk andpoint ?
#tapes = [site for site in getDatasetPresence( url, prim, vetos=['T0','T2','T3','Disk']) if site.endswith('MSS')]
chops,sizes = getDatasetChops(prim, chop_threshold = options.chopsize, only_blocks=blocks)
spreading = distributeToSites( chops, prim_to_distribute, n_copies = copies_needed, weights=SI.cpu_pledges, sizes=sizes)
transfer_sizes[prim] = sum(sizes)
if not spreading:
sendLog('transferor','cannot send %s to any site, it cannot fit anywhere'% prim, level='critical')
wfh.sendLog('transferor', "cannot send to any site. %s cannot seem to fit anywhere"%(prim))
staging=False
can_go = False
else:
spreading = {}
for site in prim_to_distribute:
if blocks:
spreading[site]=blocks
else:
spreading[site]=[prim]
transfer_sizes[prim] = input_sizes[prim] ## this is approximate if blocks are specified
can_go = False
wfh.sendLog('transferor', "selected CE destinations %s"%(sorted( spreading.keys())))
for (site,items) in spreading.items():
all_transfers[site].extend( items )
transfers_per_sites[site] += 1
primary_destinations.add( site )
if not allowed:
wfh.sendLog('transferor', "Not allowed to move on with")
continue
if secondary:
override_sec_destination = []
if 'SecondaryLocation' in CI.campaigns[wfh.request['Campaign']]:
override_sec_destination = CI.campaigns[wfh.request['Campaign']]['SecondaryLocation']
print wfo.name,'reads',', '.join(secondary),'in secondary'
for sec in secondary:
workflow_dependencies[sec].add( wfo.id )
if True:
## new style, failing on minbias
if not sec in destination_cache:
## this is barbbaric, and does not show the correct picture on workflow by workflow with different whitelist
destination_cache[sec],_ = getDatasetDestinations(url, sec) ## NO SITE WHITE LIST ADDED
#destination_cache[sec],_ = getDatasetDestinations(url, sec, within_sites = [SI.CE_to_SE(site) for site in sites_allowed])
## limit to the site whitelist NOW
se_allowed = [SI.CE_to_SE(site) for site in sites_allowed]
destinations = dict([(k,v) for (k,v) in destination_cache[sec].items() if site in se_allowed])
## truncate location/destination to those making up for >90% of the dataset
bad_destinations = [destinations.pop(site) for (site,info) in destinations.items() if info['data_fraction']<0.9]
sec_location = [site for (site,info) in destinations.items() if info['completion']>=95]
sec_destination = [site for site in destinations.keys() if not site in sec_location]
else:
## old style
presence = getDatasetPresence( url, sec )
sec_location = [site for site,pres in presence.items() if pres[1]>90.] ## more than 90% of the minbias at sites
subscriptions = listSubscriptions( url ,sec )
sec_destination = [site for site in subscriptions]
sec_to_distribute = [site for site in sites_allowed if not any([osite.startswith(site) for osite in sec_location])]
sec_to_distribute = [site for site in sec_to_distribute if not any([osite.startswith(site) for osite in sec_destination])]
sec_to_distribute = [site for site in sec_to_distribute if not any([osite.startswith(site) for osite in SI.sites_veto_transfer])]
if override_sec_destination:
## intersect with where we want the PU to be
not_needed_anymore = list(set(sec_to_distribute) - set(override_sec_destination))
#sendEmail("secondary superfluous","the dataset %s could be removed from %s"%( sec, not_needed_anymore ))
sendLog('transferor', "the dataset %s could be removed from %s"%( sec, not_needed_anymore ))
sec_to_distribute = list(set(sec_to_distribute) & set(override_sec_destination))
if len( sec_to_distribute )>0:
print "secondary could go to",sorted(sec_to_distribute)
sec_size = dss.get( sec )
for site in sec_to_distribute:
site_se =SI.CE_to_SE(site)
if (SI.disk[site_se]*1024.) > sec_size:
all_transfers[site].append( sec )
can_go = False
else:
print "could not send the secondary input to",site_se,"because it is too big for the available disk",SI.disk[site_se]*1024,"GB need",sec_size
if primary_destinations and site in primary_destinations:
#sendEmail('secondary input too big','%s is too big (%s) for %s (%s)'%( sec, sec_size, site_se, SI.disk[site_se]*1024))
sendLog('transferor', '%s is too big (%s) for %s (%s)'%( sec, sec_size, site_se, SI.disk[site_se]*1024), level='critical')
else:
print "the secondary input does not have to be send to site"
## is that possible to do something more
if can_go:
## no explicit transfer required this time
if staging:
## but using existing ones
wfh.sendLog('transferor', "latches on existing transfers, and nothing else, settin staging")
wfo.status = 'staging'
needs_transfer+=1
else:
wfh.sendLog('transferor', "should just be assigned now to %s"%sorted(sites_allowed))
wfo.status = 'staged'
passing_along+=1
wfh.sendLog('transferor', "setting status to %s"%wfo.status)
session.commit()
continue
else:
## there is an explicit transfer required
if staging:
## and also using an existing one
wfh.sendLog('transferor', "latches on existing transfers")
if not options.test:
wfo.status = 'staging'
wfh.sendLog('transferor', "setting status to %s"%wfo.status)
session.commit()
wfh.sendLog('transferor',"needs a transfer")
needs_transfer+=1
passing_along+=1
if no_goes:
#sendEmail("no go for managing","No go for \n"+"\n".join( no_goes ))
sendLog('transferor', "No go for \n"+"\n".join( no_goes ), level='critical')
print "accumulated transfers"
print json.dumps(all_transfers, indent=2)
fake_id=-1
wf_id_in_prestaging=set()
for (site,items_to_transfer) in all_transfers.iteritems():
items_to_transfer = list(set(items_to_transfer))
## convert to storage element
site_se = SI.CE_to_SE(site)
## site that do not want input datasets
if site in SI.sites_veto_transfer:
print site,"does not want transfers"
continue
## throttle the transfer size to T2s ? we'd be screwed by a noPU sample properly configured.
## massage a bit the items
blocks = [it for it in items_to_transfer if '#' in it]
block_datasets = list(set([it.split('#')[0] for it in blocks]))
datasets = [it for it in items_to_transfer if not '#' in it]
details_text = "Making a replica to %s (CE) %s (SE) for"%( site, site_se)
#print "\t",len(blocks),"blocks"
## remove blocks if full dataset is send out
blocks = [block for block in blocks if not block.split('#')[0] in datasets]
#print "\t",len(blocks),"needed blocks for",list(set([block.split('#')[0] for block in blocks]))
#print "\t",len(datasets),"datasets"
#print "\t",datasets
details_text += '\n\t%d blocks'%len(blocks)
details_text += '\n\t%d needed blocks for %s'%( len(blocks), sorted(list(set([block.split('#')[0] for block in blocks]))))
details_text += '\n\t%d datasets'% len(datasets)
details_text += '\n\t%s'%sorted(datasets)
items_to_transfer = blocks + datasets
if execute:
sendLog('transferor', details_text)
else:
print "Would make a replica to",site,"(CE)",site_se,"(SE) for"
print details_text
## operate the transfer
if options and options.stop:
## ask to move-on
answer = raw_input('Continue with that ?')
if not answer.lower() in ['y','yes','go']:
continue
if execute:
priority = 'normal'
cds = [ds for ds in datasets+block_datasets if ds in max_priority]
if cds and False: ## I don't think this is working. subscription should be updated on the fly and regularly for raising the priority if needed
## decide on an overall priority : that's a bit too large though
if any([max_priority[ds]>=90000 for ds in cds]):
priority = 'high'
elif all([max_priority[ds]<80000 for ds in cds]):
priority = 'low'
result = makeReplicaRequest(url, site_se, items_to_transfer, 'prestaging', priority=priority)
else:
result= {'phedex':{'request_created' : []}}
fake_id-=1
if not result:
print "ERROR Could not make a replica request for",site,items_to_transfer,"pre-staging"
continue
for phedexid in [o['id'] for o in result['phedex']['request_created']]:
new_transfer = session.query(Transfer).filter(Transfer.phedexid == int(phedexid)).first()
if not new_transfer:
new_transfer = session.query(Transfer).filter(Transfer.phedexid == -int(phedexid)).first()
print phedexid,"transfer created"
if not new_transfer:
new_transfer = Transfer( phedexid = phedexid)
session.add( new_transfer )
else:
new_transfer.phedexid = phedexid ## make it positive again
new_transfer.workflows_id = set()
for transfering in list(set(map(lambda it : it.split('#')[0], items_to_transfer))):
new_transfer.workflows_id.update( workflow_dependencies[transfering] )
new_transfer.workflows_id = list(new_transfer.workflows_id)
wf_id_in_prestaging.update(new_transfer.workflows_id)
session.commit()
## auto approve it
if execute:
approved = approveSubscription(url, phedexid, [site_se])
for wfid in wf_id_in_prestaging:
tr_wf = session.query(Workflow).get(wfid)
if tr_wf and tr_wf.status!='staging':
if execute:
tr_wf.status = 'staging'
if talk:
print "setting",tr_wf.name,"to staging"
session.commit()
def assignor(url ,specific = None, talk=True, options=None):
if userLock(): return
if duplicateLock(): return
if not componentInfo().check(): return
UC = unifiedConfiguration()
CI = campaignInfo()
#SI = siteInfo()
SI = global_SI()
#NLI = newLockInfo()
#if not NLI.free() and not options.go: return
LI = lockInfo()
if not LI.free() and not options.go: return
n_assigned = 0
n_stalled = 0
wfos=[]
fetch_from = []
if specific or options.early:
fetch_from.extend(['considered','staging'])
if specific:
fetch_from.extend(['considered-tried'])
fetch_from.extend(['staged'])
if options.from_status:
fetch_from = options.from_status.split(',')
print "Overriding to read from",fetch_from
for status in fetch_from:
wfos.extend(session.query(Workflow).filter(Workflow.status==status).all())
## in case of partial, go for fetching a list from json ?
#if options.partial and not specific:
# pass
dataset_endpoints = json.loads(open('%s/dataset_endpoints.json'%monitor_dir).read())
aaa_mapping = json.loads(open('%s/equalizor.json'%monitor_pub_dir).read())['mapping']
all_stuck = set()
all_stuck.update( json.loads( open('%s/stuck_transfers.json'%monitor_dir).read() ))
all_stuck.update( getAllStuckDataset())
max_per_round = UC.get('max_per_round').get('assignor',None)
max_cpuh_block = UC.get('max_cpuh_block')
random.shuffle( wfos )
for wfo in wfos:
if options.limit and (n_stalled+n_assigned)>options.limit:
break
if max_per_round and (n_stalled+n_assigned)>max_per_round:
break
if specific:
if not any(map(lambda sp: sp in wfo.name,specific.split(','))): continue
#if not specific in wfo.name: continue
print "\n\n"
wfh = workflowInfo( url, wfo.name)
if options.priority and int(wfh.request['RequestPriority']) < options.priority:
continue
options_text=""
if options.early: options_text+=", early option is ON"
if options.partial:
options_text+=", partial option is ON"
options_text+=", good fraction is %.2f"%options.good_enough
wfh.sendLog('assignor',"%s to be assigned%s"%(wfo.name, options_text))
## the site whitelist takes into account siteInfo, campaignInfo, memory and cores
(lheinput,primary,parent,secondary, sites_allowed) = wfh.getSiteWhiteList()
output_tiers = list(set([o.split('/')[-1] for o in wfh.request['OutputDatasets']]))
is_stuck = (all_stuck & primary)
if is_stuck:
wfh.sendLog('assignor',"%s are stuck input"%(','.join( is_stuck)))
## check if by configuration we gave it a GO
no_go = False
if not wfh.go(log=True) and not options.go:
no_go = True
allowed_secondary = {}
assign_parameters = {}
check_secondary = False
for campaign in wfh.getCampaigns():
if campaign in CI.campaigns:
assign_parameters.update( CI.campaigns[campaign] )
if campaign in CI.campaigns and 'secondaries' in CI.campaigns[campaign]:
if CI.campaigns[campaign]['secondaries']:
allowed_secondary.update( CI.campaigns[campaign]['secondaries'] )
check_secondary = True
if campaign in CI.campaigns and 'banned_tier' in CI.campaigns[campaign]:
banned_tier = list(set(CI.campaigns[campaign]['banned_tier']) & set(output_tiers))
if banned_tier:
no_go=True
wfh.sendLog('assignor','These data tiers %s are not allowed'%(','.join( banned_tier)))
sendLog('assignor','These data tiers %s are not allowed'%(','.join( banned_tier)), level='critical')
if secondary and check_secondary:
if (set(secondary)&set(allowed_secondary.keys())!=set(secondary)):
wfh.sendLog('assignor','%s is not an allowed secondary'%(', '.join(set(secondary)-set(allowed_secondary.keys()))))
sendLog('assignor','%s is not an allowed secondary'%(', '.join(set(secondary)-set(allowed_secondary.keys()))), level='critical')
if not options.go:
no_go = True
## then get whether there is something more to be done by secondary
for sec in secondary:
if sec in allowed_secondary:# and 'parameters' in allowed_secondary[sec]:
assign_parameters.update( allowed_secondary[sec] )
if no_go:
n_stalled+=1
continue
## check on current status for by-passed assignment
if wfh.request['RequestStatus'] !='assignment-approved':
if not options.test:
wfh.sendLog('assignor',"setting %s away and skipping"%wfo.name)
## the module picking up from away will do what is necessary of it
wfo.wm_status = wfh.request['RequestStatus']
wfo.status = 'away'
session.commit()
continue
else:
print wfo.name,wfh.request['RequestStatus']
## retrieve from the schema, dbs and reqMgr what should be the next version
version=wfh.getNextVersion()
if not version:
if options and options.ProcessingVersion:
version = options.ProcessingVersion
else:
wfh.sendLog('assignor',"cannot decide on version number")
n_stalled+=1
wfo.status = 'trouble'
session.commit()
continue
original_sites_allowed = copy.deepcopy( sites_allowed )
wfh.sendLog('assignor',"Site white list %s"%sorted(sites_allowed))
override_sec_location = CI.get(wfh.request['Campaign'], 'SecondaryLocation', [])
blocks = []
if 'BlockWhitelist' in wfh.request:
blocks = wfh.request['BlockWhitelist']
if 'RunWhitelist' in wfh.request and wfh.request['RunWhitelist']:
## augment with run white list
for dataset in primary:
blocks = list(set( blocks + getDatasetBlocks( dataset, runs=wfh.request['RunWhitelist'] ) ))
wfh.sendLog('assignor',"Allowed %s"%sorted(sites_allowed))
secondary_locations=None
primary_aaa = options.primary_aaa
secondary_aaa = options.secondary_aaa
do_partial = False #options.good_enough if options.partial else 0
if 'Campaign' in wfh.request and wfh.request['Campaign'] in CI.campaigns:
assign_parameters.update( CI.campaigns[wfh.request['Campaign']] )
if 'primary_AAA' in assign_parameters:
primary_aaa = primary_aaa or assign_parameters['primary_AAA']
if 'secondary_AAA' in assign_parameters:
secondary_aaa = secondary_aaa or assign_parameters['secondary_AAA']
if 'partial_copy' in assign_parameters:
## can this only work if there is a stuck input ? maybe not
## this is a number. 0 means no
print "Could do partial disk copy assignment"
if is_stuck or options.partial:
do_partial = assign_parameters['partial_copy']
wfh.sendLog('assignor',"Overiding partial copy assignment to %.2f fraction"% do_partial)
#sendEmail('stuck input to assignment','%s is stuck for assigning %s and going fractional'%(','.join( is_stuck), wfo.name))
do_partial = options.good_enough if options.partial else do_partial
for sec in list(secondary):
if override_sec_location:
print "We don't care where the secondary is"
print "Cannot pass for now"
#sendEmail("tempting to pass sec location check","but we cannot yet IMO")
#pass
if secondary_aaa:
#just continue without checking
continue
presence = getDatasetPresence( url, sec )
print sec
print json.dumps(presence, indent=2)
one_secondary_locations = [site for (site,(there,frac)) in presence.items() if frac>98.]
