def retrievePandaJobs(job, jIDs):
'''
methods for retrieving panda job ids of panda jobs given a jobDefId
'''
from pandatools import Client
ick = False
jstatus = ''
num_pjobs = 0
logger.debug("retrievePandaJobs jIDs=%s" % jIDs)
# get status from Panda server
rc, jobsStatus = Client.getFullJobStatus(jIDs, False)
if rc:
logger.error('Return code %d retrieving job status information.', rc)
raise BackendError(
'Jedi', 'Return code %d retrieving job status information.' % rc)
for status in jobsStatus:
if not status: continue
jstatus = status.jobStatus
if status.jobStatus == None:
logger.warning('No panda jobs expected')
job.backend.pandajobs = []
elif status.jobStatus in [
"defined", "activated", "running", "failed", "finished",
"holding", "assigned"
]:
logger.debug('Panda jobs are running')
logger.debug("PandaID: %d" % status.PandaID)
pjobj = JediPandaJob()
pjobj.id = status.PandaID
pjobj.url = 'http://panda.cern.ch/?job=%d' % status.PandaID
pjobj.jopSpec = dict(zip(status._attributes, status.values()))
for k in pjobj.jobSpec.keys():
if type(pjobj.jobSpec[k]) not in [type(''), type(1)]:
pjobj.jobSpec[k] = str(pjobj.jobSpec[k])
if pjobj not in job.backend.pandajobs:
job.backend.pandajobs.append(pjobj)
else:
logger.debug("Panda job %s already exists locally" % pjobj.id)
num_pjobs += 1
else:
logger.warning(
"getFullJobStatus returned unsupported status %s for Panda job %s "
% (status.jobStatus, status.PandaID))
ick = True
return (ick, jstatus, num_pjobs)
def retrievePandaJobs(job, jIDs):
'''
methods for retrieving panda job ids of panda jobs given a jobDefId
'''
from pandatools import Client
ick = False
jstatus = ''
num_pjobs = 0
logger.debug("retrievePandaJobs jIDs=%s" %jIDs)
# get status from Panda server
rc, jobsStatus = Client.getFullJobStatus(jIDs,False)
if rc:
logger.error('Return code %d retrieving job status information.',rc)
raise BackendError('Jedi','Return code %d retrieving job status information.' % rc)
for status in jobsStatus:
if not status: continue
jstatus = status.jobStatus
if status.jobStatus == None:
logger.warning('No panda jobs expected')
job.backend.pandajobs = []
elif status.jobStatus in [ "defined", "activated", "running", "failed", "finished", "holding", "assigned"]:
logger.debug('Panda jobs are running')
logger.debug("PandaID: %d" % status.PandaID)
pjobj = JediPandaJob()
pjobj.id = status.PandaID
pjobj.url = 'http://panda.cern.ch/?job=%d' % status.PandaID
pjobj.jopSpec = dict(zip(status._attributes,status.values()))
for k in pjobj.jobSpec.keys():
if type(pjobj.jobSpec[k]) not in [type(''),type(1)]:
pjobj.jobSpec[k]=str(pjobj.jobSpec[k])
if pjobj not in job.backend.pandajobs:
job.backend.pandajobs.append(pjobj)
else:
logger.debug("Panda job %s already exists locally" % pjobj.id)
num_pjobs += 1
else:
logger.warning("getFullJobStatus returned unsupported status %s for Panda job %s " %(status.jobStatus, status.PandaID) )
ick = True
return (ick, jstatus, num_pjobs)
def status(self, JobID, forceUpdate=False):
# get logger
tmpLog = PLogger.getPandaLogger()
# check proxy
self.gridPassPhrase, self.vomsFQAN = PsubUtils.checkGridProxy(
self.gridPassPhrase, False, self.verbose)
# get job info from local repository
job = self.getJobInfo(JobID)
if job == None:
# not found
return None
# update if needed
if job.dbStatus != 'frozen' or forceUpdate:
tmpLog.info("Getting status for JobID=%s ..." % JobID)
# get status from Panda server
status, pandaIDstatus = Client.getPandIDsWithJobID(
JobID, verbose=self.verbose)
if status != 0:
tmpLog.error("Failed to get status for JobID=%s" % JobID)
return None
# get one job to set computingSite which may have changed due to rebrokerage
pandaJob = None
if pandaIDstatus != {}:
tmpPandaIDs = pandaIDstatus.keys()
tmpPandaIDs.sort()
status, tmpPandaJobs = Client.getFullJobStatus(
tmpPandaIDs[:1], verbose=self.verbose)
if status != 0:
tmpLog.error("Failed to get PandaJobs for %s" % JobID)
return None
pandaJob = tmpPandaJobs[0]
# convert to local job spec
job = PdbUtils.convertPtoD([],
pandaIDstatus,
job,
pandaJobForSiteID=pandaJob)
# update DB
try:
PdbUtils.updateJobDB(job, self.verbose)
except:
tmpLog.error("Failed to update local repository for JobID=%s" %
JobID)
return None
tmpLog.info("Updated JobID=%s" % JobID)
# return
return job
if status != 0:
print "Skip..."
continue
pandaJob = None
if pandaIDstatus != {}:
tmpPandaIDs = pandaIDstatus.keys()
tmpPandaIDs.sort()
nIDs = len(tmpPandaIDs)
nIDs_step = (nIDs + 9) / 10
for i, tmpID in enumerate(tmpPandaIDs):
if prog_options.verbose:
if i % nIDs_step == 0:
print i, '/', nIDs
status, tmpPandaJob = Client.getFullJobStatus([tmpID],
verbose=verbose2)
CPUTime = getattr(tmpPandaJob[0], "cpuConsumptionTime")
startTime = getattr(tmpPandaJob[0], "startTime")
endTime = getattr(tmpPandaJob[0], "endTime")
parentID = getattr(tmpPandaJob[0], "parentID")
jobsetID = getattr(tmpPandaJob[0], "jobsetID")
if prog_options.verbose:
print status, getattr(tmpPandaJob[0],
"jobStatus"), CPUTime, startTime, endTime
if getattr(tmpPandaJob[0],
"jobStatus") == "finished" and "runGen" in getattr(
tmpPandaJob[0], "transformation"):
TotalCPUTime += CPUTime
deltaT = endTime - startTime
TotalRealTime += deltaT.days * 86400 + deltaT.seconds + deltaT.microseconds * 1e-6
TotalJobs += 1
for r in res:
ids.append(r[0])
print 'got', len(ids), 'new pandaIDs'
print
except cx_Oracle.DatabaseError, exc:
error, = exc.args
print "updatePandaWAN.py - problem in establishing connection to db"
print "updatePandaWAN.py Oracle-Error-Code:", error.code
print "updatePandaWAN.py Oracle-Error-Message:", error.message
print 'leaving only first 200 in the list.'
del ids[2000:]
from pandatools import Client
status,jobSpec = Client.getFullJobStatus(ids,False)
print 'got back from panda:', len(jobSpec)
try:
connection = cx_Oracle.Connection(line)
cursor = cx_Oracle.Cursor(connection)
print 'Connection established.'
for i in jobSpec:
if i is None:
print "problem..."
