def testGrouper(self):
"""
Test the grouper function (returns chunk of an iterable)
"""
listChunks = [i for i in grouper(list(range(0, 7)), 3)] # Want list(range) for python 3
iterChunks = [i for i in grouper(xrange(0, 7), 3)] # xrange becomes range in python 3
for a, b in itertools.izip_longest(listChunks, iterChunks):
self.assertEqual(a, b)
self.assertEqual(listChunks[-1], [6])
def testGrouper(self):
"""
Test the grouper function (returns chunk of an iterable)
"""
listChunks = [i for i in grouper(list(range(0, 7)), 3)
] # Want list(range) for python 3
iterChunks = [i for i in grouper(xrange(0, 7), 3)
] # xrange becomes range in python 3
for a, b in itertools.izip_longest(listChunks, iterChunks):
self.assertEqual(a, b)
self.assertEqual(listChunks[-1], [6])
def _getLumiList(self, blockName=None, lfns=None, validFileOnly=1):
"""
currently only take one lfn but dbs api need be updated
"""
try:
if blockName:
lumiLists = self.dbs.listFileLumis(block_name=blockName, validFileOnly=validFileOnly)
elif lfns:
lumiLists = []
for slfn in grouper(lfns, 50):
lumiLists.extend(self.dbs.listFileLumiArray(logical_file_name = slfn))
else:
# shouldn't call this with both blockName and lfns empty
# but still returns empty dict for that case
return {}
except dbsClientException as ex:
msg = "Error in "
msg += "DBSReader.listFileLumiArray(%s)\n" % lfns
msg += "%s\n" % formatEx3(ex)
raise DBSReaderError(msg)
lumiDict = {}
for lumisItem in lumiLists:
lumiDict.setdefault(lumisItem['logical_file_name'], [])
item = {}
item["RunNumber"] = lumisItem['run_num']
item['LumiSectionNumber'] = lumisItem['lumi_section_num']
lumiDict[lumisItem['logical_file_name']].append(item)
return lumiDict
def _getLumiList(self, blockName=None, lfns=None, validFileOnly=1):
"""
currently only take one lfn but dbs api need be updated
"""
try:
if blockName:
lumiLists = self.dbs.listFileLumis(block_name=blockName,
validFileOnly=validFileOnly)
elif lfns:
lumiLists = []
for slfn in grouper(lfns, 50):
lumiLists.extend(
self.dbs.listFileLumiArray(logical_file_name=slfn))
except dbsClientException as ex:
msg = "Error in "
msg += "DBSReader.listFileLumiArray(%s)\n" % lfns
msg += "%s\n" % formatEx3(ex)
raise DBSReaderError(msg)
lumiDict = {}
for lumisItem in lumiLists:
lumiDict.setdefault(lumisItem['logical_file_name'], [])
item = {}
item["RunNumber"] = lumisItem['run_num']
item['LumiSectionNumber'] = lumisItem['lumi_section_num']
lumiDict[lumisItem['logical_file_name']].append(item)
return lumiDict
def algorithm(self, parameters):
"""
get information from wmbs, workqueue and local couch
"""
try:
data = self.fwjrAPI.getFWJRByArchiveStatus('ready', limit=self.numDocsRetrievePerPolling)['rows']
logging.info("Found %i not archived documents from FWRJ db to upload to WMArchive.", len(data))
for slicedData in grouper(data, self.numDocsUploadPerCall):
jobIDs = []
archiveDocs = []
for job in slicedData:
doc = createArchiverDoc(job)
archiveDocs.append(doc)
jobIDs.append(job["id"])
response = self.wmarchiver.archiveData(archiveDocs)
# Partial success is not allowed either all the insert is successful or none is
if response[0]['status'] == "ok" and len(response[0]['ids']) == len(jobIDs):
archiveIDs = response[0]['ids']
for docID in jobIDs:
self.fwjrAPI.updateArchiveUploadedStatus(docID)
logging.info("...successfully uploaded %d docs", len(jobIDs))
logging.debug("JobIDs uploaded: %s", jobIDs)
logging.debug("Archived IDs returned: %s", archiveIDs)
else:
logging.warning("Upload failed and it will be retried in the next cycle: %s: %s.",
response[0]['status'], response[0]['reason'])
logging.debug("failed JobIds %s", jobIDs)
except Exception as ex:
logging.error("Error occurred, will retry later:")
logging.error(str(ex))
logging.error("Trace back: \n%s", traceback.format_exc())
def listFilesInBlockWithParents(self, fileBlockName, lumis = True, validFileOnly = 1):
"""
_listFilesInBlockWithParents_
Get a list of files in the named fileblock including
the parents of that file.
TODO: lumis can be false when lumi splitting is not required
However WMBSHelper expect file['LumiList'] to get the run number
so for now it will be always true.
"""
if not self.blockExists(fileBlockName):
msg = "DBSReader.listFilesInBlockWithParents(%s): No matching data"
raise DBSReaderError(msg % fileBlockName)
try:
#TODO: shoud we get only valid block for this?
files = self.dbs.listFileParents(block_name = fileBlockName)
fileDetails = self.listFilesInBlock(fileBlockName, lumis, validFileOnly)
except dbsClientException as ex:
msg = "Error in "
msg += "DBSReader.listFilesInBlockWithParents(%s)\n" % (
fileBlockName,)
msg += "%s\n" % formatEx3(ex)
raise DBSReaderError(msg)
childByParents = defaultdict(list)
for f in files:
# Probably a child can have more than 1 parent file
for fp in f['parent_logical_file_name']:
childByParents[fp].append(f['logical_file_name'])
parentsLFNs = childByParents.keys()
parentFilesDetail = []
#TODO: slicing parentLFNs util DBS api is handling that.