#one_secondary_locations = [site for (site,(there,frac)) in presence.items() if there]
if secondary_locations==None:
secondary_locations = one_secondary_locations
else:
secondary_locations = list(set(secondary_locations) & set(one_secondary_locations))
## reduce the site white list to site with secondary only
#sites_allowed = [site for site in sites_allowed if any([osite.startswith(site) for osite in one_secondary_locations])]
sites_allowed = [site for site in sites_allowed if SI.CE_to_SE(site) in one_secondary_locations]
wfh.sendLog('assignor',"From secondary requirement, now Allowed%s"%sorted(sites_allowed))
initial_sites_allowed = copy.deepcopy( sites_allowed ) ## keep track of this, after secondary input location restriction : that's how you want to operate it
sites_all_data = copy.deepcopy( sites_allowed )
sites_with_data = copy.deepcopy( sites_allowed )
sites_with_any_data = copy.deepcopy( sites_allowed )
primary_locations = None
available_fractions = {}
set_lfn = '/store/mc' ## by default
endpoints = set()
for prim in list(primary):
if prim in dataset_endpoints:
print "endpoints from stagor",dataset_endpoints[prim]
endpoints.update( dataset_endpoints[prim] )
set_lfn = getLFNbase( prim )
presence = getDatasetPresence( url, prim , only_blocks=blocks)
if talk:
print prim
print json.dumps(presence, indent=2)
available_fractions[prim] = getDatasetBlocksFraction(url, prim, sites = [SI.CE_to_SE(site) for site in sites_allowed] , only_blocks = blocks)
#sites_all_data = [site for site in sites_with_data if any([osite.startswith(site) for osite in [psite for (psite,(there,frac)) in presence.items() if there]])]
#sites_with_data = [site for site in sites_with_data if any([osite.startswith(site) for osite in [psite for (psite,frac) in presence.items() if frac[1]>90.]])]
sites_all_data = [site for site in sites_with_data if SI.CE_to_SE(site) in [psite for (psite,(there,frac)) in presence.items() if there]]
sites_with_data = [site for site in sites_with_data if SI.CE_to_SE(site) in [psite for (psite,frac) in presence.items() if frac[1]>90.]]
sites_with_any_data = [site for site in sites_with_any_data if SI.CE_to_SE(site) in presence.keys()]
wfh.sendLog('assignor',"Holding the data but not allowed %s"%sorted(list(set([se_site for se_site in presence.keys() if not SI.SE_to_CE(se_site) in sites_allowed]))))
if primary_locations==None:
primary_locations = presence.keys()
else:
primary_locations = list(set(primary_locations) & set(presence.keys() ))
sites_with_data = list(set(sites_with_data))
sites_with_any_data = list(set(sites_with_any_data))
opportunistic_sites=[]
down_time = False
## opportunistic running where any piece of data is available
if secondary_locations or primary_locations:
## intersection of both any pieces of the primary and good IO
#opportunistic_sites = [SI.SE_to_CE(site) for site in list((set(secondary_locations) & set(primary_locations) & set(SI.sites_with_goodIO)) - set(sites_allowed))]
if secondary_locations and primary_locations:
opportunistic_sites = [SI.SE_to_CE(site) for site in list((set(secondary_locations) & set(primary_locations)) - set([SI.CE_to_SE(site) for site in sites_allowed]))]
elif primary_locations:
opportunistic_sites = [SI.SE_to_CE(site) for site in list(set(primary_locations) - set([SI.CE_to_SE(site) for site in sites_allowed]))]
else:
opportunistic_sites = []
wfh.sendLog('assignor',"We could be running in addition at %s"% sorted(opportunistic_sites))
if any([osite in SI.sites_not_ready for osite in opportunistic_sites]):
wfh.sendLog('assignor',"One of the usable site is in downtime %s"%([osite for osite in opportunistic_sites if osite in SI.sites_not_ready]))
down_time = True
## should this be send back to considered ?
## should be 2 but for the time-being let's lower it to get things going
copies_wanted,cpuh = wfh.getNCopies()
wfh.sendLog('assignor',"we need %s CPUh"%cpuh)
if cpuh>max_cpuh_block and not options.go:
#sendEmail('large workflow','that wf %s has a large number of CPUh %s, not assigning, please check the logs'%(wfo.name, cpuh))#,destination=['*****@*****.**'])
sendLog('assignor','%s requires a large numbr of CPUh %s , not assigning, please check with requester'%( wfo.name, cpuh), level='critical')
wfh.sendLog('assignor',"Requiring a large number of CPUh %s, not assigning"%cpuh)
continue
if 'Campaign' in wfh.request and wfh.request['Campaign'] in CI.campaigns and 'maxcopies' in CI.campaigns[wfh.request['Campaign']]:
copies_needed_from_campaign = CI.campaigns[wfh.request['Campaign']]['maxcopies']
copies_wanted = min(copies_needed_from_campaign, copies_wanted)
if not options.early:
less_copies_than_requested = UC.get("less_copies_than_requested")
copies_wanted = max(1,copies_wanted-less_copies_than_requested) # take one out for the efficiency
else:
## find out whether there is a site in the whitelist, that is lacking jobs and reduce to 1 copy needed to get things going
pass
wfh.sendLog('assignor',"needed availability fraction %s"% copies_wanted)
## should also check on number of sources, if large enough, we should be able to overflow most, efficiently
## default back to white list to original white list with any data
wfh.sendLog('assignor',"Allowed sites :%s"% sorted(sites_allowed))
if primary_aaa:
## remove the sites not reachable localy if not in having the data
if not sites_all_data:
wfh.sendLog('assignor',"Overiding the primary on AAA setting to Off")
primary_aaa=False
else:
aaa_grid = set(sites_all_data)
for site in list(aaa_grid):
aaa_grid.update( aaa_mapping.get(site,[]) )
sites_allowed = list(set(initial_sites_allowed) & aaa_grid)
wfh.sendLog('assignor',"Selected to read primary through xrootd %s"%sorted(sites_allowed))
if not primary_aaa:
sites_allowed = sites_with_any_data
wfh.sendLog('assignor',"Selected for any data %s"%sorted(sites_allowed))
### check on endpoints for on-going transfers
if do_partial:
if endpoints:
end_sites = [SI.SE_to_CE(s) for s in endpoints]
sites_allowed = list(set(sites_allowed + end_sites))
if down_time and not any(osite in SI.sites_not_ready for osite in end_sites):
print "Flip the status of downtime, since our destinations are good"
down_time = False
print "with added endpoints",sorted(end_sites)
else:
print "Cannot do partial assignment without knowin the endpoints"
n_stalled+=1
continue
#if not len(sites_allowed):
# if not options.early:
# wfh.sendLog('assignor',"cannot be assign with no matched sites")
# sendLog('assignor','%s has no whitelist'% wfo.name, level='critical')
# n_stalled+=1
# continue
low_pressure = SI.sites_low_pressure(0.4)
## if any of the site allowed is low pressure : reduce to 1 copy so that it gets started
allowed_and_low = sorted(set(low_pressure) & set(sites_allowed))
if allowed_and_low:
wfh.sendLog('assignor',"The workflow can run at %s under low pressure currently"%( ','.join( allowed_and_low )))
copies_wanted = max(1., copies_wanted-1.)
if available_fractions and not all([available>=copies_wanted for available in available_fractions.values()]):
not_even_once = not all([available>=1. for available in available_fractions.values()])
above_good = all([available >= do_partial for available in available_fractions.values()])
wfh.sendLog('assignor',"The input dataset is not available %s times, only %s"%( copies_wanted, available_fractions.values()))
if down_time and not options.go and not options.early:
wfo.status = 'considered'
session.commit()
wfh.sendLog('assignor',"sending back to considered because of site downtime, instead of waiting")
#sendEmail( "cannot be assigned due to downtime","%s is not sufficiently available, due to down time of a site in the whitelist. check the assignor logs. sending back to considered."% wfo.name)
sendLog('assignor','%s is not sufficiently available, due to down time of a site in the whitelist. sending back to considered.'%( wfo.name ), level='delay')
n_stalled+=1
continue
#pass
print json.dumps(available_fractions)
if (options.go and not_even_once) or not options.go:
known = []
try:
known = json.loads(open('cannot_assign.json').read())
except:
pass
if not wfo.name in known and not options.limit and not options.go and not options.early and not (do_partial and above_good):
wfh.sendLog('assignor',"cannot be assigned, %s is not sufficiently available.\n %s"%(wfo.name,json.dumps(available_fractions)))
#sendEmail( "cannot be assigned","%s is not sufficiently available.\n %s"%(wfo.name,json.dumps(available_fractions)))
known.append( wfo.name )
open('cannot_assign.json','w').write(json.dumps( known, indent=2))
if options.early:
if wfo.status == 'considered':
wfh.sendLog('assignor',"setting considered-tried")
wfo.status = 'considered-tried'
session.commit()
else:
print "tried but status is",wfo.status
if do_partial and above_good:
print "Will move on with partial locations"
else:
n_stalled+=1
continue
if not len(sites_allowed):
if not options.early:
wfh.sendLog('assignor',"cannot be assign with no matched sites")
sendLog('assignor','%s has no whitelist'% wfo.name, level='critical')
n_stalled+=1
continue
t1_only = [ce for ce in sites_allowed if ce.startswith('T1')]
if t1_only:
# try to pick from T1 only first
sites_out = [SI.pick_dSE([SI.CE_to_SE(ce) for ce in t1_only])]
else:
# then pick any otherwise
sites_out = [SI.pick_dSE([SI.CE_to_SE(ce) for ce in sites_allowed])]
wfh.sendLog('assignor',"Placing the output on %s"%sites_out)
parameters={
'SiteWhitelist' : sites_allowed,
'NonCustodialSites' : sites_out,
'AutoApproveSubscriptionSites' : list(set(sites_out)),
'AcquisitionEra' : wfh.acquisitionEra(),
'ProcessingString' : wfh.processingString(),
'MergedLFNBase' : set_lfn,
'ProcessingVersion' : version,
}
if primary_aaa:
parameters['TrustSitelists'] = True
wfh.sendLog('assignor',"Reading primary through xrootd at %s"%sorted(sites_allowed))
if secondary_aaa:
parameters['TrustPUSitelists'] = True
wfh.sendLog('assignor',"Reading secondary through xrootd at %s"%sorted(sites_allowed))
if 'parameters' in assign_parameters:
parameters.update( assign_parameters['parameters'] )
## plain assignment here
team='production'
if os.getenv('UNIFIED_TEAM'): team = os.getenv('UNIFIED_TEAM')
if options and options.team:
team = options.team
if False and 'T2_CH_CERN' in parameters['SiteWhitelist']:
## add some check on
### the amount pending to HLT
### the size of the request
### the priority of the request (maybe not if we decide to overflow during runs)
parameters['SiteWhitelist'] = ['T2_CH_CERN_HLT']
team = 'hlt'
## reduce the splitting by factor of 4, regardless of type of splitting
sendEmail("sending work to HLT","%s was assigned to HLT"%wfo.name)
##parse options entered in command line if any
if options:
for key in reqMgrClient.assignWorkflow.keys:
v=getattr(options,key)
if v!=None:
if type(v)==str and ',' in v:
parameters[key] = filter(None,v.split(','))
else:
parameters[key] = v
if lheinput:
## throttle reading LHE article
wfh.sendLog('assignor', 'Setting the number of events per job to 500k max')
parameters['EventsPerJob'] = 500000
## pick up campaign specific assignment parameters
#parameters.update( CI.parameters(wfh.request['Campaign']) )
parameters.update( assign_parameters.get('parameters',{}) )
if not options.test:
parameters['execute'] = True
split_check = wfh.checkWorkflowSplitting()
if split_check!=True:
parameters.update( split_check )
if 'NoGo' in split_check.values():
wfh.sendLog('assignor', "Failing splitting check")
sendLog('assignor','the workflow %s is failing the splitting check. Verify in the logs'% wfo.name, level='critical')
n_stalled+=1
continue
if 'EventBased' in split_check.values():
wfh.sendLog('assignor', "Falling back to event splitting.")