continue
ti=i.pilotTiming.split('|')
if len(ti)!=5:
print i.PandaID,i.pilotTiming,ti
#cursor.execute("DELETE FROM ATLAS_WANHCTEST.result where pandaid=:pid",{'pid':i.PandaID});
continue
def retry(self,JobsetID,newSite=False,newOpts={},noSubmit=False,ignoreDuplication=False,useJobsetID=False,retryBuild=False,reproduceFiles=[],unsetRetryID=False):
# get logger
tmpLog = PLogger.getPandaLogger()
# check proxy
self.gridPassPhrase,self.vomsFQAN = PsubUtils.checkGridProxy(
self.gridPassPhrase,
False,
self.verbose,
useCache=True)
# force update just in case
self.status(JobsetID,True)
# set an empty map since mutable default value is used
if newOpts == {}:
newOpts = {}
# get jobset
newJobsetID = -1
jobList = self.getJobIDsWithSetID(JobsetID)
if jobList == None:
# works only for jobsetID
if useJobsetID:
return
# works with jobID
isJobset = False
jobList = [JobsetID]
else:
isJobset = True
tmpMsg = "ID=%s is composed of JobID=" % JobsetID
for tmpJobID in jobList:
tmpMsg += '%s,' % tmpJobID
tmpMsg = tmpMsg[:-1]
tmpLog.info(tmpMsg)
for JobID in jobList:
# get job info from local repository
localJob = self.getJobInfo(JobID)
if localJob == None:
tmpLog.warning("JobID=%s not found in local repository. Synchronization may be needed" % JobID)
return None
# for JEDI
if localJob.isJEDI():
status,out = Client.retryTask(
localJob.jediTaskID,
verbose=self.verbose,
properErrorCode=True,
newParams=newOpts)
if status != 0:
tmpLog.error(status)
tmpLog.error(out)
tmpLog.error("Failed to retry TaskID=%s" % localJob.jediTaskID)
return False
tmpStat,tmpDiag = out
if (not tmpStat in [0,True] and newOpts == {}) or (newOpts != {} and tmpStat != 3):
tmpLog.error(tmpDiag)
tmpLog.error("Failed to retry TaskID=%s" % localJob.jediTaskID)
return False
tmpLog.info(tmpDiag)
continue
# skip running job
if localJob.dbStatus != 'frozen':
tmpLog.info('Retry failed subjobs in running jobId=%s' % JobID)
status,out = Client.retryFailedJobsInActive(JobID,verbose=self.verbose)
if status != 0:
tmpLog.error(status)
tmpLog.error(out)
tmpLog.error("Failed to retry JobID=%s" % JobID)
else:
job = self.status(JobID)
if isJobset:
continue
else:
return
# skip already retried
if localJob.retryID != '0':
if isJobset:
tmpLog.info('Skip JobID=%s since already retried by JobID=%s JobsetID=%s' % \
(JobID,localJob.retryID,localJob.retryJobsetID))
continue
else:
tmpLog.warning('This job was already retried by JobID=%s' % localJob.retryID)
return
# check status of buildJob
if not retryBuild and not localJob.buildStatus in ['','finished']:
tmpMsgStr = 'Cannot retry since status of buildJob %s is %s (!= finished). ' \
% (localJob.PandaID.split(',')[0],localJob.buildStatus)
tmpMsgStr += 'Please execute %s with the same input/output datasets (or containers). ' % localJob.jobType
tmpMsgStr += 'It will run only on failed/cancelled/unused input files '
tmpMsgStr += 'and append output files to the output dataset container. '
tmpMsgStr += 'Or you may set retryBuild=True in pbook.retry() '
tmpLog.warning(tmpMsgStr)
if isJobset:
continue
else:
return
# check opts for newSite
if newSite or newOpts != {}:
if not localJob.outDS.endswith('/') and not newOpts.has_key('outDS') and not newOpts.has_key('--outDS'):
tmpLog.warning('You need to specify --outDS in newOpts to retry at new site unless container is used as output')
return
# get list of failed jobs
pandaIDs = localJob.PandaID.split(',')
statusList= localJob.jobStatus.split(',')
jobList = []
for idx in range(len(pandaIDs)):
# check status unless reproduce files
if reproduceFiles == [] and not statusList[idx] in ['failed','cancelled']:
continue
jobList.append(pandaIDs[idx])
# no failed job
if jobList == []:
if isJobset:
tmpLog.info('Skip JobID=%s since no failed jobs' % JobID)
continue
else:
tmpLog.info('No failed jobs to be retried for JobID=%s' % JobID)
return
# get full job spec
tmpLog.info("Retrying JobID=%s ..." % JobID)
tmpLog.info("Getting job info")
idxJL = 0
nQuery = 500
pandaJobs = []
while idxJL < len(jobList):
# avoid burst query
tmpLog.info(" %5s/%s" % (idxJL,len(jobList)))
status,oTmp = Client.getFullJobStatus(
jobList[idxJL:idxJL+nQuery],
verbose=self.verbose)
if status != 0:
tmpLog.error(status)
tmpLog.error(oTmp)
tmpLog.error("Cannot get job info from Panda server")
return
pandaJobs += oTmp
idxJL += nQuery
time.sleep(1)
tmpLog.info(" %5s/%s" % (len(jobList),len(jobList)))
# get PandaIDs to reproduce files
if reproduceFiles != []:
# change wildcard to .* for regexp
reproduceFilePatt = []
for tmpReproduceFile in reproduceFiles:
if '*' in tmpReproduceFile:
tmpReproduceFile = tmpReproduceFile.replace('*','.*')
reproduceFilePatt.append(tmpReproduceFile)
# get list of jobs which produced interesting files
tmpJobList = []
tmpPandaJobs = []
for tmpPandaJob in pandaJobs:
# check names
tmpMatchFlag = False
for tmpFile in tmpPandaJob.Files:
if tmpFile.type == 'output' and tmpFile.status == 'ready':
for tmpReproduceFile in reproduceFilePatt:
# normal matching
if tmpReproduceFile == tmpFile.lfn:
tmpMatchFlag = True
break
# wild card
if '*' in tmpReproduceFile and \
re.search('^'+tmpReproduceFile,tmpFile.lfn) != None:
tmpMatchFlag = True
break
if tmpMatchFlag:
break
# append
if tmpMatchFlag:
tmpJobList.append(tmpPandaJob.PandaID)
tmpPandaJobs.append(tmpPandaJob)
# use new list
jobList = tmpJobList
pandaJobs = tmpPandaJobs
if jobList == []:
tmpLog.info("No jobs to reproduce files : Jobs in JobID=%s didn't produce lost files" % JobID)
continue
# jobdefID
newJobdefID = PsubUtils.readJobDefID()
# reset some parameters
retryJobs = []
retrySite = None
retryElement = None
retryDestSE = None
outDsName = None
shadowList = []
oldLibDS = None
newLibDS = None
newLibTgz = None
rebroMap = {}
for idx in range(len(jobList)):
job = pandaJobs[idx]
# skip exired
if job == None:
tmpLog.warning("Could not retry jobs older than 30 days : JobID=%s (PandaID=%s) expired" \
% (JobID,jobList[idxJob]))
return
# skip jobs reassigned by rebrokerage
if (job.jobStatus == 'cancelled' and job.taskBufferErrorCode in [105,'105']) or \
(job.jobStatus == 'failed' and job.taskBufferErrorCode in [106,'106']):
# extract JobIDs of reassigned jobs
tmpM = re.search('JobsetID=(\d+) JobID=(\d+)',job.taskBufferErrorDiag)
if tmpM != None:
tmpRebKey = (tmpM.group(1),tmpM.group(2))
if not rebroMap.has_key(tmpRebKey):
rebroMap[tmpRebKey] = 0
# count # of reassigned jobs
rebroMap[tmpRebKey] += 1
continue
# get shadow list
if (not ignoreDuplication) and outDsName == None and job.prodSourceLabel == 'user':
# look for dataset for log since it doesn't have suffix even when --individualOutDS is used
for tmpFile in job.Files:
if tmpFile.type == 'log':
outDsName = tmpFile.dataset
break
# output dataset was not found
if outDsName == None:
tmpLog.error("Could not get output dataset name for JobID=%s (PandaID=%s)" \
% (JobID,job.PandaID))
return
# get files in shadow
if outDsName.endswith('/'):
shadowList = Client.getFilesInShadowDataset(
outDsName,
Client.suffixShadow,
self.verbose)
else:
# disable duplication check mainly for old overlay jobs since non-signal files are wrongly skipped
#shadowList = Client.getFilesInShadowDatasetOld(outDsName,Client.suffixShadow,self.verbose)
pass
# unify sitename
if retrySite == None:
retrySite = job.computingSite
retryElement = job.computingElement