#sendEmail("Fallback to EventBased","the workflow %s is too heavy to be processed as it is. Fallback to EventBased splitting"%wfo.name)
sendLog('assignor','the workflow %s is too heavy to be processed as it is. Fallback to EventBased splitting ?'%wfo.name, level='critical')
## we have a problem here, that EventBased should never be used as a backup
if not options.go:
n_stalled+=1
continue
continue ## skip all together
elif 'EventsPerJob' in split_check.values():
wfh.sendLog('assignor', "Modifying the number of events per job")
#sendEmail("Modifying the job per events","the workflow %s is too heavy in number of jobs explosion"%wfo.name)
sendLog('assignor',"the workflow %s is too heavy in number of jobs explosion"%wfo.name, level='critical')
elif 'EventsPerLumi' in split_check.values():
wfh.sendLog('assignor', "Modifying the number of events per lumi to be able to process this")
# Handle run-dependent MC
pstring = wfh.processingString()
if 'PU_RD' in pstring:
numEvents = wfh.getRequestNumEvents()
eventsPerLumi = [getDatasetEventsPerLumi(prim) for prim in primary]
eventsPerLumi = sum(eventsPerLumi)/float(len(eventsPerLumi))
reqJobs = 500
if 'PU_RD2' in pstring:
reqJobs = 2000
eventsPerJob = int(numEvents/(reqJobs*1.4))
lumisPerJob = int(eventsPerJob/eventsPerLumi)
if lumisPerJob==0:
#sendEmail("issue with event splitting for run-dependent MC","%s needs to be split by event with %s per job"%(wfo.name, eventsPerJob))
sendLog('assignor', "%s needs to be split by event with %s per job"%(wfo.name, eventsPerJob), level='critical')
wfh.sendLog('assignor', "%s needs to be split by event with %s per job"%(wfo.name, eventsPerJob))
parameters['EventsPerJob'] = eventsPerJob
else:
spl = wfh.getSplittings()[0]
eventsPerJobEstimated = spl['events_per_job'] if 'events_per_job' in spl else None
eventsPerJobEstimated = spl['avg_events_per_job'] if 'avg_events_per_job' in spl else None
if eventsPerJobEstimated and eventsPerJobEstimated > eventsPerJob:
#sendEmail("setting lumi splitting for run-dependent MC","%s was assigned with %s lumis/job"%( wfo.name, lumisPerJob))
sendLog('assignor',"%s was assigned with %s lumis/job"%( wfo.name, lumisPerJob), level='critical')
wfh.sendLog('assignor',"%s was assigned with %s lumis/job"%( wfo.name, lumisPerJob))
parameters['LumisPerJob'] = lumisPerJob
else:
#sendEmail("leaving splitting untouched for PU_RD*","please check on "+wfo.name)
sendLog('assignor',"leaving splitting untouched for %s, please check on %s"%( pstring, wfo.name), level='critical')
wfh.sendLog('assignor',"leaving splitting untouched for PU_RD*, please check.")
result = reqMgrClient.assignWorkflow(url, wfo.name, team, parameters)
# set status
if not options.test:
if result:
wfo.status = 'away'
session.commit()
n_assigned+=1
wfh.sendLog('assignor',"Properly assigned\n%s"%(json.dumps( parameters, indent=2)))
try:
## refetch information and lock output
new_wfi = workflowInfo( url, wfo.name)
(_,prim,_,sec) = new_wfi.getIO()
for secure in list(prim)+list(sec)+new_wfi.request['OutputDatasets']:
## lock all outputs flat
#NLI.lock( secure )
LI.lock( secure, reason = 'assigning')
#for site in [SI.CE_to_SE(site) for site in sites_allowed]:
# for output in new_wfi.request['OutputDatasets']:
# LI.lock( output, site, 'dataset in production')
# for primary in prim:
# LI.lock( primary, site, 'dataset used in input')
# for secondary in sec:
# LI.lock( secondary, site, 'required for mixing' )
except Exception as e:
print "fail in locking output"
print str(e)
sendEmail("failed locking of output",str(e))
else:
wfh.sendLog('assignor',"Failed to assign. Please check the logs")
print "ERROR could not assign",wfo.name
else:
pass
print "Assignment summary:"
sendLog('assignor',"Assigned %d Stalled %s"%(n_assigned, n_stalled))
def spawn_harvesting(url, wfi, sites_for_DQMHarvest):
SI = global_SI()
all_OK = {}
requests = []
outputs = wfi.request['OutputDatasets']
if ('EnableHarvesting' in wfi.request
and not wfi.request['EnableHarvesting']) and (
'DQMConfigCacheID' in wfi.request
and wfi.request['DQMConfigCacheID']):
if not 'MergedLFNBase' in wfi.request:
print "f****d up"
sendEmail('screwed up wl cache',
'%s wl cache is bad' % (wfi.request['RequestName']))
all_OK['fake'] = False
return all_OK, requests
wfi = workflowInfo(url, wfi.request['RequestName'])
dqms = [out for out in outputs if '/DQM' in out]
#if not all([in_full[dqm_input] for dqm_input in dqms]):
# wfi.sendLog('closor',"will not be able to assign the harvesting: holding up")
# for dqm_input in dqms:
# all_OK[dqm_input] = False
## raise the subscription to high priority
for dqm_input in dqms:
## handle it properly
harvesting_schema = {
'Requestor': os.getenv('USER'),
'RequestType': 'DQMHarvest',
'Group': 'DATAOPS'
}
copy_over = [
'AcquisitionEra',
'ProcessingString',
'DQMUploadUrl',
'CMSSWVersion',
'CouchDBName',
'CouchWorkloadDBName',
'ConfigCacheUrl',
'DbsUrl',
'inputMode',
'DQMConfigCacheID',
'OpenRunningTimeout',
'ScramArch',
'CMSSWVersion',
'Campaign',
'Memory', #dummy
'SizePerEvent', #dummy
'GlobalTag', #dummy
]
for item in copy_over:
if item in wfi.request:
harvesting_schema[item] = copy.deepcopy(wfi.request[item])
else:
print item, "is not in initial schema"
harvesting_schema['InputDataset'] = dqm_input
harvesting_schema['TimePerEvent'] = 1
harvesting_schema['PrepID'] = 'Harvest-' + wfi.request['PrepID']
if len(wfi.request['RequestString']) > 60:
wfi.request['RequestString'] = wfi.request[
'RequestString'][:60]
print "truncating request string", wfi.request['RequestString']
harvesting_schema[
'RequestString'] = 'HARVEST-' + wfi.request['RequestString']
harvesting_schema['DQMHarvestUnit'] = 'byRun'
harvesting_schema['RequestPriority'] = min(
wfi.request['RequestPriority'] * 10, 999999)
harvest_request = reqMgrClient.submitWorkflow(
url, harvesting_schema)
if not harvest_request:
print "Error in making harvesting for", wfi.request[
'RequestName']
print "schema"
print json.dumps(harvesting_schema, indent=2)
harvest_request = reqMgrClient.submitWorkflow(
url, harvesting_schema)
if not harvest_request:
print "Error twice in harvesting for", wfi.request[
'RequestName']
print "schema"
print json.dumps(harvesting_schema, indent=2)
if harvest_request:
requests.append(harvest_request)
## should we protect for setting approved ? no, it's notified below, assignment will fail, likely
data = reqMgrClient.setWorkflowApproved(url, harvest_request)
print "created", harvest_request, "for harvesting of", dqm_input
wfi.sendLog(
'closor', "created %s for harvesting of %s" %
(harvest_request, dqm_input))
## assign it directly
team = wfi.request['Team']
parameters = {
'SiteWhitelist': [
SI.SE_to_CE(se)
for se in wfi.request['NonCustodialSites']
],
'AcquisitionEra':
wfi.acquisitionEra(),
'ProcessingString':
wfi.processingString(),
'MergedLFNBase':
wfi.request['MergedLFNBase'],
'ProcessingVersion':
wfi.request['ProcessingVersion'],
'execute':
True
}
#if in_full[dqm_input]:
# print "using full copy at",in_full[dqm_input]
# parameters['SiteWhitelist'] = [SI.SE_to_CE(se) for se in in_full[dqm_input]]
#else:
# print "cannot do anything if not having a full copy somewhere"
# all_OK[dqm_input]=False
# continue
parameters['SiteWhitelist'] = sites_for_DQMHarvest
result = reqMgrClient.assignWorkflow(url, harvest_request,
team, parameters)
if not result:
#sendEmail('harvesting request created','%s was created at announcement of %s in %s, failed to assign'%(harvest_request, dqm_input, wfi.request['RequestName']), destination=[wfi.request['Requestor']+'@cern.ch'])
wfi.sendLog(
'closor',
'%s was created at announcement of %s in %s, failed to assign'
% (harvest_request, dqm_input,
wfi.request['RequestName']))
sendLog(
'closor',
'%s was created at announcement of %s in %s, failed to assign'
% (harvest_request, dqm_input,
wfi.request['RequestName']),
level='critical')
else:
#sendEmail('harvesting request assigned','%s was created at announcement of %s in %s, and assigned'%(harvest_request, dqm_input, wfi.request['RequestName']), destination=[wfi.request['Requestor']+'@cern.ch'])
wfi.sendLog(
'closor',
'%s was created at announcement of %s in %s, and assigned'
% (harvest_request, dqm_input,
wfi.request['RequestName']))
else:
#print "could not make the harvesting for",wfo.name,"not announcing"
wfi.sendLog('closor', "could not make the harvesting request")
sendLog('closor',
"could not make the harvesting request for %s" %
wfi.request['RequestName'],
level='critical')
all_OK[dqm_input] = False
return (all_OK, requests)
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 closor(url, specific=None, options=None):
if userLock(): return
mlock = moduleLock()
if mlock() and not options.manual: return
up = componentInfo(soft=['mcm', 'wtc'])
if not up.check(): return
UC = unifiedConfiguration()
CI = campaignInfo()
BI = batchInfo()
CloseI = closeoutInfo()
all_late_files = []
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()
if specific:
wfs = [wfo for wfo in wfs if specific in wfo.name]
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_extreme_warnings = defaultdict(set)
batch_goodness = UC.get("batch_goodness")
closers = []
print len(wfs), "closing"
th_start = time.mktime(time.gmtime())
for iwfo, wfo in enumerate(wfs):
if specific and not specific in wfo.name: continue
if not options.manual and (
'cmsunified_task_HIG-RunIIFall17wmLHEGS-05036__v1_T_200712_005621_4159'
.lower() in (wfo.name).lower() or
'pdmvserv_task_HIG-RunIISummer16NanoAODv7-03979__v1_T_200915_013748_1986'
.lower() in (wfo.name).lower()):
continue
closers.append(
CloseBuster(
wfo=wfo,
url=url,
CI=CI,
UC=UC,
jump_the_line=jump_the_line,
batch_goodness=batch_goodness,
batch_go=batch_go,
#stats = stats,
batch_warnings=batch_warnings,
batch_extreme_warnings=batch_extreme_warnings,
all_late_files=all_late_files,
held=held,
))
run_threads = ThreadHandler(threads=closers,
n_threads=options.threads,
sleepy=10,
timeout=None,
verbose=True,
label='closor')
run_threads.start()
## waiting on all to complete
while run_threads.is_alive():
#print "Waiting on closing threads",time.asctime(time.gmtime())
time.sleep(5)
JC = JIRAClient() if up.status.get('jira', False) else None
print len(
run_threads.threads), "finished thread to gather information from"
failed_threads = 0
for to in run_threads.threads:
if to.failed:
failed_threads += 1
continue
if to.outs:
for outO in to.outs:
out = outO.datasetname
odb = session.query(Output).filter(
Output.datasetname == out).first()
if not odb:
print "adding an output object", out
session.add(outO)
else:
odb.date = outO.date
if to.to_status:
to.wfo.status = to.to_status
if JC and to.to_status == "done" and to.wfi:
jiras = JC.find({"prepid": to.wfi.request['PrepID']})
for jira in jiras:
JC.close(jira.key)
if to.to_wm_status:
to.wfo.wm_status = to.to_wm_status
if to.closing:
CloseI.pop(to.wfo.name)
session.commit()
th_stop = time.mktime(time.gmtime())
if wfs:
time_spend_per_workflow = (th_stop - th_start) / float(len(wfs))
print "Average time spend per workflow is", time_spend_per_workflow
if float(failed_threads / run_threads.n_threads) > 0:
sendLog('checkor',
'%d/%d threads have failed, better check this out' %
(failed_threads, run_threads.n_threads),
level='critical')
sendEmail(
'checkor', '%d/%d threads have failed, better check this out' %
(failed_threads, run_threads.n_threads))
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')
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"
if batch_extreme_warnings[bname]:
subject = "Low Statistics for %s" % bname
issues = "The following datasets have outstanding completion (<50%%) issues:\n\n"
issues += "\n".join(sorted(batch_extreme_warnings[bname]))
issues += "\n\n"
elif 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 = ""
text += "Dear all,\n\n"
text += "A batch of release validation workflows has finished.\n\n"
text += "Batch ID:\n\n"
text += "%s\n\n" % (bname)
text += "Detail of the workflows\n\n"
text += "https://dmytro.web.cern.ch/dmytro/cmsprodmon/requests.php?campaign=%s\n\n" % (
bname)
text += "%s\n\n" % (issues)
text += "This is an automated message.\n\n"
text += ""
to = ['*****@*****.**']
sendEmail(subject, text, destination=to)
## just announced ; take it out now.
BI.pop(bname)
deleteCampaignConfig(bname)
if os.path.isfile('.closor_stop'):
print "The loop on workflows was shortened"
sendEmail('closor',
'Closor loop was shortened artificially using .closor_stop')
os.system('rm -f .closor_stop')
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 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 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 injector(url, options, specific):
mlock = moduleLock()
if mlock(): return
use_mcm = True
up = componentInfo(soft=['mcm', 'wtc'])
if not up.check(): return
use_mcm = up.status['mcm']
UC = unifiedConfiguration()
transform_keywords = UC.get('convert_to_stepchain')
workflows = getWorkflows(url, status=options.wmstatus, user=options.user)
for user in UC.get("user_rereco"):
workflows.extend(
getWorkflows(url,
status=options.wmstatus,
user=user,
rtype="ReReco"))
for user in (options.user_relval.split(',')
if options.user_relval else UC.get("user_relval")):
workflows.extend(
getWorkflows(url,
status=options.wmstatus,
user=user,
rtype="TaskChain"))
for user in (options.user_storeresults.split(',') if
options.user_storeresults else UC.get("user_storeresults")):
workflows.extend(
getWorkflows(url,
status=options.wmstatus,
user=user,
rtype="StoreResults"))
print len(workflows), "in line"
cannot_inject = set()
to_convert = set()
status_cache = defaultdict(str)
## browse for assignment-approved requests, browsed for ours, insert the diff
for wf in workflows:
if specific and not specific in wf: continue
exists = session.query(Workflow).filter(Workflow.name == wf).first()
if not exists:
wfi = workflowInfo(url, wf)
## check first that there isn't related here with something valid
can_add = True
## first try at finding a match
familly = session.query(Workflow).filter(
Workflow.name.contains(wfi.request['PrepID'])).all()
if not familly:
pids = wfi.getPrepIDs()
req_familly = []
for pid in pids:
req_familly.extend(getWorkflowById(url, pid, details=True))
familly = []
print len(req_familly), "members"
for req_member in req_familly:
#print "member",req_member['RequestName']
owfi = workflowInfo(url,
req_member['RequestName'],
request=req_member)
other_pids = owfi.getPrepIDs()
if set(pids) == set(other_pids):
## this is a real match
familly.extend(
session.query(Workflow).filter(
Workflow.name ==
req_member['RequestName']).all())
for lwfo in familly:
if lwfo:
## we have it already
if not lwfo.status in [
'forget', 'trouble', 'forget-unlock',
'forget-out-unlock'
]:
wfi.sendLog(
'injector', "Should not put %s because of %s %s" %
(wf, lwfo.name, lwfo.status))
sendLog('injector',
"Should not put %s because of %s %s" %
(wf, lwfo.name, lwfo.status),
level='critical')
print "Should not put", wf, "because of", lwfo.name, lwfo.status
cannot_inject.add(wf)
can_add = False
## add a check on validity of input datasets
_, prim, par, sec = wfi.getIO()
for d in list(prim) + list(par) + list(sec):
if not d in status_cache:
status_cache[d] = getDatasetStatus(d)
if status_cache[d] != 'VALID':
wfi.sendLog(
'injector', "One of the input is not VALID. %s : %s" %
(d, status_cache[d]))
sendLog('injector',
"One of the input of %s is not VALID. %s : %s" %
(wf, d, status_cache[d]),
level='critical')
can_add = False
## check for any file in phedex, to verify existence
_, ph_files, _, _ = getDatasetFiles(url, d)
if not ph_files and not ('StoreResults'
== wfi.request.setdefault(
'RequestType', None)):
wfi.sendLog(
'injector',
"One of the input has no file in phedex: %s" % d)
sendLog('injector',
"One of the input has no file in phedex: %s" % d,
level='critical')
can_add = False
### ban some workflow that you don't like anymore
#outputs = wfi.request['OutputDatasets']
if not can_add: continue
## temporary hack to transform specific taskchain into stepchains
#good_for_stepchain = wfi.isGoodToConvertToStepChain( keywords = transform_keywords)
good_for_stepchain = wfi.isGoodToConvertToStepChain(keywords=None)
## match keywords and technical constraints
#if (not options.no_convert) and good_for_stepchain and not wfi.isRelval():
# to_convert.add( wf )
# wfi.sendLog('injector','Transforming %s TaskChain into StepChain'%wf)
# #sendEmail('convertion to stepchain','Transforming %s TaskChain into StepChain'%wf)
wfi.sendLog('injector', "considering %s" % wf)
new_wf = Workflow(name=wf,
status=options.setstatus,
wm_status=options.wmstatus)
session.add(new_wf)
session.commit()
time.sleep(0.5)
else:
#print "already have",wf
pass
if cannot_inject:
#sendEmail('workflow duplicates','These workflow cannot be added in because of duplicates \n\n %s'%( '\n'.join(cannot_inject)))
sendLog(
'injector',
'These workflow cannot be added in because of duplicates \n\n %s' %
('\n'.join(cannot_inject)),
level='warning')
for wf in to_convert:
os.system(
'./Unified/rejector.py --clone --to_step --comments \"Transform to StepChain\" %s'
% wf)
## passing a round of invalidation of what needs to be invalidated
if use_mcm and (options.invalidate or True):
invalidator(url)
no_replacement = set()
#print "getting all transfers"
#all_transfers=session.query(Transfer).all()
#print "go!"
## pick up replacements
for wf in session.query(Workflow).filter(
Workflow.status == 'trouble').all():
print wf.name
if specific and not specific in wf.name: continue
print wf.name
wfi = workflowInfo(url, wf.name)
wl = wfi.request #getWorkLoad(url, wf.name)
familly = getWorkflowById(url, wl['PrepID'])
true_familly = []
for member in familly:
if member == wf.name: continue
fwl = getWorkLoad(url, member)
if options.replace:
if member != options.replace: continue
else:
if fwl['RequestDate'] < wl['RequestDate']: continue
if fwl['RequestType'] == 'Resubmission': continue
if fwl['RequestStatus'] in ['None', None, 'new']: continue
if fwl['RequestStatus'] in [
'rejected', 'rejected-archived', 'aborted',
'aborted-archived'
]:
continue
true_familly.append(fwl)
if len(true_familly) == 0:
#sendLog('injector','%s had no replacement'%wf.name, level='critical')
if wfi.isRelval():
#wfi.sendLog('injector','the workflow was found in trouble with no replacement. As a relval, there is no clean way to handle this.')
wfi.sendLog(
'injector',
'the workflow was found in trouble with no replacement. As a relval, there is no clean way to handle this. Setting forget'
)
wf.status = 'forget'
session.commit()
else:
wfi.sendLog(
'injector',
'the workflow was found in trouble with no replacement')
no_replacement.add(wf.name)
continue
else:
wfi.sendLog(
'injector',
'the workflow was found in trouble and has a replacement')
print wf.name, "has", len(familly), "familly members"
print wf.name, "has", len(true_familly), "true familly members"
##we cannot have more than one of them !!! pick the last one
if len(true_familly) > 1:
#sendEmail('multiple wf','please take a look at injector for %s'%wf.name)
sendLog('injector',
'Multiple wf in line, will take the last one for %s \n%s' %
(wf.name, ', '.join(fwl['RequestName']
for fwl in true_familly)),
level='critical')
for fwl in true_familly[-1:]:
member = fwl['RequestName']
new_wf = session.query(Workflow).filter(
Workflow.name == member).first()
if not new_wf:
sendLog('injector',
"putting %s as replacement of %s" % (member, wf.name))
status = 'away'
if fwl['RequestStatus'] in ['assignment-approved']:
status = 'considered'
new_wf = Workflow(name=member,
status=status,
wm_status=fwl['RequestStatus'])
wf.status = 'forget'
session.add(new_wf)
else:
if new_wf.status == 'forget': continue
sendLog(
'injector',
"getting %s as replacement of %s" % (new_wf.name, wf.name))
wf.status = 'forget'
for tr in session.query(TransferImp).filter(
TransferImp.workflow_id == wf.id).all():
## get all transfer working for the old workflow
existing = session.query(TransferImp).filter(
TransferImp.phedexid == tr.phedexid).filter(
TransferImp.workflow_id == new_wf.id).all()
tr.active = False ## disable the old one
if not existing:
## create the transfer object for the new dependency
tri = TransferImp(phedexid=tr.phedexid, workflow=new_wf)
session.add(tri)
session.commit()
## don't do that automatically
#wf.status = 'forget'
session.commit()
if no_replacement:
#sendEmail('workflow with no replacement','%s \n are dangling there'%( '\n'.join(no_replacement)))
sendLog('injector',
'workflow with no replacement\n%s \n are dangling there' %
('\n'.join(no_replacement)),
level='critical')
def assignor(url, specific=None, talk=True, options=None):
if userLock(): return
if duplicateLock(): return
if not componentInfo().check(): return
UC = unifiedConfiguration()
CI = campaignInfo()
#SI = siteInfo()
SI = global_SI()
#NLI = newLockInfo()
#if not NLI.free() and not options.go: return
LI = lockInfo()
if not LI.free() and not options.go: return
n_assigned = 0
n_stalled = 0
wfos = []
fetch_from = []
if specific or options.early:
fetch_from.extend(['considered', 'staging'])
if specific:
fetch_from.extend(['considered-tried'])
if options.early:
print "Option Early is on"
fetch_from.extend(['staged'])
if options.from_status:
fetch_from = options.from_status.split(',')
print "Overriding to read from", fetch_from
for status in fetch_from:
print "getting wf in", status
wfos.extend(
session.query(Workflow).filter(Workflow.status == status).all())
print len(wfos)