retryDestSE = job.destinationSE
# reset
job.jobStatus = None
job.commandToPilot = None
job.startTime = None
job.endTime = None
job.attemptNr = 1+job.attemptNr
for attr in job._attributes:
if attr.endswith('ErrorCode') or attr.endswith('ErrorDiag'):
setattr(job,attr,None)
job.transExitCode = None
job.computingSite = retrySite
job.computingElement = retryElement
job.destinationSE = retryDestSE
job.dispatchDBlock = None
if not unsetRetryID:
job.jobExecutionID = JobID
job.jobDefinitionID = newJobdefID
job.parentID = job.PandaID
if job.jobsetID != ['NULL',None,-1]:
if not unsetRetryID:
job.sourceSite = job.jobsetID
job.jobsetID = newJobsetID
skipInputList = []
numUsedFiles = 0
# loop over all files
for file in job.Files:
file.rowID = None
if file.type == 'input':
# protection against wrong sync which doesn't update buildStatus correctly
if not retryBuild and file.lfn.endswith('.lib.tgz') and file.GUID == 'NULL':
tmpLog.warning('GUID for %s is unknown. Cannot retry when corresponding buildJob failed' \
% file.lfn)
return
if not retryBuild or not file.lfn.endswith('.lib.tgz'):
file.status = 'ready'
# set new lib dataset
if retryBuild and file.lfn.endswith('.lib.tgz'):
if newLibTgz != None:
file.lfn = newLibTgz
file.dataset = newLibDS
file.dispatchDBlock = newLibDS
# check with shadow for non lib.tgz/DBR
tmpDbrMatch = re.search('^DBRelease-.*\.tar\.gz$',file.lfn)
if tmpDbrMatch == None and not file.lfn.endswith('.lib.tgz'):
if file.lfn in shadowList:
skipInputList.append(file)
else:
numUsedFiles += 1
elif file.type in ('output','log'):
file.destinationSE = retryDestSE
file.destinationDBlock = re.sub('_sub\d+$','',file.destinationDBlock)
# add retry num
if file.dataset.endswith('/') or job.prodSourceLabel == 'panda':
oldOutDsName = file.destinationDBlock
retryDsPatt = '_r'
if reproduceFiles != []:
retryDsPatt = '_rp'
retryMatch = re.search(retryDsPatt+'(\d+)$',file.destinationDBlock)
if retryMatch == None:
file.destinationDBlock += (retryDsPatt+'1')
else:
tmpDestinationDBlock = re.sub(retryDsPatt+'(\d+)$','',file.destinationDBlock)
file.destinationDBlock = tmpDestinationDBlock + retryDsPatt + '%d' % (1+int(retryMatch.group(1)))
if job.processingType == 'usermerge':
job.jobParameters = job.jobParameters.replace(' %s ' % oldOutDsName,
' %s ' % file.destinationDBlock)
# use new dataset name for buildXYZ
if job.prodSourceLabel == 'panda':
if file.lfn.endswith('.lib.tgz'):
# get new libDS and lib.tgz names
oldLibDS = file.dataset
file.dataset = file.destinationDBlock
newLibDS = file.dataset
file.lfn = re.sub(oldLibDS,newLibDS,file.lfn)
newLibTgz = file.lfn
else:
file.dataset = file.destinationDBlock
# add attempt nr
oldName = file.lfn
if job.prodSourceLabel == 'panda' and file.lfn.endswith('.lib.tgz'):
continue
else:
# append attempt number at the tail
file.lfn = re.sub("\.\d+$","",file.lfn)
file.lfn = "%s.%d" % (file.lfn,job.attemptNr)
newName = file.lfn
# modify jobParameters
job.jobParameters = re.sub("'%s'" % oldName ,"'%s'" % newName,
job.jobParameters)
# look for output in trf
oldGenelicName = re.sub('\.\d+$','',oldName)
match = re.search(oldGenelicName+'(\.\d+)*(%20|")',job.jobParameters)
if match != None:
job.jobParameters = job.jobParameters.replace(match.group(0),newName+match.group(2))
# change lib.tgz name
if retryBuild and newLibDS != None:
job.jobParameters = re.sub(oldLibDS,newLibDS,job.jobParameters)
# change destinationDBlock
if job.prodSourceLabel == 'panda':
job.destinationDBlock = newLibDS
# all files are used by others
if numUsedFiles == 0 and skipInputList != []:
continue
# remove skipped files
strSkipped = ''
for tmpFile in skipInputList:
strSkipped += '%s,' % tmpFile.lfn
job.Files.remove(tmpFile)
strSkipped = strSkipped[:-1]
# modify jobpar
if strSkipped != '':
optionToSkipFiles = '--skipInputByRetry'
if not optionToSkipFiles in job.jobParameters:
# just append
job.jobParameters += "%s=%s " % (optionToSkipFiles,strSkipped)
else:
# extract already skipped files
tmpMatch = re.search("(%s=[^ ]+)",job.jobParameters)
if tmpMatch == None:
tmpLog.error("Failed to extract arg of %s for PandaID=%s" \
% (optionToSkipFiles,job.PandaID))
return
# replace
job.jobParameters = re.sub(tmpMatch.group(1),"%s,%s" % (tmpMatch.group(1),optionToSkipFiles),
job.jobParameters)
if self.verbose:
tmpLog.debug(job.jobParameters)
# append
retryJobs.append(job)
# info on rebrokeage
if rebroMap != {}:
for tmpRebKey,tmpRebNumJobs in rebroMap.iteritems():
tmpRebSetID,tmpRebJobID = tmpRebKey
tmpLog.info('Skip %s jobs since JobID=%s JobsetID=%s already reassigned them to another site' % \
(tmpRebNumJobs,tmpRebJobID,tmpRebSetID))
if retryJobs == []:
tmpLog.info("No more jobs to be retried for JobID=%s" % JobID)
if isJobset:
continue
else:
return
# all input files were or are being used by other jobs
if retryJobs == []:
tmpLog.info('All input files were or are being used by other jobs for the same output. No jobs to be retried. If you need to ignore duplication check (e.g., using the same EVNT file for multiple simulation subjobs), set ignoreDuplication=True. i.e. retry(123,ignoreDuplication=True)')
if isJobset:
continue
else:
return
# check voms role
if not retryJobs[0].workingGroup in ['NULL',None,'']:
# VOMS role was used
if not "--workingGroup" in job.metadata:
# extract voms roles from metadata
match = re.search("--voms( |=)[ \"]*([^ \"]+)",job.metadata)
if match != None:
vomsRoles = match.group(2)
else:
vomsRoles = "atlas:/atlas/%s/Role=production" % retryJobs[0].workingGroup
# regenerate proxy with VOMS roles
try:
tmpLog.info("Checking proxy role to resubmit %s jobs" % retryJobs[0].workingGroup)
self.gridPassPhrase,self.vomsFQAN = PsubUtils.checkGridProxy(
self.gridPassPhrase,
False,
self.verbose,vomsRoles,
useCache=True)
except:
tmpLog.error("Failed to generate a proxy with %s" % vomsRoles)
return
# check runtime env for new site submission
if (newSite or newOpts != {}):
if retryJobs[0].processingType == 'pathena' or '--useAthenaPackages' in retryJobs[0].metadata:
from pandatools import AthenaUtils
stA,retA = AthenaUtils.getAthenaVer()
if not stA:
tmpLog.error("Failed to get Athena rel/cache version in current runtime env")
return
athenaVer = retA['athenaVer']
cacheVer = retA['cacheVer']
nightVer = retA['nightVer']
wrongSetup = False
if retryJobs[0].AtlasRelease != 'Atlas-%s' % athenaVer:
wrongSetup = True
errMsg = "Current Athena version Atlas-%s is inconsitent with the previous submission %s. " % (athenaVer,retryJobs[0].AtlasRelease)
elif retryJobs[0].homepackage != 'AnalysisTransforms'+cacheVer+nightVer:
wrongSetup = True
errMsg = "Current cache version %s is inconsitent with the previous submission. " % cacheVer.replace('-','').replace('_','-')
if wrongSetup:
errMsg += 'You need to have the same runtime env as before since all job spec need to be re-created to send jobs to a new site. '
errMsg += 'Please setup Athena correctly and restart pbook'
tmpLog.error(errMsg)
return
# test mode
if noSubmit:
continue
# invoke pathena/prun to send job to new site
if (newSite or newOpts != {}) and retryJobs[0].processingType != 'usermerge':
# set parent jobID and jobsetID
newOpts['provenanceID'] = retryJobs[0].jobExecutionID
newOpts['panda_parentJobsetID'] = retryJobs[0].sourceSite
tmpLog.info("Constructing job spec again to be sent to another site ...")