## in case of partial, go for fetching a list from json ?
#if options.partial and not specific:
# pass
dataset_endpoints = json.loads(
open('%s/dataset_endpoints.json' % monitor_dir).read())
aaa_mapping = json.loads(
open('%s/equalizor.json' % monitor_pub_dir).read())['mapping']
all_stuck = set()
all_stuck.update(
json.loads(open('%s/stuck_transfers.json' % monitor_pub_dir).read()))
all_stuck.update(getAllStuckDataset())
max_per_round = UC.get('max_per_round').get('assignor', None)
max_cpuh_block = UC.get('max_cpuh_block')
##order by priority instead of random
if options.early:
cache = sorted(getWorkflows(url, 'assignment-approved', details=True),
key=lambda r: r['RequestPriority'])
cache = [r['RequestName'] for r in cache]
def rank(wfn):
return cache.index(wfn) if wfn in cache else 0
wfos = sorted(wfos, key=lambda wfo: rank(wfo.name), reverse=True)
print "10 first", [wfo.name for wfo in wfos[:10]]
print "10 last", [wfo.name for wfo in wfos[-10:]]
else:
random.shuffle(wfos)
for wfo in wfos:
if options.limit and (n_stalled + n_assigned) > options.limit:
break
if max_per_round and (n_stalled + n_assigned) > max_per_round:
break
if specific:
if not any(map(lambda sp: sp in wfo.name, specific.split(','))):
continue
#if not specific in wfo.name: continue
print "\n\n"
wfh = workflowInfo(url, wfo.name)
if wfh.request['RequestStatus'] in [
'rejected', 'aborted', 'aborted-completed', 'aborted-archived',
'rejected-archived'
] and wfh.isRelval():
wfo.status = 'forget'
session.commit()
n_stalled += 1
continue
if options.priority and int(
wfh.request['RequestPriority']) < options.priority:
continue
options_text = ""
if options.early: options_text += ", early option is ON"
if options.partial:
options_text += ", partial option is ON"
options_text += ", good fraction is %.2f" % options.good_enough
wfh.sendLog('assignor',
"%s to be assigned%s" % (wfo.name, options_text))
## the site whitelist takes into account siteInfo, campaignInfo, memory and cores
(lheinput, primary, parent, secondary,
sites_allowed) = wfh.getSiteWhiteList()
output_tiers = list(
set([o.split('/')[-1] for o in wfh.request['OutputDatasets']]))
if not output_tiers:
n_stalled += 1
wfh.sendLog('assignor', 'There is no output at all')
sendLog('assignor',
'Workflow %s has no output at all' % (wfo.name),
level='critical')
continue
is_stuck = (all_stuck & primary)
if is_stuck:
wfh.sendLog('assignor',
"%s are stuck input" % (','.join(is_stuck)))
## check if by configuration we gave it a GO
no_go = False
if not wfh.go(log=True) and not options.go:
no_go = True
allowed_secondary = {}
assign_parameters = {}
check_secondary = (not wfh.isRelval())
for campaign in wfh.getCampaigns():
if campaign in CI.campaigns:
assign_parameters.update(CI.campaigns[campaign])
if campaign in CI.campaigns and 'secondaries' in CI.campaigns[
campaign]:
if CI.campaigns[campaign]['secondaries']:
allowed_secondary.update(
CI.campaigns[campaign]['secondaries'])
check_secondary = True
if campaign in CI.campaigns and 'banned_tier' in CI.campaigns[
campaign]:
banned_tier = list(
set(CI.campaigns[campaign]['banned_tier'])
& set(output_tiers))
if banned_tier:
no_go = True
wfh.sendLog(
'assignor', 'These data tiers %s are not allowed' %
(','.join(banned_tier)))
sendLog('assignor',
'These data tiers %s are not allowed' %
(','.join(banned_tier)),
level='critical')
if secondary and check_secondary:
if (set(secondary) & set(allowed_secondary.keys()) !=
set(secondary)):
wfh.sendLog(
'assignor', '%s is not an allowed secondary' %
(', '.join(set(secondary) -
set(allowed_secondary.keys()))))
sendLog(
'assignor',
'%s is not an allowed secondary' %
(', '.join(set(secondary) -
set(allowed_secondary.keys()))),
level='critical')
if not options.go:
no_go = True
## then get whether there is something more to be done by secondary
for sec in secondary:
if sec in allowed_secondary: # and 'parameters' in allowed_secondary[sec]:
assign_parameters.update(allowed_secondary[sec])
if no_go:
n_stalled += 1
## make a very loud noise if >100k priority stalled
continue
## check on current status for by-passed assignment
if wfh.request['RequestStatus'] != 'assignment-approved':
if not options.test:
wfh.sendLog('assignor',
"setting %s away and skipping" % wfo.name)
## the module picking up from away will do what is necessary of it
wfo.wm_status = wfh.request['RequestStatus']
wfo.status = 'away'
session.commit()
continue
else:
print wfo.name, wfh.request['RequestStatus']
## retrieve from the schema, dbs and reqMgr what should be the next version
version = wfh.getNextVersion()
if not version:
if options and options.ProcessingVersion:
version = options.ProcessingVersion
else:
wfh.sendLog('assignor', "cannot decide on version number")
n_stalled += 1
wfo.status = 'trouble'
session.commit()
continue
original_sites_allowed = copy.deepcopy(sites_allowed)
wfh.sendLog('assignor', "Site white list %s" % sorted(sites_allowed))
override_sec_location = CI.get(wfh.request['Campaign'],
'SecondaryLocation', [])
blocks = wfh.getBlockWhiteList()
rwl = wfh.getRunWhiteList()
if rwl:
## augment with run white list
for dataset in primary:
blocks = list(set(blocks +
getDatasetBlocks(dataset, runs=rwl)))
lwl = wfh.getLumiWhiteList()
if lwl:
## augment with lumi white list
for dataset in primary:
blocks = list(
set(blocks + getDatasetBlocks(dataset, lumis=lwl)))
wfh.sendLog('assignor', "Allowed %s" % sorted(sites_allowed))
secondary_locations = None
primary_aaa = options.primary_aaa
secondary_aaa = options.secondary_aaa
do_partial = False #options.good_enough if options.partial else 0
if 'Campaign' in wfh.request and wfh.request[
'Campaign'] in CI.campaigns:
assign_parameters.update(CI.campaigns[wfh.request['Campaign']])
if 'primary_AAA' in assign_parameters:
primary_aaa = primary_aaa or assign_parameters['primary_AAA']
if 'secondary_AAA' in assign_parameters:
secondary_aaa = secondary_aaa or assign_parameters['secondary_AAA']
if 'partial_copy' in assign_parameters:
## can this only work if there is a stuck input ? maybe not
## this is a number. 0 means no
print "Could do partial disk copy assignment"
if is_stuck or options.partial:
do_partial = assign_parameters['partial_copy']
wfh.sendLog(
'assignor',
"Overiding partial copy assignment to %.2f fraction" %
do_partial)
#sendEmail('stuck input to assignment','%s is stuck for assigning %s and going fractional'%(','.join( is_stuck), wfo.name))
do_partial = options.good_enough if options.partial else do_partial
for sec in list(secondary):
if override_sec_location:
print "We don't care where the secondary is"
print "Cannot pass for now"
#sendEmail("tempting to pass sec location check","but we cannot yet IMO")
#pass
presence = getDatasetPresence(url, sec)
print sec
print json.dumps(presence, indent=2)
one_secondary_locations = [
site for (site, (there, frac)) in presence.items()
if frac > 98.
]
if secondary_aaa:
if not one_secondary_locations:
sec_availability = getDatasetBlocksFraction(url, sec)
if sec_availability >= 1. and options.go:
## there is at least one copy of each block on disk. We should go ahead and let it go.
wfh.sendLog(
'assignor',
"The secondary %s is available %s times on disk, and usable"
% (sec, sec_availability))
else:
## not even a copy on disk anywhere !!!!
sites_allowed = [] ## will block the assignment
wfh.sendLog(
'assignor',
"The secondary %s is nowhere on disk" % sec)
#just continue without checking
continue
#one_secondary_locations = [site for (site,(there,frac)) in presence.items() if there]
if secondary_locations == None:
secondary_locations = one_secondary_locations
else:
secondary_locations = list(
set(secondary_locations) & set(one_secondary_locations))
## reduce the site white list to site with secondary only
#sites_allowed = [site for site in sites_allowed if any([osite.startswith(site) for osite in one_secondary_locations])]
sites_allowed = [
site for site in sites_allowed
if SI.CE_to_SE(site) in one_secondary_locations
]
wfh.sendLog(
'assignor', "From/after secondary requirement, now Allowed%s" %
sorted(sites_allowed))
initial_sites_allowed = copy.deepcopy(
sites_allowed
) ## keep track of this, after secondary input location restriction : that's how you want to operate it
sites_all_data = copy.deepcopy(sites_allowed)
sites_with_data = copy.deepcopy(sites_allowed)
sites_with_any_data = copy.deepcopy(sites_allowed)
primary_locations = None
available_fractions = {}
set_lfn = '/store/mc' ## by default
endpoints = set()
for prim in list(primary):
if prim in dataset_endpoints:
print "endpoints from stagor", dataset_endpoints[prim]
endpoints.update(dataset_endpoints[prim])
set_lfn = getLFNbase(prim)
presence = getDatasetPresence(url, prim, only_blocks=blocks)
if talk:
print prim
print json.dumps(presence, indent=2)
available_fractions[prim] = getDatasetBlocksFraction(
url,
prim,
sites=[SI.CE_to_SE(site) for site in sites_allowed],
only_blocks=blocks)
if primary_aaa:
available_fractions[prim] = getDatasetBlocksFraction(
url, prim, only_blocks=blocks)
sites_all_data = [
site for site in sites_with_data if SI.CE_to_SE(site) in [
psite for (psite, (there, frac)) in presence.items()
if there
]
]
if primary_aaa:
sites_all_data = list(
set([
SI.SE_to_CE(psite)
for (psite, (there, frac)) in presence.items() if there
]))
sites_with_data = [
site for site in sites_with_data if SI.CE_to_SE(site) in
[psite for (psite, frac) in presence.items() if frac[1] > 90.]
]
sites_with_any_data = [
site for site in sites_with_any_data
if SI.CE_to_SE(site) in presence.keys()
]
if primary_aaa:
sites_with_any_data = list(
set([SI.SE_to_CE(psite) for psite in presence.keys()]))
wfh.sendLog(
'assignor', "Holding the data but not allowed %s" % sorted(
list(
set([
se_site for se_site in presence.keys()
if not SI.SE_to_CE(se_site) in sites_allowed
]))))
if primary_locations == None:
primary_locations = presence.keys()
else:
primary_locations = list(
set(primary_locations) & set(presence.keys()))
sites_with_data = list(set(sites_with_data))
sites_with_any_data = list(set(sites_with_any_data))
opportunistic_sites = []
down_time = False
## opportunistic running where any piece of data is available
if secondary_locations or primary_locations:
## intersection of both any pieces of the primary and good IO
#opportunistic_sites = [SI.SE_to_CE(site) for site in list((set(secondary_locations) & set(primary_locations) & set(SI.sites_with_goodIO)) - set(sites_allowed))]
if secondary_locations and primary_locations:
opportunistic_sites = [
SI.SE_to_CE(site) for site in
list((set(secondary_locations) & set(primary_locations)) -
set([SI.CE_to_SE(site) for site in sites_allowed]))
]
elif primary_locations:
opportunistic_sites = [
SI.SE_to_CE(site) for site in list(
set(primary_locations) -
set([SI.CE_to_SE(site) for site in sites_allowed]))
]
else:
opportunistic_sites = []
wfh.sendLog(
'assignor', "We could be running in addition at %s" %
sorted(opportunistic_sites))
if any(
[osite in SI.sites_not_ready
for osite in opportunistic_sites]):
wfh.sendLog(
'assignor', "One of the usable site is in downtime %s" % ([
osite for osite in opportunistic_sites
if osite in SI.sites_not_ready
]))
down_time = True
## should this be send back to considered ?
## should be 2 but for the time-being let's lower it to get things going
copies_wanted, cpuh = wfh.getNCopies()
wfh.sendLog('assignor', "we need %s CPUh" % cpuh)
if cpuh > max_cpuh_block and not options.go:
#sendEmail('large workflow','that wf %s has a large number of CPUh %s, not assigning, please check the logs'%(wfo.name, cpuh))#,destination=['*****@*****.**'])
sendLog(
'assignor',
'%s requires a large numbr of CPUh %s , not assigning, please check with requester'
% (wfo.name, cpuh),
level='critical')
wfh.sendLog(
'assignor',
"Requiring a large number of CPUh %s, not assigning" % cpuh)
continue
if 'Campaign' in wfh.request and wfh.request[
'Campaign'] in CI.campaigns and 'maxcopies' in CI.campaigns[
wfh.request['Campaign']]:
copies_needed_from_campaign = CI.campaigns[
wfh.request['Campaign']]['maxcopies']
copies_wanted = min(copies_needed_from_campaign, copies_wanted)
if not options.early:
less_copies_than_requested = UC.get("less_copies_than_requested")
copies_wanted = max(
1, copies_wanted -
less_copies_than_requested) # take one out for the efficiency
else:
## find out whether there is a site in the whitelist, that is lacking jobs and reduce to 1 copy needed to get things going
pass
wfh.sendLog('assignor',
"needed availability fraction %s" % copies_wanted)
## should also check on number of sources, if large enough, we should be able to overflow most, efficiently
## default back to white list to original white list with any data
wfh.sendLog('assignor', "Allowed sites :%s" % sorted(sites_allowed))
if primary_aaa:
## remove the sites not reachable localy if not in having the data
if not sites_all_data:
wfh.sendLog('assignor',
"Overiding the primary on AAA setting to Off")
primary_aaa = False
else:
aaa_grid = set(sites_all_data)
for site in list(aaa_grid):
aaa_grid.update(aaa_mapping.get(site, []))
sites_allowed = list(set(initial_sites_allowed) & aaa_grid)
wfh.sendLog(
'assignor', "Selected to read primary through xrootd %s" %
sorted(sites_allowed))
isStoreResults = ('StoreResults' == wfh.request.setdefault(
'RequestType', None))
if isStoreResults:
if 'MergedLFNBase' in wfh.request:
set_lfn = wfh.request['MergedLFNBase']
else:
n_stalled += 1
wfh.sendLog(
'assignor',
"Cannot assign StoreResults request because MergedLFN is missing"
)
sendLog(
'assignor',
'Cannot assign StoreResults request because MergedLFN is missing',
level='critical')
continue
if not primary_aaa:
if not isStoreResults:
sites_allowed = sites_with_any_data
else:
## if we are dealing with a StoreResults request, we don't need to check dataset availability and
## should use the SiteWhiteList set in the original request
if 'SiteWhitelist' in wfh.request:
sites_allowed = wfh.request['SiteWhitelist']
else:
wfh.sendLog(
'assignor',
"Cannot assign StoreResults request because SiteWhitelist is missing"
)
sendLog(
'assignor',
'Cannot assign StoreResults request because SiteWhitelist is missing',
level='critical')
n_stalled += 1
continue
available_fractions = {}
wfh.sendLog('assignor',
"Selected for any data %s" % sorted(sites_allowed))
### check on endpoints for on-going transfers
if do_partial:
if endpoints:
end_sites = [SI.SE_to_CE(s) for s in endpoints]
sites_allowed = list(set(sites_allowed + end_sites))
if down_time and not any(osite in SI.sites_not_ready
for osite in end_sites):
print "Flip the status of downtime, since our destinations are good"
down_time = False
print "with added endpoints", sorted(end_sites)
else:
print "Cannot do partial assignment without knowin the endpoints"
n_stalled += 1
continue
#if not len(sites_allowed):
# if not options.early:
# wfh.sendLog('assignor',"cannot be assign with no matched sites")
# sendLog('assignor','%s has no whitelist'% wfo.name, level='critical')
# n_stalled+=1
# continue
low_pressure = SI.sites_low_pressure(0.4)
## if any of the site allowed is low pressure : reduce to 1 copy so that it gets started
allowed_and_low = sorted(set(low_pressure) & set(sites_allowed))
if allowed_and_low:
wfh.sendLog(
'assignor',
"The workflow can run at %s under low pressure currently" %
(','.join(allowed_and_low)))
copies_wanted = max(1., copies_wanted - 1.)
if available_fractions and not all([
available >= copies_wanted
for available in available_fractions.values()
]):
not_even_once = not all([
available >= 1. for available in available_fractions.values()
])
above_good = all([
available >= do_partial
for available in available_fractions.values()
])
wfh.sendLog(
'assignor',
"The input dataset is not available %s times, only %s" %
(copies_wanted, available_fractions.values()))
if down_time and not options.go and not options.early:
wfo.status = 'considered'
session.commit()
wfh.sendLog(
'assignor',
"sending back to considered because of site downtime, instead of waiting"
)
#sendEmail( "cannot be assigned due to downtime","%s is not sufficiently available, due to down time of a site in the whitelist. check the assignor logs. sending back to considered."% wfo.name)
sendLog(
'assignor',
'%s is not sufficiently available, due to down time of a site in the whitelist. sending back to considered.'