comStat= PsubUtils.execWithModifiedParams(retryJobs,newOpts,self.verbose,newSite)
if comStat == 0:
# update database
time.sleep(2)
self.sync()
else:
tmpLog.error("Failed to submit jobs to Panda server")
return
# register datasets
tmpOutDsLocation = Client.PandaSites[retryJobs[-1].computingSite]['ddm']
addedDataset = []
shadowDSname = None
for tmpFile in retryJobs[-1].Files:
if tmpFile.type in ['output','log'] and tmpFile.dataset.endswith('/'):
# add shadow
"""
removed shadow
if shadowDSname == None and tmpFile.type == 'log':
shadowDSname = "%s%s" % (tmpFile.destinationDBlock,Client.suffixShadow)
Client.addDataset(shadowDSname,self.verbose)
"""
# add datasets
if not tmpFile.destinationDBlock in addedDataset:
# create dataset
Client.addDataset(
tmpFile.destinationDBlock,
self.verbose,
location=tmpOutDsLocation,
dsCheck=False)
# add to container
Client.addDatasetsToContainer(
tmpFile.dataset,
[tmpFile.destinationDBlock],
self.verbose)
# append
addedDataset.append(tmpFile.destinationDBlock)
# register libDS
if retryBuild and newLibDS != None:
Client.addDataset(
newLibDS,
self.verbose,
location=tmpOutDsLocation,
dsCheck=False)
# submit
tmpLog.info("Submitting job ...")
status,out = Client.submitJobs(retryJobs,verbose=self.verbose)
if out == None or status != 0:
tmpLog.error(status)
tmpLog.error(out)
tmpLog.error("Failed to submit jobs to Panda server")
return
# update database
pandaIDstatus = {}
newJobID = None
for items in out:
# get newJobID
if newJobID == None:
newJobID = items[1]
# check PandaID
PandaID = items[0]
if PandaID == 'NULL':
tmpLog.error("Panda server returned wrong IDs. It may have a temporary problem")
return
# set newJobsetID
if newJobsetID in [None,-1]:
newJobsetID = items[2]['jobsetID']
# dummy statuso
pandaIDstatus[PandaID] = ('defined','NULL')
# set retry ID
if not unsetRetryID:
localJob.retryID = newJobID
if not newJobsetID in [None,-1,'NULL']:
localJob.retryJobsetID = newJobsetID
try:
PdbUtils.updateJobDB(localJob,self.verbose)
except:
tmpLog.error("Failed to set retryID for JobID=%s" % JobID)
return
# set new paramers
newLocalJob = PdbUtils.convertPtoD(retryJobs,pandaIDstatus)
newLocalJob.JobID = newJobID
if not newJobsetID in [None,-1,'NULL']:
newLocalJob.groupID = newJobsetID
newLocalJob.creationTime = datetime.datetime.utcnow()
# insert to DB
try:
PdbUtils.insertJobDB(newLocalJob,self.verbose)
except:
tmpLog.error("Failed to insert JobID=%s to local repository" % newJobID)
return
# write new jobdefID
PsubUtils.writeJobDefID(newJobID)
# done
tmpMsg = 'Done. New JobID=%s' % newJobID
if not newJobsetID in [None,-1,'NULL']:
tmpMsg += " JobsetID=%s" % newJobsetID
tmpLog.info(tmpMsg)
def status(self,JobID,forceUpdate=False):
# get logger
tmpLog = PLogger.getPandaLogger()
# check proxy
self.gridPassPhrase,self.vomsFQAN = PsubUtils.checkGridProxy(
self.gridPassPhrase,
False,
self.verbose,
useCache=True)
# get job info from local repository
job = self.getJobInfo(JobID)
if job == None:
# not found
return None
# update if needed
if job.dbStatus != 'frozen' or forceUpdate:
if not job.isJEDI():
tmpLog.info("Getting status for JobID=%s ..." % JobID)
# get status from Panda server
status,pandaIDstatus = Client.getPandIDsWithJobID(JobID,verbose=self.verbose)
if status != 0:
tmpLog.error("Failed to get status for ID=%s" % JobID)
return None
# get one job to set computingSite which may have changed due to rebrokerage
pandaJob = None
if pandaIDstatus != {}:
tmpPandaIDs = pandaIDstatus.keys()
tmpPandaIDs.sort()
status,tmpPandaJobs = Client.getFullJobStatus(
tmpPandaIDs[:1],
verbose=self.verbose)
if status != 0:
tmpLog.error("Failed to get PandaJobs for %s" % JobID)
return None
pandaJob = tmpPandaJobs[0]
# convert to local job spec
job = PdbUtils.convertPtoD([],pandaIDstatus,job,pandaJobForSiteID=pandaJob)
# check merge job generation
status = self.setMergeJobStatus(job,forceUpdate)
if not status:
return None
else:
tmpLog.info("Getting status for TaskID=%s ..." % job.jediTaskID)
# get JEDI task
status,jediTaskDict = Client.getJediTaskDetails(
{'jediTaskID':job.jediTaskID},
False,
True,
verbose=self.verbose)
if status != 0:
tmpLog.error("Failed to get task details for %s" % JobID)
return
# convert JEDI task
job = PdbUtils.convertJTtoD(jediTaskDict,job)
# update DB
try:
PdbUtils.updateJobDB(job,self.verbose)
except:
tmpLog.error("Failed to update local repository for JobID=%s" % JobID)
return None
if not job.isJEDI():
tmpLog.info("Updated JobID=%s" % JobID)
else:
tmpLog.info("Updated TaskID=%s ..." % job.jediTaskID)
# return
return job
def master_updateMonitoringInformation(jobs):
'''Monitor jobs'''
from pandatools import Client
#active_status = [ None, 'defined', 'unknown', 'assigned', 'waiting', 'activated', 'sent', 'starting', 'running', 'holding', 'transferring' ]
submitting_status = [ ]
active_status = [ None, 'registered', 'waiting', 'defined', 'pending', 'assigning', 'ready', 'scouting', 'running', 'holding', 'merging', 'prepared', 'aborting', 'finishing' ]
inactive_status = [ 'finished', 'aborted', 'broken', 'failed', 'done' ]
# Find jobs to be monitored
jobdict = {}
for job in jobs:
# add a delay as Panda can be a little slow in sorting out a new Task
if job.backend.id and job.backend.status in active_status and ( (datetime.datetime.utcnow() - job.time.timestamps["submitted"]).seconds > 120):
jobdict[job.backend.id] = job
logger.debug("jobdict = %s" %jobdict)
# Monitor active Jedi tasks
allJobIDs = jobdict.keys()
pandaJobIDs = {}
for jID in allJobIDs:
status, jediTaskDict = Client.getJediTaskDetails({'jediTaskID': jID},False,True,verbose=False)
if status != 0:
logger.error("Failed to get task details for %s" % jID)
#raise BackendError('Jedi','Return code %d retrieving job status information.' % status)
continue
# Retrieve job
job = jobdict[jediTaskDict['jediTaskID']]
# Store associated Panda jobs
if job.backend.pandajobs:
pandaJobIDs[job.backend.id] = [pj.id for pj in job.backend.pandajobs]
else:
pandaJobIDs[jediTaskDict['jediTaskID']] = jediTaskDict['PandaID']
logger.debug("jID = %s, pandaJobIDs = %s" % (jID, pandaJobIDs))
# Fill the output data dataset list
if 'outDS' in jediTaskDict and jediTaskDict['outDS'] != '':
for ds in jediTaskDict['outDS'].split(','):
if not ds in job.outputdata.datasetList:
job.outputdata.datasetList.append(ds)
# Jedi job status has changed
if job.backend.status != jediTaskDict['status']:
logger.debug('Job %s has changed status from %s to %s',job.getFQID('.'),job.backend.status, jediTaskDict['status'])