% (wfo.name),
level='delay')
n_stalled += 1
continue
#pass
print json.dumps(available_fractions)
if (options.go and not_even_once) or not options.go:
known = []
try:
known = json.loads(open('cannot_assign.json').read())
except:
pass
if not wfo.name in known and not options.limit and not options.go and not options.early and not (
do_partial and above_good):
wfh.sendLog(
'assignor',
"cannot be assigned, %s is not sufficiently available.\n %s"
% (wfo.name, json.dumps(available_fractions)))
#sendEmail( "cannot be assigned","%s is not sufficiently available.\n %s"%(wfo.name,json.dumps(available_fractions)))
known.append(wfo.name)
open('cannot_assign.json',
'w').write(json.dumps(known, indent=2))
if options.early:
if wfo.status == 'considered':
wfh.sendLog('assignor', "setting considered-tried")
wfo.status = 'considered-tried'
session.commit()
else:
print "tried but status is", wfo.status
if do_partial and above_good:
print "Will move on with partial locations"
else:
n_stalled += 1
continue
if not len(sites_allowed):
if not options.early:
wfh.sendLog('assignor',
"cannot be assign with no matched sites")
sendLog('assignor',
'%s has no whitelist' % wfo.name,
level='critical')
n_stalled += 1
continue
t1_only = [ce for ce in sites_allowed if ce.startswith('T1')]
if t1_only:
# try to pick from T1 only first
sites_out = [SI.pick_dSE([SI.CE_to_SE(ce) for ce in t1_only])]
else:
# then pick any otherwise
sites_out = [
SI.pick_dSE([SI.CE_to_SE(ce) for ce in sites_allowed])
]
wfh.sendLog('assignor', "Placing the output on %s" % sites_out)
parameters = {
'SiteWhitelist': sites_allowed,
'NonCustodialSites': sites_out,
'AutoApproveSubscriptionSites': list(set(sites_out)),
'AcquisitionEra': wfh.acquisitionEra(),
'ProcessingString': wfh.processingString(),
'MergedLFNBase': set_lfn,
'ProcessingVersion': version,
}
if primary_aaa:
parameters['TrustSitelists'] = True
wfh.sendLog(
'assignor',
"Reading primary through xrootd at %s" % sorted(sites_allowed))
if secondary_aaa:
parameters['TrustPUSitelists'] = True
wfh.sendLog(
'assignor', "Reading secondary through xrootd at %s" %
sorted(sites_allowed))
## plain assignment here
team = 'production'
if os.getenv('UNIFIED_TEAM'): team = os.getenv('UNIFIED_TEAM')
if options and options.team:
team = options.team
parameters['Team'] = team
if lheinput:
## throttle reading LHE article
wfh.sendLog('assignor',
'Setting the number of events per job to 500k max')
parameters['EventsPerJob'] = 500000
def pick_options(options, parameters):
##parse options entered in command line if any
if options:
for key in reqMgrClient.assignWorkflow.keys:
v = getattr(options, key)
if v != None:
if type(v) == str and ',' in v:
parameters[key] = filter(None, v.split(','))
else:
parameters[key] = v
def pick_campaign(assign_parameters, parameters):
## pick up campaign specific assignment parameters
parameters.update(assign_parameters.get('parameters', {}))
if options.force_options:
pick_campaign(assign_parameters, parameters)
pick_options(options, parameters)
else:
## campaign parameters update last
pick_options(options, parameters)
pick_campaign(assign_parameters, parameters)
if not options.test:
parameters['execute'] = True
hold_split, split_check = wfh.checkSplitting()
if hold_split and not options.go:
if split_check:
wfh.sendLog(
'assignor', 'Holding on to the change in splitting %s' %
('\n\n'.join([str(i) for i in split_check])))
else:
wfh.sendLog('assignor', 'Change of splitting is on hold')
n_stalled += 1
continue
if split_check == None or split_check == False:
n_stalled += 1
continue
elif split_check:
## operate all recommended changes
reqMgrClient.setWorkflowSplitting(url, wfo.name, split_check)
wfh.sendLog(
'assignor', 'Applying the change in splitting %s' %
('\n\n'.join([str(i) for i in split_check])))
split_check = True ## bypass completely and use the above
# Handle run-dependent MC
pstring = wfh.processingString()
if 'PU_RD' in pstring:
numEvents = wfh.getRequestNumEvents()
eventsPerLumi = [getDatasetEventsPerLumi(prim) for prim in primary]
eventsPerLumi = sum(eventsPerLumi) / float(len(eventsPerLumi))
reqJobs = 500
if 'PU_RD2' in pstring:
reqJobs = 2000
eventsPerJob = int(numEvents / (reqJobs * 1.4))
lumisPerJob = int(eventsPerJob / eventsPerLumi)
if lumisPerJob == 0:
#sendEmail("issue with event splitting for run-dependent MC","%s needs to be split by event with %s per job"%(wfo.name, eventsPerJob))
sendLog('assignor',
"%s needs to be split by event with %s per job" %
(wfo.name, eventsPerJob),
level='critical')
wfh.sendLog(
'assignor',
"%s needs to be split by event with %s per job" %
(wfo.name, eventsPerJob))
parameters['EventsPerJob'] = eventsPerJob
else:
spl = wfh.getSplittings()[0]
eventsPerJobEstimated = spl[
'events_per_job'] if 'events_per_job' in spl else None
eventsPerJobEstimated = spl[
'avg_events_per_job'] if 'avg_events_per_job' in spl else None
if eventsPerJobEstimated and eventsPerJobEstimated > eventsPerJob:
#sendEmail("setting lumi splitting for run-dependent MC","%s was assigned with %s lumis/job"%( wfo.name, lumisPerJob))
sendLog('assignor',
"%s was assigned with %s lumis/job" %
(wfo.name, lumisPerJob),
level='critical')
wfh.sendLog(
'assignor', "%s was assigned with %s lumis/job" %
(wfo.name, lumisPerJob))
parameters['LumisPerJob'] = lumisPerJob
else:
#sendEmail("leaving splitting untouched for PU_RD*","please check on "+wfo.name)
sendLog(
'assignor',
"leaving splitting untouched for %s, please check on %s"
% (pstring, wfo.name),
level='critical')
wfh.sendLog(
'assignor',
"leaving splitting untouched for PU_RD*, please check."
)
if isHEPCloudReady(url) and wfh.isGoodForNERSC():
parameters['Team'] = 'hepcloud'
parameters['SiteWhitelist'] = ['T3_US_NERSC']
if primary:
parameters['TrustSitelists'] = True
if secondary:
parameters['TrustPUSitelists'] = True
sendEmail("sending work to hepcloud",
"pleasse check on %s" % wfh.request['RequestName'],
destination=['*****@*****.**'])
## make sure to autoapprove all NonCustodialSites
parameters['AutoApproveSubscriptionSites'] = list(
set(parameters['NonCustodialSites'] +
parameters.get('AutoApproveSubscriptionSites', [])))
result = reqMgrClient.assignWorkflow(
url, wfo.name, None,
parameters) ## team is not relevant anymore here
# set status
if not options.test:
if result:
wfo.status = 'away'
session.commit()
n_assigned += 1
wfh.sendLog(
'assignor', "Properly assigned\n%s" %
(json.dumps(parameters, indent=2)))
try:
## refetch information and lock output
new_wfi = workflowInfo(url, wfo.name)
(_, prim, _, sec) = new_wfi.getIO()
for secure in list(prim) + list(
sec) + new_wfi.request['OutputDatasets']:
## lock all outputs
LI.lock(secure, reason='assigning')
except Exception as e:
print "fail in locking output"
print str(e)
sendEmail("failed locking of output", str(e))
else:
wfh.sendLog(
'assignor',
"Failed to assign %s.\n%s \n Please check the logs" %
(wfo.name, reqMgrClient.assignWorkflow.errorMessage))
sendLog('assignor',
"Failed to assign %s.\n%s \n Please check the logs" %
(wfo.name, reqMgrClient.assignWorkflow.errorMessage),
level='critical')
print "ERROR could not assign", wfo.name
else:
pass
print "Assignment summary:"
sendLog('assignor', "Assigned %d Stalled %s" % (n_assigned, n_stalled))
if n_stalled and not options.go and not options.early:
sendLog('assignor',
"%s workflows cannot be assigned. Please take a look" %
(n_stalled),
level='critical')
def singleRecovery(url, task, initial, actions, do=False):
print "Inside single recovery!"
payload = {
"Requestor": os.getenv('USER'),
"Group": 'DATAOPS',
"RequestType": "Resubmission",
"ACDCServer": initial['ConfigCacheUrl'],
"ACDCDatabase": "acdcserver",
"OriginalRequestName": initial['RequestName'],
"OpenRunningTimeout": 0
}
copy_over = [
'PrepID', 'Campaign', 'RequestPriority', 'TimePerEvent',
'SizePerEvent', 'Group', 'Memory', 'RequestString', 'CMSSWVersion'
]
for c in copy_over:
if c in initial:
payload[c] = copy.deepcopy(initial[c])
else:
print c, "not in the initial payload"
#a massage ? boost the recovery over the initial wf
# payload['RequestPriority'] *= 10
#Max priority is 1M
payload['RequestPriority'] = min(500000, payload['RequestPriority'] *
2) ## never above 500k
#change parameters based on actions here
if actions:
for action in actions:
if action.startswith('mem') and actions[action] != "" and actions[
action] != 'Same':
#if multicore parameter is also used, need to scale memory by the new number of cores
if 'multicore' in actions and actions['multicore'] != "":
continue
## Taskchains needs to be treated special to set the memory to all tasks
set_to = int(actions[action])
if 'TaskChain' in initial:
mem_dict = {}
it = 1
while True:
t = 'Task%d' % it
it += 1
if t in initial:
tname = payload.setdefault(t,
initial[t])['TaskName']
mem = mem_dict.setdefault(tname,
payload[t]['Memory'])
mem_dict[tname] = set_to
else:
break
payload['Memory'] = mem_dict
print "Memory set to: ", json.dumps(mem_dict, indent=2)
else:
payload['Memory'] = set_to
print "Memory set to: ", set_to
if action.startswith('multicore') and actions[action] != "":
set_to = int(actions[action])
## Taskchains needs to be treated special to set the multicore and memory values to all tasks
if 'TaskChain' in initial:
mem_dict = payload['Memory'] if type(
payload['Memory']) == dict else {}
core_dict = {}
it = 1
while True:
t = 'Task%d' % it
it += 1
if t in initial:
tname = payload.setdefault(t,
initial[t])['TaskName']
mem = mem_dict.setdefault(tname,
payload[t]['Memory'])
#Need to scale the memory by the new number of cores
initial_cores = payload[t].setdefault(
'Multicore', 1)
if 'memory' in actions and actions[
'memory'] != "" and actions[
'memory'] != 'Same':
mem = actions['memory']
fraction_constant = 0.4
mem_per_core_c = int((1 - fraction_constant) *
mem / float(initial_cores))
mem_dict[tname] = int(mem +
(set_to - initial_cores) *
mem_per_core_c)
core_dict[tname] = set_to
print "For ", t
print "Multicore set to ", set_to
print "Memory set to ", mem_dict[tname]
else:
break
payload['Memory'] = mem_dict
payload['Multicore'] = core_dict
else:
#Need to scale the memory by the new number of cores
initial_cores = initial.setdefault('Multicore', 1)
mem = payload['Memory']
if 'memory' in actions and actions[
'memory'] != "" and actions['memory'] != 'Same':
mem = actions['memory']
fraction_constant = 0.4
mem_per_core_c = int(
(1 - fraction_constant) * mem / float(initial_cores))
payload['Multicore'] = set_to
payload['Memory'] = int(mem + (set_to - initial_cores) *
mem_per_core_c)
print "Multicore set to ", set_to
print "Memory set to ", payload['Memory']
if action.startswith('split'):
split_alert = (initial['RequestType'] in ['MonteCarlo'])
for key in initial:
if key == 'SplittingAlgo' and (initial[key]
in ['EventBased']):
split_alert = True
elif key.startswith('Task') and key != 'TaskChain':
for key2 in initial[key]:
if key2 == 'TaskName':
this_taskname = initial[key][key2]
recover_task = task.split('/')[-1]
print "For recovery of task", recover_task
print "Looking at task", this_taskname
if (recover_task == this_taskname) and (
initial[key]['SplittingAlgo']
in ['EventBased']):
## the task to be recovered is actually of the wrong type to allow change of splitting
sendLog(
'actor',
'To recover on %s, changing the splitting on %s is not really allowed and this will be ignored instead of failing acdc.'
%
(task, initial[key]['SplittingAlgo']),
level='critical')
## do not send an alert and stop the acdc
#split_alert = True
if split_alert:
sendLog('actor',
'Cannot change splitting for %s' %
initial['RequestName'],
level='critical')
print "I should not be doing splitting for this type of request", initial[
'RequestName']
return None
acdc_round = 0
initial_string = payload['RequestString']
if initial_string.startswith('ACDC'):
if initial_string[4].isdigit():
acdc_round = int(initial_string[4])
acdc_round += 1
initial_string = initial_string.replace('ACDC_', '').replace(
'ACDC%d_' % (acdc_round - 1), '')
payload['RequestString'] = 'ACDC%d_%s' % (acdc_round, initial_string)
payload['InitialTaskPath'] = task
if not do:
print json.dumps(payload, indent=2)
return None
print "ACDC payload"
# print json.dumps( payload , indent=2)
print actions
## submit here
acdc = reqMgrClient.submitWorkflow(url, payload)
if not acdc:
print "Error in making ACDC for", initial["RequestName"]
acdc = reqMgrClient.submitWorkflow(url, payload)
if not acdc:
print "Error twice in making ACDC for", initial["RequestName"]
sendLog('actor',
'Failed twice in making ACDCs for %s!' %
initial['RequestName'],
level='critical')
return None
## change splitting if requested
if actions:
for action in actions:
if action.startswith('split'):
acdcInfo = workflowInfo(url, acdc)
splittings = acdcInfo.getSplittingsNew(strip=True)
if actions[action] != 'Same' and actions[action] != 'max':
factor = int(
actions[action][0:-1]) if 'x' in actions[action] else 2
for split in splittings:
split_par = split['splitParams']
if split['splitAlgo'] in ['EventBased']:
sendLog(
'actor',
"Changing the splitting on %s for %s is not permitted. Not changing."