job.backend.status = jediTaskDict['status']
job.backend.reason = jediTaskDict['statistics']
# Now update Jedi job status
if jediTaskDict['status'] in ['registered', 'waiting', 'defined', 'pending', 'assigning', 'ready']:
job.updateStatus('submitted')
elif jediTaskDict['status'] in ['scouting', 'running', 'holding', 'merging', 'prepared' ]:
job.updateStatus('running')
elif jediTaskDict['status'] in ['done']:
job.updateStatus('completed')
elif jediTaskDict['status'] in ['failed', 'finished']:
job.updateStatus('failed')
elif jediTaskDict['status'] in [ 'aborted', 'broken', 'cancelled' ] and job.status not in ['completed','failed']:
job.updateStatus('killed')
else:
logger.warning('Unexpected Jedi task status %s', jediTaskDict['status'])
# Check if associated Panda job exist and monitor them
if not job.backend.pandajobs:
jdefids = pandaJobIDs[jID]
# skip if there are no Panda jobs yet
if not jdefids:
continue
tot_num_mjobs = 0
do_master_update = True
ick,status,num_mjobs = retrievePandaJobs(job, jdefids)
logger.debug('retrievePandaJobs returns: %s %s' % (repr(ick),status))
if not ick:
logger.debug('Panda job retrival failure for Jedi task %s with PandaIds %s' % (job.backend.id, jdefids))
do_master_update = False
tot_num_mjobs += num_mjobs
logger.debug('Job %s retrieved %d Panda jobs' % (job.getFQID('.'),tot_num_mjobs) )
# Now monitor the already attached Panda jobs
else:
jdefids = [ pj.id for pj in job.backend.pandajobs ]
rc, jobsStatus = Client.getFullJobStatus(jdefids,False)
if rc:
logger.error('Return code %d retrieving job status information.',rc)
raise BackendError('Jedi','Return code %d retrieving job status information.' % rc)
for status in jobsStatus:
if not status: continue
for pjob in job.backend.pandajobs:
if pjob.id == status.PandaID:
# skip if no status change
if pjob.status == status.jobStatus:
continue
# Else update job record
pjob.jobSpec = dict(zip(status._attributes,status.values()))
for k in pjob.jobSpec.keys():
if type(pjob.jobSpec[k]) not in [type(''),type(1)]:
pjob.jobSpec[k]=str(pjob.jobSpec[k])
logger.debug('Job %s with Panda job %s has changed status from %s to %s',job.getFQID('.'),pjob.id, pjob.status,status.jobStatus)
pjob.status = status.jobStatus
pjob.exitcode = str(status.transExitCode)
pjob.piloterrorcode = str(status.pilotErrorCode)
pjob.reason = ''
for k in pjob.jobSpec.keys():
if k.endswith('ErrorDiag') and pjob.jobSpec[k]!='NULL':
pjob.reason += '%s: %s, '%(k,str(pjob.jobSpec[k]))
#if job.backend.jobSpec['transExitCode'] != 'NULL':
pjob.reason += 'transExitCode: %s'%pjob.jobSpec['transExitCode']
if status.jobStatus in ['defined','unknown','assigned','waiting','activated','sent']:
logger.debug('Panda job %s %s' % (pjob.id, status.jobStatus))
elif status.jobStatus in ['starting','running','holding','transferring', 'merging']:
logger.debug('Panda job %s %s '% (pjob.id, status.jobStatus))
elif status.jobStatus in ['finished']:
logger.debug('Panda job %s %s '% (pjob.id, status.jobStatus))
elif status.jobStatus == 'failed':
logger.debug('Panda job %s %s '% (pjob.id, status.jobStatus))
# check for server side retry
if 'taskBufferErrorDiag' in pjob.jobSpec and pjob.jobSpec['taskBufferErrorDiag'].find("PandaID=") != -1:
# grab the new panda ID
newPandaID = long(pjob.jobSpec['taskBufferErrorDiag'].split("=")[1])
pjob.id = newPandaID
pjob.status = None
pjob.url = 'http://panda.cern.ch/?job=%d'%newPandaID
elif status.jobStatus == 'cancelled' and pjob.status not in ['completed','failed']: # bug 67716
logger.debug('Panda job %s cancelled'%pjob.id)
if 'taskBufferErrorDiag' in pjob.jobSpec and "rebrokerage" in pjob.jobSpec['taskBufferErrorDiag']:
newPandaID = checkForRebrokerage(pjob.jobSpec['taskBufferErrorDiag'])
logger.warning("Subjob rebrokered by Panda server. Job %d moved to %d."%(pjob.id, newPandaID))
pjob.id = newPandaID
pjob.status = None
else:
logger.warning('Unexpected job status %s',status.jobStatus)
def master_updateMonitoringInformation(jobs):
'''Monitor jobs'''
from pandatools import Client
#active_status = [ None, 'defined', 'unknown', 'assigned', 'waiting', 'activated', 'sent', 'starting', 'running', 'holding', 'transferring' ]
submitting_status = []
active_status = [
None, 'registered', 'waiting', 'defined', 'pending', 'assigning',
'ready', 'scouting', 'running', 'holding', 'merging', 'prepared',
'aborting', 'finishing'
]
inactive_status = ['finished', 'aborted', 'broken', 'failed', 'done']
# Find jobs to be monitored
jobdict = {}
for job in jobs:
# add a delay as Panda can be a little slow in sorting out a new Task
if job.backend.id and job.backend.status in active_status and (
(datetime.datetime.utcnow() -
job.time.timestamps["submitted"]).seconds > 120):
jobdict[job.backend.id] = job
logger.debug("jobdict = %s" % jobdict)
# Monitor active Jedi tasks
allJobIDs = jobdict.keys()
pandaJobIDs = {}
for jID in allJobIDs:
status, jediTaskDict = Client.getJediTaskDetails(
{'jediTaskID': jID}, False, True, verbose=False)
if status != 0:
logger.error("Failed to get task details for %s" % jID)
#raise BackendError('Jedi','Return code %d retrieving job status information.' % status)
continue
# Retrieve job
job = jobdict[jediTaskDict['jediTaskID']]
# Store associated Panda jobs
if job.backend.pandajobs:
pandaJobIDs[job.backend.id] = [
pj.id for pj in job.backend.pandajobs
]
else:
pandaJobIDs[
jediTaskDict['jediTaskID']] = jediTaskDict['PandaID']
logger.debug("jID = %s, pandaJobIDs = %s" % (jID, pandaJobIDs))
# Fill the output data dataset list
if 'outDS' in jediTaskDict and jediTaskDict['outDS'] != '':
for ds in jediTaskDict['outDS'].split(','):
if not ds in job.outputdata.datasetList:
job.outputdata.datasetList.append(ds)
# Jedi job status has changed
if job.backend.status != jediTaskDict['status']:
logger.debug('Job %s has changed status from %s to %s',
job.getFQID('.'), job.backend.status,
jediTaskDict['status'])
job.backend.status = jediTaskDict['status']
job.backend.reason = jediTaskDict['statistics']
# Now update Jedi job status
if jediTaskDict['status'] in [
'registered', 'waiting', 'defined', 'pending',
'assigning', 'ready'
]:
job.updateStatus('submitted')
elif jediTaskDict['status'] in [
'scouting', 'running', 'holding', 'merging', 'prepared'
]:
job.updateStatus('running')
elif jediTaskDict['status'] in ['done']:
job.updateStatus('completed')
elif jediTaskDict['status'] in ['failed', 'finished']:
job.updateStatus('failed')
elif jediTaskDict['status'] in [
'aborted', 'broken', 'cancelled'
] and job.status not in ['completed', 'failed']:
job.updateStatus('killed')
else:
logger.warning('Unexpected Jedi task status %s',
jediTaskDict['status'])
# Check if associated Panda job exist and monitor them
if not job.backend.pandajobs:
jdefids = pandaJobIDs[jID]
# skip if there are no Panda jobs yet
if not jdefids:
continue
tot_num_mjobs = 0
do_master_update = True
ick, status, num_mjobs = retrievePandaJobs(job, jdefids)
logger.debug('retrievePandaJobs returns: %s %s' %
(repr(ick), status))
if not ick:
logger.debug(
'Panda job retrival failure for Jedi task %s with PandaIds %s'
% (job.backend.id, jdefids))
do_master_update = False
tot_num_mjobs += num_mjobs
logger.debug('Job %s retrieved %d Panda jobs' %
(job.getFQID('.'), tot_num_mjobs))
# Now monitor the already attached Panda jobs
else:
jdefids = [pj.id for pj in job.backend.pandajobs]
rc, jobsStatus = Client.getFullJobStatus(jdefids, False)
if rc:
logger.error(
'Return code %d retrieving job status information.',
rc)
raise BackendError(
'Jedi',
'Return code %d retrieving job status information.' %
rc)
for status in jobsStatus:
if not status: continue
for pjob in job.backend.pandajobs:
if pjob.id == status.PandaID:
# skip if no status change
if pjob.status == status.jobStatus:
continue
# Else update job record
pjob.jobSpec = dict(
zip(status._attributes, status.values()))
for k in pjob.jobSpec.keys():
if type(pjob.jobSpec[k]) not in [
type(''), type(1)
]:
pjob.jobSpec[k] = str(pjob.jobSpec[k])
logger.debug(
'Job %s with Panda job %s has changed status from %s to %s',
job.getFQID('.'), pjob.id, pjob.status,
status.jobStatus)
pjob.status = status.jobStatus
pjob.exitcode = str(status.transExitCode)
pjob.piloterrorcode = str(status.pilotErrorCode)
pjob.reason = ''
for k in pjob.jobSpec.keys():
if k.endswith('ErrorDiag'
) and pjob.jobSpec[k] != 'NULL':
pjob.reason += '%s: %s, ' % (
k, str(pjob.jobSpec[k]))
#if job.backend.jobSpec['transExitCode'] != 'NULL':
pjob.reason += 'transExitCode: %s' % pjob.jobSpec[
'transExitCode']
if status.jobStatus in [
'defined', 'unknown', 'assigned',
'waiting', 'activated', 'sent'
]:
logger.debug('Panda job %s %s' %
(pjob.id, status.jobStatus))
elif status.jobStatus in [
'starting', 'running', 'holding',
'transferring', 'merging'
]:
logger.debug('Panda job %s %s ' %
(pjob.id, status.jobStatus))
elif status.jobStatus in ['finished']:
logger.debug('Panda job %s %s ' %
(pjob.id, status.jobStatus))
elif status.jobStatus == 'failed':
logger.debug('Panda job %s %s ' %
(pjob.id, status.jobStatus))
# check for server side retry
if 'taskBufferErrorDiag' in pjob.jobSpec and pjob.jobSpec[
'taskBufferErrorDiag'].find(
"PandaID=") != -1:
# grab the new panda ID
newPandaID = long(
pjob.jobSpec['taskBufferErrorDiag'].
split("=")[1])
pjob.id = newPandaID
pjob.status = None
pjob.url = 'http://panda.cern.ch/?job=%d' % newPandaID
elif status.jobStatus == 'cancelled' and pjob.status not in [
'completed', 'failed'
]: # bug 67716
logger.debug('Panda job %s cancelled' %
pjob.id)
if 'taskBufferErrorDiag' in pjob.jobSpec and "rebrokerage" in pjob.jobSpec[
'taskBufferErrorDiag']:
newPandaID = checkForRebrokerage(
pjob.jobSpec['taskBufferErrorDiag'])
logger.warning(
"Subjob rebrokered by Panda server. Job %d moved to %d."
% (pjob.id, newPandaID))
pjob.id = newPandaID
pjob.status = None
else:
logger.warning('Unexpected job status %s',
status.jobStatus)
def retry(self,
JobsetID,
newSite=False,
newOpts={},
noSubmit=False,
ignoreDuplication=True):
# get logger
tmpLog = PLogger.getPandaLogger()
# check proxy
self.gridPassPhrase, self.vomsFQAN = PsubUtils.checkGridProxy(
self.gridPassPhrase, False, self.verbose)
# get jobset
newJobsetID = -1
jobList = self.getJobIDsWithSetID(JobsetID)
if jobList == None:
isJobset = False
jobList = [JobsetID]
else:
isJobset = True
tmpMsg = "JobsetID=%s is composed of JobID=" % JobsetID
for tmpJobID in jobList:
tmpMsg += '%s,' % tmpJobID
tmpMsg = tmpMsg[:-1]
tmpLog.info(tmpMsg)
for JobID in jobList:
# get job info from local repository
localJob = self.getJobInfo(JobID)
if localJob == None:
tmpLog.warning(
"JobID=%s not found in local repository. Synchronization may be needed"
% JobID)
return None
# skip running job
if localJob.dbStatus != 'frozen':
tmpLog.warning('Cannot retry running jobs')
if isJobset:
continue
else:
return
# skip already retried
if localJob.retryID != '0':
if isJobset:
tmpLog.info('Skip JobID=%s since already retried by JobID=%s JobsetID=%s' % \
(JobID,localJob.retryID,localJob.retryJobsetID))
continue
else:
tmpLog.warning('This job was already retried by JobID=%s' %
localJob.retryID)
return
# check status of buildJob
if not localJob.buildStatus in ['', 'finished']:
tmpMsgStr = 'Cannot retry since status of buildJob %s is %s (!= finished). ' \
% (localJob.PandaID.split(',')[0],localJob.buildStatus)
tmpMsgStr += 'Please execute %s with the same input/output datasets (or containers). ' % localJob.jobType
tmpMsgStr += 'It will run only on failed/cancelled/unused input files '
tmpMsgStr += 'and append output files to the output dataset container'
tmpLog.warning(tmpMsgStr)
if isJobset:
continue
else:
return
# check opts for newSite
if newSite:
if not localJob.outDS.endswith('/') and not newOpts.has_key(
'outDS') and not newOpts.has_key('--outDS'):
tmpLog.warning(
'You need to specify --outDS in newOpts to retry at new site unless container is used as output'
)
return
# get list of failed jobs
pandaIDs = localJob.PandaID.split(',')
statusList = localJob.jobStatus.split(',')
jobList = []
for idx in range(len(pandaIDs)):
# check status
if not statusList[idx] in ['failed', 'cancelled']:
continue
jobList.append(pandaIDs[idx])
# no failed job
if jobList == []:
if isJobset:
tmpLog.info('Skip JobID=%s since no failed jobs' % JobID)
continue
else:
tmpLog.info('No failed jobs to be retried for JobID=%s' %
JobID)
return
# get full job spec
tmpLog.info("Retrying JobID=%s ..." % JobID)
tmpLog.info("Getting job info")
idxJL = 0
nQuery = 500
pandaJobs = []
while idxJL < len(jobList):
# avoid burst query
tmpLog.info(" %5s/%s" % (idxJL, len(jobList)))
status, oTmp = Client.getFullJobStatus(jobList[idxJL:idxJL +
nQuery],
verbose=self.verbose)
if status != 0:
tmpLog.error(status)
tmpLog.error(oTmp)
tmpLog.error("Cannot get job info from Panda server")
return
pandaJobs += oTmp
idxJL += nQuery
time.sleep(1)
tmpLog.info(" %5s/%s" % (len(jobList), len(jobList)))
# jobdefID
newJobdefID = PsubUtils.readJobDefID()
# reset some parameters
retryJobs = []
retrySite = None
retryElement = None
retryDestSE = None
outDsName = None
shadowList = []
for idx in range(len(jobList)):
job = pandaJobs[idx]
# skip exired
if job == None:
tmpLog.warning("Could not retry jobs older than 30 days : JobID=%s (PandaID=%s) expired" \
% (JobID,jobList[idxJob]))
return
# get shadow list
if (not ignoreDuplication
) and outDsName == None and job.prodSourceLabel == 'user':
# look for dataset for log since it doesn't have suffix even when --individualOutDS is used
for tmpFile in job.Files:
if tmpFile.type == 'log':
outDsName = tmpFile.dataset
break
# output dataset was not found
if outDsName == None:
tmpLog.error("Could not get output dataset name for JobID=%s (PandaID=%s)" \
% (JobID,job.PandaID))
return
# get files in shadow
if outDsName.endswith('/'):
shadowList = Client.getFilesInShadowDataset(
outDsName, Client.suffixShadow, self.verbose)
else:
# disable duplication check mainly for old overlay jobs since non-signal files are wrongly skipped
#shadowList = Client.getFilesInShadowDatasetOld(outDsName,Client.suffixShadow,self.verbose)
pass
# unify sitename
if retrySite == None:
retrySite = job.computingSite
retryElement = job.computingElement
retryDestSE = job.destinationSE
# reset
job.jobStatus = None
job.commandToPilot = None
job.startTime = None
job.endTime = None
job.attemptNr = 1 + job.attemptNr
for attr in job._attributes:
if attr.endswith('ErrorCode') or attr.endswith(
'ErrorDiag'):
setattr(job, attr, None)
job.transExitCode = None
job.computingSite = retrySite
job.computingElement = retryElement
job.destinationSE = retryDestSE
job.dispatchDBlock = None
job.jobExecutionID = JobID
job.jobDefinitionID = newJobdefID
job.parentID = job.PandaID
if job.jobsetID != ['NULL', None, -1]:
job.sourceSite = job.jobsetID
job.jobsetID = newJobsetID
skipInputList = []
numUsedFiles = 0
for file in job.Files:
file.rowID = None
if file.type == 'input':
# protection against wrong sync which doesn't update buildStatus correctly
if file.lfn.endswith(
'.lib.tgz') and file.GUID == 'NULL':
tmpLog.warning('GUID for %s is unknown. Cannot retry when corresponding buildJob failed' \
% file.lfn)
return
file.status = 'ready'
# check with shadow for non lib.tgz/DBR
tmpDbrMatch = re.search('^DBRelease-.*\.tar\.gz$',
file.lfn)
if tmpDbrMatch == None and not file.lfn.endswith(
'.lib.tgz'):
if file.lfn in shadowList:
skipInputList.append(file)
else:
numUsedFiles += 1
elif file.type in ('output', 'log'):
file.destinationSE = retryDestSE
file.destinationDBlock = re.sub(
'_sub\d+$', '', file.destinationDBlock)
# add retry num
if file.dataset.endswith('/'):
retryMatch = re.search('_r(\d+)$',
file.destinationDBlock)
if retryMatch == None:
file.destinationDBlock += '_r1'
else:
tmpDestinationDBlock = re.sub(
'_r(\d+)$', '', file.destinationDBlock)
file.destinationDBlock = tmpDestinationDBlock + '_r%d' % (
1 + int(retryMatch.group(1)))
# add attempt nr
oldName = file.lfn
file.lfn = re.sub("\.\d+$", "", file.lfn)
file.lfn = "%s.%d" % (file.lfn, job.attemptNr)
newName = file.lfn
# modify jobParameters
job.jobParameters = re.sub("'%s'" % oldName,
"'%s'" % newName,
job.jobParameters)
# look for output in trf
oldGenelicName = re.sub('\.\d+$', '', oldName)
match = re.search(oldGenelicName + '(\.\d+)*(%20|")',
job.jobParameters)
if match != None:
job.jobParameters = job.jobParameters.replace(
match.group(0), newName + match.group(2))
# all files are used by others
if numUsedFiles == 0 and skipInputList != []:
continue
# remove skipped files
strSkipped = ''
for tmpFile in skipInputList:
strSkipped += '%s,' % tmpFile.lfn
job.Files.remove(tmpFile)
strSkipped = strSkipped[:-1]
# modify jobpar
if strSkipped != '':
optionToSkipFiles = '--skipInputByRetry'
if not optionToSkipFiles in job.jobParameters:
# just append
job.jobParameters += "%s=%s " % (optionToSkipFiles,
strSkipped)
else:
# extract already skipped files
tmpMatch = re.search("(%s=[^ ]+)", job.jobParameters)
if tmpMatch == None:
tmpLog.error("Failed to extract arg of %s for PandaID=%s" \
% (optionToSkipFiles,job.PandaID))
return
# replace
job.jobParameters = re.sub(
tmpMatch.group(1),
"%s,%s" % (tmpMatch.group(1), optionToSkipFiles),
job.jobParameters)
if self.verbose:
tmpLog.debug(job.jobParameters)
# append
retryJobs.append(job)
# all input files were or are being used by other jobs
if retryJobs == []:
tmpLog.info(
'All input files were or are being used by other jobs for the same output. No jobs to be retried. If you need to ignore duplication check (e.g., using the same EVNT file for multiple simulation subjobs), set ignoreDuplication=True. i.e. retry(123,ignoreDuplication=True)'
)
return
# check voms role
if not retryJobs[0].workingGroup in ['NULL', None, '']:
# VOMS role was used
if not "--workingGroup" in job.metadata:
# extract voms roles from metadata
match = re.search("--voms( |=)[ \"]*([^ \"]+)",
job.metadata)
if match != None:
vomsRoles = match.group(2)
else:
vomsRoles = "atlas:/atlas/%s/Role=production" % retryJobs[
0].workingGroup
# regenerate proxy with VOMS roles
try:
tmpLog.info("Checking proxy role to resubmit %s jobs" %
retryJobs[0].workingGroup)
self.gridPassPhrase, self.vomsFQAN = PsubUtils.checkGridProxy(
self.gridPassPhrase, False, self.verbose,
vomsRoles)
except:
tmpLog.error("Failed to generate a proxy with %s" %
vomsRoles)
return
# check runtime env for new site submission
if newSite:
if retryJobs[
0].processingType == 'pathena' or '--useAthenaPackages' in retryJobs[
0].metadata:
from pandatools import AthenaUtils
stA, retA = AthenaUtils.getAthenaVer()
if not stA:
tmpLog.error(
"Failed to get Athena rel/cache version in current runtime env"
)
return
athenaVer = retA['athenaVer']
cacheVer = retA['cacheVer']
nightVer = retA['nightVer']
wrongSetup = False
if retryJobs[0].AtlasRelease != 'Atlas-%s' % athenaVer:
wrongSetup = True
errMsg = "Current Athena version Atlas-%s is inconsitent with the previous submission %s. " % (
athenaVer, retryJobs[0].AtlasRelease)
elif retryJobs[
0].homepackage != 'AnalysisTransforms' + cacheVer + nightVer:
wrongSetup = True
errMsg = "Current cache version %s is inconsitent with the previous submission. " % cacheVer.replace(
'-', '').replace('_', '-')
if wrongSetup:
errMsg += 'You need to have the same runtime env as before since all job spec need to be re-created to send jobs to a new site. '
errMsg += 'Please setup Athena correctly and restart pbook'
tmpLog.error(errMsg)
return
# test mode
if noSubmit:
continue
# invoke pathena/prun to send job to new site
if newSite:
tmpLog.info(
"Constrcuting job spec again to be sent to another site ..."