% (split['splitAlgo'], initial["RequestName"]),
level='critical')
continue
for act in [
'avg_events_per_job', 'events_per_job',
'lumis_per_job'
]:
if act in split_par:
print "Changing %s (%d) by a factor %d" % (
act, split_par[act], factor),
split_par[act] /= factor
print "to", split_par[act]
break
#split['requestName'] = acdc
#print "changing the splitting of",acdc
#print json.dumps( split, indent=2 )
#print reqMgrClient.setWorkflowSplitting(url, acdc, split )
elif 'max' in actions[action]:
for split in splittings:
split_par = split['splitParams']
for act in [
'avg_events_per_job', 'events_per_job',
'lumis_per_job'
]:
if act in split_par:
print "Changing %s (%d) " % (act,
split_par[act]),
split_par[act] = 1
print "to max splitting ", split_par[act]
break
#split['requestName'] = acdc
#print "changing the splitting of",acdc
#print json.dumps( split, indent=2 )
#print reqMgrClient.setWorkflowSplitting(url, acdc, split )
print "changing the splitting of", acdc
print json.dumps(splittings, indent=2)
done = reqMgrClient.setWorkflowSplitting(url, acdc, splittings)
## check on done == True
data = reqMgrClient.setWorkflowApproved(url, acdc)
print data
return acdc
def batchor( url ):
UC = unifiedConfiguration()
SI = global_SI()
## get all workflows in assignment-approved with SubRequestType = relval
all_wfs = []
for user in UC.get("user_relval"):
all_wfs.extend( getWorkflows(url, 'assignment-approved', details=True, user=user, rtype='TaskChain') )
wfs = filter( lambda r :r['SubRequestType'] == 'RelVal' if 'SubRequestType' in r else False, all_wfs)
## need a special treatment for those
hi_wfs = filter( lambda r :r['SubRequestType'] == 'HIRelVal' if 'SubRequestType' in r else False, all_wfs)
by_campaign = defaultdict(set)
by_hi_campaign = defaultdict(set)
for wf in wfs:
print "Relval:",wf['RequestName'], wf['Campaign']
#by_campaign[wf['Campaign']].add( wf['RequestName'] )
by_campaign[wf['Campaign']].add( wf['PrepID'] )
for wf in hi_wfs:
print "HI Relval:",wf['RequestName'], wf['Campaign']
#by_hi_campaign[wf['Campaign']].add( wf['RequestName'] )
by_hi_campaign[wf['Campaign']].add( wf['PrepID'] )
default_setup = {
"go" :True,
"parameters" : {
"SiteWhitelist": [ "T1_US_FNAL" ],
"MergedLFNBase": "/store/relval",
"Team" : "relval",
"NonCustodialGroup" : "RelVal"
},
"custodial" : "T1_US_FNAL_MSS",
"custodial_override" : ["DQMIO"],
"phedex_group" : "RelVal",
"lumisize" : -1,
"fractionpass" : 0.0,
"maxcopies" : 1
}
default_hi_setup = copy.deepcopy( default_setup )
add_on = {}
batches = json.loads( open('batches.json').read() )
relval_routing = UC.get('relval_routing')
def pick_one_site( p):
## modify the parameters on the spot to have only one site
if "parameters" in p and "SiteWhitelist" in p["parameters"] and len(p["parameters"]["SiteWhitelist"])>1:
choose_from = list(set(p["parameters"]["SiteWhitelist"]) & set(SI.sites_ready))
picked = random.choice( choose_from )
print "picked",picked,"from",choose_from
p["parameters"]["SiteWhitelist"] = [picked]
for campaign in by_campaign:
if campaign in batches: continue
## get a bunch of information
setup = copy.deepcopy( default_setup )
for key in relval_routing:
if key in campaign:
## augment with the routing information
augment_with = relval_routing[key]
print "Modifying the batch configuration because of keyword",key
print "with",augment_with
setup = deep_update( setup, augment_with )
#if 'cc7' in campaign: setup["parameters"]["SiteWhitelist"] = ["T2_US_Nebraska"]
pick_one_site( setup )
add_on[campaign] = setup
sendLog('batchor','Adding the relval campaigns %s with parameters \n%s'%( campaign, json.dumps( setup, indent=2)),level='critical')
if not campaign in batches: batches[campaign] = []
batches[campaign] = list(set(list(copy.deepcopy( by_campaign[campaign] )) + batches[campaign] ))
for campaign in by_hi_campaign:
if campaign in batches: continue
## get a bunch of information
setup = copy.deepcopy( default_hi_setup )
hi_site = random.choice(["T1_DE_KIT","T1_FR_CCIN2P3"])
setup["parameters"]["SiteWhitelist"]=[ hi_site ]
#setup["parameters"]["SiteWhitelist"]=["T1_DE_KIT","T1_FR_CCIN2P3"]
pick_one_site( setup )
add_on[campaign] = setup
sendLog('batchor','Adding the HI relval campaigns %s with parameters \n%s'%( campaign, json.dumps( setup, indent=2)),level='critical')
if not campaign in batches: batches[campaign] = []
batches[campaign] = list(set(list(copy.deepcopy( by_hi_campaign[campaign] )) + batches[campaign] ))
open('batches.json','w').write( json.dumps( batches , indent=2 ) )
## open the campaign configuration
campaigns = json.loads( open('campaigns.relval.json').read() )
## protect for overwriting ??
for new_campaign in list(set(add_on.keys())-set(campaigns.keys())):
## this is new, and can be announced as such
print new_campaign,"is new stuff"
subject = "Request of RelVal samples batch %s"% new_campaign
text="""Dear all,
A new batch of relval workflows was requested.
Batch ID:
%s
Details of the workflows:
https://dmytro.web.cern.ch/dmytro/cmsprodmon/requests.php?campaign=%s
This is an automated message"""%( new_campaign,
new_campaign,
)
print subject
print text
to = ['*****@*****.**']
sendEmail(subject, text, destination=to)
sendLog('batchor',text, level='critical')
## go through all existing campaigns and remove the ones not in use anymore ?
for old_campaign in campaigns.keys():
all_in_batch = getWorkflowByCampaign(url, old_campaign, details=True)
is_batch_done = all(map(lambda s : not s in ['completed','running-open','running-closed','acquired','assigned','assignment-approved'], [wf['RequestStatus']for wf in all_in_batch]))
## check all statuses
if is_batch_done:
#print "batch",old_campaign,"can be closed or removed if necessary"
#campaigns[old_campaign]['go'] = False ## disable
campaigns.pop( old_campaign ) ## or just drop it all together ?
print "batch",old_campaign," configuration was removed"
## merge all anyways
campaigns.update( add_on )
## write it out for posterity
open('campaigns.json.updated','w').write(json.dumps( campaigns , indent=2))
## read back
rread = json.loads(open('campaigns.json.updated').read())
os.system('mv campaigns.json.updated campaigns.relval.json')
def actor(url, options=None):
mlock = moduleLock(wait=False, silent=True)
if mlock(): return
if userLock('actor'): return
up = componentInfo(soft=['mcm'])
if not up.check(): return
# CI = campaignInfo()
SI = siteInfo()
UC = unifiedConfiguration()
WC = wtcClient()
WI = wtcInfo()
JC = JIRAClient()
action_list = WC.get_actions()
if action_list is None:
print "Not able to load action list"
sendLog('actor', 'Not able to load action list', level='critical')
return
if options.actions:
action_list = json.loads(open(options.actions).read())
print json.dumps(action_list, indent=2)
if not action_list:
print "EMPTY!"
return
wf_list = action_list.keys()
print json.dumps(sorted(wf_list), indent=2)
if options.spec:
wf_list = [wf for wf in wf_list if options.spec in wf]
max_per_round = UC.get('max_per_round').get('actor', None)
if max_per_round:
random.shuffle(wf_list)
wf_list = wf_list[:max_per_round]
for wfname in wf_list:
print '-' * 100
print "Looking at", wfname, "for recovery options"
to_clone = False
to_acdc = False
to_force = False
to_hold = False
something_to_do = False
tasks = action_list[wfname].get('Parameters', None)
to_acdc = action_list[wfname].get('Action', None) == 'acdc'
to_clone = action_list[wfname].get('Action', None) == 'clone'
to_force = action_list[wfname].get(
'Action', None) == 'special' and action_list[wfname].get(
'Parameters', {}).get('action', None) in ['by-pass', 'bypass']
to_hold = action_list[wfname].get(
'Action', None) == 'special' and action_list[wfname].get(
'Parameters', {}).get('action', None) in ['onhold', 'on-hold']
if not to_acdc and not to_clone and not to_force and not to_hold:
sendLog(
'actor',
'Action submitted for something other than acdc, clone, bypass or hold for workflow %s'
% wfname,
level='critical')
print json.dumps(action_list[wfname], indent=2)
continue
if not tasks and to_acdc:
sendLog('actor',
'Empty action submitted for workflow %s' % wfname,
level='critical')
print "Moving on. Parameters is blank for " + wfname
continue
wfi = workflowInfo(url, wfname)
recover = True
message_to_ops = ""
message_to_user = ""
#===========================================================
if to_clone and options.do:
print "Let's try kill and clone: "
wfi.sendLog('actor', 'Going to clone %s' % wfname)
comment = ""
if 'comment' in tasks: comment = ", reason: " + tasks['comment']
wfi.sendLog(
'actor',
"invalidating the workflow by traffic controller %s" % comment)
#Reject all workflows in the family
inv_results = invalidate(url,
wfi,
only_resub=False,
with_output=True)
all_good = all(inv_results)
if all_good:
wfi.sendLog('actor', "%s and children are rejected" % wfname)
else:
wfi.sendLog('actor',
"Failed to reject the request and dependents")
sendLog('actor',
'Failed to reject the familly of %s' % wfname,
level='critical')
continue
cloned = None
try:
cloned = singleClone(url, wfname, tasks, comment, options.do)
except Exception as e:
sendLog(
'actor',
'Failed to create clone for %s! Check logs for more information. Action will need to be resubmitted.'
% wfname,
level='critical')
wfi.sendLog('actor', 'Failed to create clone for %s!' % wfname)
print str(e)
##let's not remove the action other the workflow goes to "trouble" and the WTC cannot set the action again
if not cloned:
recover = False
wfi.sendLog('actor', 'Failed to create clone for %s!' % wfname)
sendLog('actor',
'Failed to create clone for %s!' % wfname,
level='critical')
else:
wfi.sendLog('actor',
"Workflow %s cloned into %s" % (wfname, cloned))
## set to trouble for swift replacement
for wfo in session.query(Workflow).filter(
Workflow.name == wfname).all():
wfo.status = 'trouble'
session.commit()
#===========================================================
elif to_force:
wfi.sendLog(
'actor',
'Force-completing from workflow traffic controler request')
WI.add(action='force',
keyword=wfname,
user=action_list[wfname].get('user', 'unified'))
elif to_hold:
wfi.sendLog('actor',
'Holding on workflow traffic controler request')
WI.add(action='hold',
keyword=wfname,
user=action_list[wfname].get('user', 'unified'))
#===========================================================
elif to_acdc:
if 'AllSteps' in tasks:
allTasksDefaults = tasks['AllSteps']
tasks.pop('AllSteps')
for setting in allTasksDefaults:
for task in tasks:
if setting in tasks[task]:
tasks[task][setting] = allTasksDefaults[setting]
else:
tasks[task].append(
{setting: allTasksDefaults[setting]})
print "Tasks is "
print json.dumps(tasks, indent=2)
all_tasks = wfi.getAllTasks()
## need a way to verify that this is the first round of ACDC, since the second round will have to be on the ACDC themselves
try:
WMErr = wfi.getWMErrors()
# print WMErr
except:
sendLog(
'actor',
'Cannot create ACDCS for %s because WMErr cannot be reached.'
% wfname,
level='critical')
continue
if not WMErr:
wfi.sendLog('actor', 'WMErrors is blank for %s.' % wfname)
print "FYI getWMErrors is blank. Presumably there are only unreported errors"
# continue
try:
where_to_run, missing_to_run, missing_to_run_at = wfi.getRecoveryInfo(
)
print "Where to run = "
print where_to_run
if not where_to_run:
sendLog(
'actor',
'Cannot create ACDCS for %s because recovery info cannot be found.'
% wfname,
level='critical')
continue
except:
sendLog(
'actor',
'Cannot create ACDCS for %s because recovery info cannot be found.'
% wfname,
level='critical')
print "Moving on. Cannot access recovery info for " + wfname
continue
if not where_to_run:
sendLog(
'actor',
'Cannot create ACDCS for %s because site list cannot be found.'