)
comStat = PsubUtils.execWithModifiedParams(
retryJobs, newOpts, self.verbose)
if comStat == 0:
# update database
time.sleep(2)
self.sync()
else:
tmpLog.error("Failed to submit jobs to Panda server")
return
# register datasets
tmpOutDsLocation = Client.PandaSites[
retryJobs[-1].computingSite]['ddm']
addedDataset = []
shadowDSname = None
for tmpFile in retryJobs[-1].Files:
if tmpFile.type in ['output', 'log'
] and tmpFile.dataset.endswith('/'):
# add shadow
if shadowDSname == None and tmpFile.type == 'log':
shadowDSname = "%s%s" % (tmpFile.destinationDBlock,
Client.suffixShadow)
Client.addDataset(shadowDSname, self.verbose)
# add datasets
if not tmpFile.destinationDBlock in addedDataset:
# create dataset
Client.addDataset(tmpFile.destinationDBlock,
self.verbose,
location=tmpOutDsLocation)
# add to container
Client.addDatasetsToContainer(
tmpFile.dataset, [tmpFile.destinationDBlock],
self.verbose)
# append
addedDataset.append(tmpFile.destinationDBlock)
# submit
tmpLog.info("Submitting job ...")
status, out = Client.submitJobs(retryJobs, verbose=self.verbose)
if out == None or status != 0:
tmpLog.error(status)
tmpLog.error(out)
tmpLog.error("Failed to submit jobs to Panda server")
return
# update database
pandaIDstatus = {}
newJobID = None
for items in out:
# get newJobID
if newJobID == None:
newJobID = items[1]
# check PandaID
PandaID = items[0]
if PandaID == 'NULL':
tmpLog.error(
"Panda server returned wrong IDs. It may have a temporary problem"
)
return
# set newJobsetID
if newJobsetID in [None, -1]:
newJobsetID = items[2]['jobsetID']
# dummy statuso
pandaIDstatus[PandaID] = ('defined', 'NULL')
# set retry ID
localJob.retryID = newJobID
if not newJobsetID in [None, -1, 'NULL']:
localJob.retryJobsetID = newJobsetID
try:
PdbUtils.updateJobDB(localJob, self.verbose)
except:
tmpLog.error("Failed to set retryID for JobID=%s" % JobID)
return
# set new paramers
newLocalJob = PdbUtils.convertPtoD(retryJobs, pandaIDstatus)
newLocalJob.JobID = newJobID
if not newJobsetID in [None, -1, 'NULL']:
newLocalJob.groupID = newJobsetID
newLocalJob.creationTime = datetime.datetime.utcnow()
# insert to DB
try:
PdbUtils.insertJobDB(newLocalJob, self.verbose)
except:
tmpLog.error("Failed to insert JobID=%s to local repository" %
newJobID)
return
# write new jobdefID
PsubUtils.writeJobDefID(newJobID)
# done
tmpMsg = 'Done. New JobID=%s' % newJobID
if not newJobsetID in [None, -1, 'NULL']:
tmpMsg += " JobsetID=%s" % newJobsetID
tmpLog.info(tmpMsg)
def sync(self):
# get logger
tmpLog = PLogger.getPandaLogger()
tmpLog.info("Synchronizing local repository ...")
# check proxy
self.gridPassPhrase, self.vomsFQAN = PsubUtils.checkGridProxy(
self.gridPassPhrase, False, self.verbose)
# get JobIDs in local repository
localJobIDs = PdbUtils.getListOfJobIDs()
# get recent JobIDs from panda server
syncTime = datetime.datetime.utcnow().strftime('%Y-%m-%d %H:%M:%S')
# set sync time for the first attempt
bookConf = BookConfig.getConfig()
if self.restoreDB:
# reset last_synctime to restore database
bookConf.last_synctime = ''
# disable
self.restoreDB = False
tmpLog.info(
"It may take several minutes to restore local repository ...")
if bookConf.last_synctime == '':
bookConf.last_synctime = datetime.datetime.utcnow(
) - datetime.timedelta(days=180)
bookConf.last_synctime = bookConf.last_synctime.strftime(
'%Y-%m-%d %H:%M:%S')
status, remoteJobIDs = Client.getJobIDsInTimeRange(
bookConf.last_synctime, verbose=self.verbose)
if status != 0:
tmpLog.error("Failed to get JobIDs from panda server")
return
tmpLog.info("Got %s jobs to be updated" % len(remoteJobIDs))
# insert if missing
for remoteJobID in remoteJobIDs:
# check local status
job = None
if remoteJobID in localJobIDs:
# get job info from local repository
job = PdbUtils.readJobDB(remoteJobID, self.verbose)
# skip if frozen
if job.dbStatus == 'frozen':
continue
tmpLog.info("Updating JobID=%s ..." % remoteJobID)
# get PandaIDs
status, pandaIDstatus = Client.getPandIDsWithJobID(
remoteJobID, verbose=self.verbose)
if status != 0:
tmpLog.error("Failed to get PandaIDs for %s" % remoteJobID)
return
pandaIDs = pandaIDstatus.keys()
pandaIDs.sort()
# get full JobSpec
pandaJobs = []
pandaFileInfo = {}
pandaJobForSiteID = None
if job == None:
tmpIDs = [pandaIDs[0], pandaIDs[-1]]
status, pandaJobs = Client.getFullJobStatus(
tmpIDs, verbose=self.verbose)
if status != 0:
tmpLog.error("Failed to get PandaJobs for %s" %
remoteJobID)
return
# get slimmed file info
status, pandaFileInfo = Client.getSlimmedFileInfoPandaIDs(
pandaIDs, verbose=self.verbose)
if status != 0:
tmpLog.error("Failed to get file info for %s" %
remoteJobID)
return
else:
# get one job to set computingSite which may have changed due to rebrokerage
status, tmpPandaJobs = Client.getFullJobStatus(
[pandaIDs[0]], verbose=self.verbose)
if status != 0:
tmpLog.error("Failed to get PandaJobs for %s" %
remoteJobID)
return
pandaJobForSiteID = tmpPandaJobs[0]
# convert to local job spec
localJob = PdbUtils.convertPtoD(pandaJobs, pandaIDstatus, job,
pandaFileInfo, pandaJobForSiteID)
# update database
if not remoteJobID in localJobIDs:
# insert to DB
try:
PdbUtils.insertJobDB(localJob, self.verbose)
except:
tmpLog.error(
"Failed to insert JobID=%s to local repository" %
remoteJobID)
return
# set retryID
if not localJob.provenanceID in [0, '0']:
try:
PdbUtils.setRetryID(localJob, self.verbose)
except:
tmpLog.error(
"Failed to set retryID for JobID=%s in local repository"
% remoteJobID)
return
else:
# update
try:
PdbUtils.updateJobDB(localJob, self.verbose)
except:
tmpLog.error(
"Failed to update local repository for JobID=%s" %
remoteJobID)
return
# update sync time
bookConf = BookConfig.getConfig()
bookConf.last_synctime = syncTime
BookConfig.updateConfig(bookConf)
tmpLog.info("Synchronization Completed")