% wfname,
level='critical')
print "Moving on. where to run is blank"
continue
message_to_ops = ""
message_to_user = ""
num_tasks_to_recover = 0
if WMErr:
for task in WMErr:
if 'LogCollect' in task: continue
if 'Cleanup' in task: continue
if not 'jobfailed' in WMErr[task]:
continue
else:
num_tasks_to_recover += 1
# print "Task to recover: " + task
if not num_tasks_to_recover:
print "\tno error for", wfname
# recover = False
if 'LheInputFiles' in wfi.request and wfi.request['LheInputFiles']:
## we do not try to recover pLHE
sendLog(
'actor',
'Cannot create ACDCS for %s because it is a pLHE workflow.'
% wfname,
level='critical')
print "We don't try to recover pLHE. Moving on."
recover = False
# sendEmail('cannot submit action', '%s is a pLHE workflow. We do not try to recover pLHE'%wfname)
# if wfi.request['RequestType'] in ['ReReco']:
# recover= False
# print 'cannot submit action. ReReco'
# sendEmail('cannot submit action', '%s is request type ReReco'%wfname)
recovering = set()
for task in tasks:
assign_to_sites = set()
print "Task names is " + task
fulltaskname = '/' + wfname + '/' + task
print "Full task name is " + fulltaskname
print where_to_run.keys()
wrong_task = False
for task_info in all_tasks:
if fulltaskname == task_info.pathName:
if task_info.taskType not in [
'Processing', 'Production', 'Merge'
]:
wrong_task = True
wfi.sendLog(
'actor',
"Skipping task %s because the taskType is %s. Can only ACDC Processing, Production, or Merge tasks"
% (fulltaskname, task_info.taskType))
if not fulltaskname in where_to_run.keys():
wrong_task = True
wfi.sendLog(
'actor',
"Skipping task %s because there is no acdc doc for it anyways."
% (fulltaskname))
if wrong_task:
continue
print tasks[task]
actions = tasks[task]
for action in actions:
if action.startswith('sites'):
if type(actions[action]) != list:
assign_to_sites = [SI.SE_to_CE(actions[action])]
else:
assign_to_sites = list(
set([
SI.SE_to_CE(site)
for site in actions[action]
]))
# if action.startswith('mem') and actions[action] != "" and actions[action] != 'Same' and wfi.request['RequestType'] in ['TaskChain']:
# recover = False;
# print "Skipping %s for now until Allie fixes memory parameter for TaskChain ACDCs."%wfname
# wfi.sendLog('actor',"Skipping %s for now until Allie fixes memory parameter for TaskChain ACDCs."%wfname)
if not 'sites' in actions:
assign_to_sites = list(
set([
SI.SE_to_CE(site)
for site in where_to_run[fulltaskname]
]))
print "Found", sorted(
assign_to_sites
), "as sites where to run the ACDC at, from the acdc doc of ", wfname
print "Going to run at", sorted(assign_to_sites)
if recover:
print "Initiating recovery"
acdc = singleRecovery(url,
fulltaskname,
wfi.request,
actions,
do=options.do)
if not acdc:
if options.do:
if recovering:
print wfname + " has been partially ACDC'ed. Needs manual attention."
sendLog(
'actor',
"%s has had %s/%s recoveries %s only" %
(wfname, len(recovering),
num_tasks_to_recover, list(recovering)),
level='critical')
wfi.sendLog(
'actor',
"%s has had %s/%s recoveries %s only" %
(wfname, len(recovering),
num_tasks_to_recover, list(recovering)))
break
else:
print wfname + " failed recovery once"
recover = False
break
else:
print "no action to take further"
# sendLog('recoveror', "ACDC for %s can be done automatically"% wfname, level='critical')
continue
else: #ACDC was made correctly. Now we have to assign it.
wfi.sendLog(
'actor',
'ACDC created for task %s. Actions taken \n%s' %
(fulltaskname, json.dumps(actions)))
jira_comment = "%s created ACDC for task %s with action %s" % (
action_list[wfname].get('user', 'unified'),
task.split('/')[-1],
json.dumps(actions),
)
reason = action_list[wfname].get('Reason', None)
if reason:
jira_comment += '\ndue to: %s' % (reason)
#team = wfi.request['Teams'][0]
team = 'production'
parameters = {
'SiteWhitelist': sorted(assign_to_sites),
'AcquisitionEra': wfi.acquisitionEra(),
'ProcessingString': wfi.processingString(),
'MergedLFNBase': wfi.request['MergedLFNBase'],
'ProcessingVersion':
wfi.request['ProcessingVersion'],
}
## hackery for ACDC merge assignment
if wfi.request[
'RequestType'] == 'TaskChain' and 'Merge' in task.split(
'/')[-1]:
parameters['AcquisitionEra'] = None
parameters['ProcessingString'] = None
## xrootd setttings on primary and secondary
if 'xrootd' in actions:
if actions['xrootd'] == 'enabled':
print "Going to assign via xrootd"
parameters['TrustSitelists'] = True
elif actions['xrootd'] == 'disabled':
parameters['TrustSitelists'] = False
elif ('TrustSitelists' in wfi.request
and wfi.request['TrustSitelists'] == 'true'):
parameters['TrustSitelists'] = True
else:
parameters['TrustSitelists'] = False
if 'secondary' in actions:
if actions['secondary'] == 'enabled':
print 'Enabling reading the secondary input via xrootd'
parameters['TrustPUSitelists'] = True
elif actions['secondary'] == 'disabled':
parameters['TrustPUSitelists'] = False
#in case secondary is blank or not set to enabled or disabled
elif 'TrustPUSitelists' in wfi.request and wfi.request[
'TrustPUSitelists']:
parameters['TrustPUSitelists'] = True
elif 'TrustPUSitelists' in wfi.request and wfi.request[
'TrustPUSitelists']:
parameters['TrustPUSitelists'] = True
if options.ass:
print "really doing the assignment of the ACDC", acdc
parameters['execute'] = True
#wfi.sendLog('actor',"%s was assigned for recovery"% acdc)
else:
print "no assignment done with this ACDC", acdc
sendLog('actor',
"%s needs to be assigned" % (acdc),
level='critical')
wfi.sendLog(
'actor',
"%s needs to be assigned by hand" % (acdc))
continue
# print parameters
result = reqMgrClient.assignWorkflow(
url, acdc, team, parameters)
if not result:
print acdc, "was not assigned"
sendLog('actor',
"%s failed to be assigned" % (acdc),
level='critical')
wfi.sendLog(
'actor',
"%s failed to get assigned for recovery" %
acdc)
else:
wfi.sendLog('actor',
"%s was assigned for recovery" % acdc)
recovering.add(acdc)
#wfi.sendLog('actor',"ACDCs created for %s"%wfname)
try:
if jira_comment:
jiras = JC.find(
{'prepid': wfi.request['PrepID']})
if len(jiras) == 1:
## put a comment on the single corresponding ticket
JC.comment(jiras[0].key, jira_comment)
JC.progress(jiras[0].key)
except Exception as e:
print "failed with JIRA"
print str(e)
#===========================================================
if recover and options.do:
r = WC.remove_action(wfname)
if not r:
sendLog(
'actor',
'not able to remove the action, interlocking the module',
level='critical')
os.system('touch %s/actor.failed-%s.lock' %
(base_eos_dir, os.getpid()))
sys.exit(-1)
## update the status with recovering removing manual
for wfo in session.query(Workflow).filter(
Workflow.name == wfname).all():
wfo.status = wfo.status.replace('manual', 'recovering')
session.commit()
if message_to_user:
print wfname, "to be notified to user(DUMMY)", message_to_user
if message_to_ops:
print 'message'
#sendEmail( "notification in recoveror" , message_to_ops, destination=['*****@*****.**'])
# sendLog('recoveror',message_to_ops,level='warning')
return
def injector(url, options, specific):
use_mcm = True
up = componentInfo( mcm = use_mcm, soft=['mcm'] )
if not up.check(): return
use_mcm = up.status['mcm']
workflows = getWorkflows(url, status=options.wmstatus, user=options.user)
workflows.extend( getWorkflows(url, status=options.wmstatus, user='******', rtype="ReReco")) ## regardless of users, pick up all ReReco on the table
print len(workflows),"in line"
cannot_inject = set()
status_cache = defaultdict(str)
## browse for assignment-approved requests, browsed for ours, insert the diff
for wf in workflows:
if specific and not specific in wf: continue
exists = session.query(Workflow).filter(Workflow.name == wf ).first()
if not exists:
wfi = workflowInfo(url, wf)
#wl = getWorkLoad(url, wf)
## check first that there isn't related here with something valid
can_add = True
## first try at finding a match
# print wfi.request
familly = session.query(Workflow).filter(Workflow.name.contains(wfi.request['PrepID'])).all()
if not familly:
#req_familly = getWorkflowById( url, wl['PrepID'])
#familly = [session.query(Workflow).filter(Workflow.name == member).first() for member in req_familly]
pids = wfi.getPrepIDs()
req_familly = []
for pid in pids:
req_familly.extend( getWorkflowById( url, pid, details=True) )
familly = []
print len(req_familly),"members"
for req_member in req_familly:
#print "member",req_member['RequestName']
owfi = workflowInfo(url, req_member['RequestName'], request=req_member)
other_pids = owfi.getPrepIDs()
if set(pids) == set(other_pids):
## this is a real match
familly.extend( session.query(Workflow).filter(Workflow.name == req_member['RequestName']).all() )
for lwfo in familly:
if lwfo:
## we have it already
if not lwfo.status in ['forget','trouble','forget-unlock','forget-out-unlock']:
sendLog('injector',"Should not put %s because of %s %s"%( wf, lwfo.name,lwfo.status ))
print "Should not put",wf,"because of",lwfo.name,lwfo.status
cannot_inject.add( wf )
can_add = False
## add a check on validity of input datasets
_,prim,par,sec = wfi.getIO()
for d in list(prim)+list(par)+list(sec):
if not d in status_cache:
status_cache[d] = getDatasetStatus(d)
if status_cache[d] != 'VALID':
wfi.sendLog('injector',"One of the input is not VALID. %s : %s"%( d, status_cache[d]))
sendLog('injector',"One of the input of %s is not VALID. %s : %s"%( wf, d, status_cache[d]))
can_add = False
if not can_add: continue
wfi.sendLog('injector',"considering %s"%wf)
new_wf = Workflow( name = wf , status = options.setstatus, wm_status = options.wmstatus)
session.add( new_wf )
session.commit()
time.sleep(0.5)
else:
#print "already have",wf
pass
if cannot_inject:
#sendEmail('workflow duplicates','These workflow cannot be added in because of duplicates \n\n %s'%( '\n'.join(cannot_inject)))
sendLog('injector','These workflow cannot be added in because of duplicates \n\n %s'%( '\n'.join(cannot_inject)), level='warning')
## passing a round of invalidation of what needs to be invalidated
if use_mcm and (options.invalidate or True):
invalidator(url)
no_replacement = set()
## pick up replacements
for wf in session.query(Workflow).filter(Workflow.status == 'trouble').all():
print wf.name
if specific and not specific in wf.name: continue
print wf.name
wfi = workflowInfo(url, wf.name )
wl = wfi.request #getWorkLoad(url, wf.name)
familly = getWorkflowById( url, wl['PrepID'] )
true_familly = []
for member in familly:
if member == wf.name: continue
fwl = getWorkLoad(url , member)
if options.replace:
if member != options.replace: continue
else:
if fwl['RequestDate'] < wl['RequestDate']: continue
if fwl['RequestType']=='Resubmission': continue
if fwl['RequestStatus'] in ['None',None,'new']: continue
if fwl['RequestStatus'] in ['rejected','rejected-archived','aborted','aborted-archived']: continue
true_familly.append( fwl )
if len(true_familly)==0:
#sendLog('injector','%s had no replacement'%wf.name, level='critical')
wfi.sendLog('injector','the workflow was found in trouble with no replacement')
no_replacement.add( wf.name )
continue
else:
wfi.sendLog('injector','the workflow was found in trouble and has a replacement')
print wf.name,"has",len(familly),"familly members"
print wf.name,"has",len(true_familly),"true familly members"
##we cannot have more than one of them !!! pick the last one
if len(true_familly)>1:
#sendEmail('multiple wf','please take a look at injector for %s'%wf.name)
sendLog('injector','Multiple wf in line, will take the last one for %s \n%s'%( wf.name, ', '.join(fwl['RequestName'] for fwl in true_familly)), level='critical')
for fwl in true_familly[-1:]:
member = fwl['RequestName']
new_wf = session.query(Workflow).filter(Workflow.name == member).first()
if not new_wf:
sendLog('injector',"putting %s as replacement of %s"%( member, wf.name))
status = 'away'
if fwl['RequestStatus'] in ['assignment-approved']:
status = 'considered'
new_wf = Workflow( name = member, status = status, wm_status = fwl['RequestStatus'])
wf.status = 'forget'
session.add( new_wf )
else:
if new_wf.status == 'forget': continue
sendLog('injector',"getting %s as replacement of %s"%( new_wf.name, wf.name ))
wf.status = 'forget'
for tr in session.query(Transfer).all():
if wf.id in tr.workflows_id:
sw = copy.deepcopy(tr.workflows_id)
sw.remove( wf.id)
sw.append(new_wf.id)
tr.workflows_id = sw
print tr.phedexid,"got",new_wf.name
if new_wf.status != 'away':
print "\t setting it considered"
new_wf.status = 'considered'
if tr.phedexid<0: ## set it back to positive
tr.phedexid = -tr.phedexid
session.commit()
## don't do that automatically
#wf.status = 'forget'
session.commit()
if no_replacement:
#sendEmail('workflow with no replacement','%s \n are dangling there'%( '\n'.join(no_replacement)))
sendLog('injector','workflow with no replacement, %s \n are dangling there'% ( '\n'.join(no_replacement)), level='critical')