#Remove slicing if DBS api handles
for pLFNs in grouper(parentsLFNs, 50):
parentFilesDetail.extend(self.dbs.listFileArray(logical_file_name = pLFNs, detail = True))
if lumis:
parentLumis = self._getLumiList(lfns = parentsLFNs)
parentsByLFN = defaultdict(list)
for pf in parentFilesDetail:
parentLFN = pf['logical_file_name']
dbsFile = remapDBS3Keys(pf, stringify = True)
if lumis:
dbsFile["LumiList"] = parentLumis[parentLFN]
for childLFN in childByParents[parentLFN]:
parentsByLFN[childLFN].append(dbsFile)
for fileInfo in fileDetails:
fileInfo["ParentList"] = parentsByLFN[fileInfo['logical_file_name']]
return fileDetails
def getClassAds(self):
"""
_getClassAds_
Grab CONDOR classAds using CONDOR-PYTHON
This looks at the schedd running on the
Submit-Host and edit/remove jobs
"""
jobInfo = {}
schedd = condor.Schedd()
try:
logging.debug("Start: Retrieving classAds using Condor Python XQuery")
itobj = schedd.xquery(
'WMAgent_JobID =!= "UNDEFINED" && WMAgent_AgentName == %s' % classad.quote(str(self.agent)),
["JobStatus", "EnteredCurrentStatus", "JobStartDate", "QDate", "DESIRED_Sites",
"ExtDESIRED_Sites", "MATCH_EXP_JOBGLIDEIN_CMSSite", "WMAgent_JobID"]
)
logging.debug("Finish: Retrieving classAds using Condor Python XQuery")
except:
msg = "Query to condor schedd failed in PyCondorPlugin"
logging.debug(msg)
return None, None
else:
for slicedAds in grouper(itobj, 1000):
for jobAd in slicedAds:
### This condition ignores jobs that are Removed, but stay in the X state
### For manual condor_rm removal, job wont be in the queue \
### and status of the jobs will be read from condor log
if jobAd["JobStatus"] == 3:
continue
else:
## For some strange race condition, schedd sometimes does not publish StartDate for a Running Job
## Get the entire classad for such a job
## Do not crash WMA, wait for next polling cycle to get all the info.
if jobAd["JobStatus"] == 2 and jobAd.get("JobStartDate") is None:
logging.debug("THIS SHOULD NOT HAPPEN. JobStartDate is MISSING from the CLASSAD.")
logging.debug("Could be caused by some race condition. Wait for the next Polling Cycle")
logging.debug("%s", str(jobAd))
continue
tmpDict = {}
tmpDict["JobStatus"] = int(jobAd.get("JobStatus", 100))
tmpDict["stateTime"] = int(jobAd["EnteredCurrentStatus"])
tmpDict["runningTime"] = int(jobAd.get("JobStartDate", 0))
tmpDict["submitTime"] = int(jobAd["QDate"])
tmpDict["DESIRED_Sites"] = jobAd["DESIRED_Sites"]
tmpDict["ExtDESIRED_Sites"] = jobAd["ExtDESIRED_Sites"]
tmpDict["runningCMSSite"] = jobAd.get("MATCH_EXP_JOBGLIDEIN_CMSSite", None)
tmpDict["WMAgentID"] = int(jobAd["WMAgent_JobID"])
jobInfo[tmpDict["WMAgentID"]] = tmpDict
logging.info("Retrieved %i classAds", len(jobInfo))
return jobInfo, schedd
def submit(self, jobs, info=None):
"""
_submit_
Submit jobs for one subscription
"""
successfulJobs = []
failedJobs = []
if len(jobs) == 0:
# Then was have nothing to do
return successfulJobs, failedJobs
schedd = htcondor.Schedd()
# Submit the jobs
for jobsReady in grouper(jobs, self.jobsPerSubmit):
cluster_ad = self.getClusterAd()
proc_ads = self.getProcAds(jobsReady)
logging.debug(
"Start: Submitting %d jobs using Condor Python SubmitMany" %
len(proc_ads))
try:
clusterId = schedd.submitMany(cluster_ad, proc_ads)
except Exception as ex:
logging.error("SimpleCondorPlugin job submission failed.")
logging.error(
"Moving on the the next batch of jobs and/or cycle....")
logging.exception(ex)
condorErrorReport = Report()
condorErrorReport.addError("JobSubmit", 61202, "CondorError",
str(ex))
for job in jobsReady:
job['fwjr'] = condorErrorReport
failedJobs.append(job)
else:
logging.debug(
"Finish: Submitting jobs using Condor Python SubmitMany")
for index, job in enumerate(jobsReady):
job['gridid'] = "%s.%s" % (clusterId, index)
job['status'] = '1'
successfulJobs.append(job)
# We must return a list of jobs successfully submitted and a list of jobs failed
logging.info(
"Done submitting jobs for this cycle in SimpleCondorPlugin")
return successfulJobs, failedJobs
def createDirectories(dirList):
"""
Create the directory if everything is sane
"""
for sdirList in grouper(dirList, 500):
cmdArgs = ['mkdir']
cmdArgs.extend(sdirList)
pipe = Popen(cmdArgs, stdout=PIPE, stderr=PIPE, shell=False)
stdout, stderr = pipe.communicate()
if not stderr == "":
msg = "Error in making directories: %s\n" % stderr
logging.error(msg)
logging.debug("Executing command %s\n" % cmdArgs)
raise CreateWorkAreaException(msg)
return
def createDirectories(dirList):
"""
Create the directory if everything is sane
"""
for sdirList in grouper(dirList, 500):
cmdArgs = ['mkdir']
cmdArgs.extend(sdirList)
pipe = Popen(cmdArgs, stdout = PIPE, stderr = PIPE, shell = False)
stdout, stderr = pipe.communicate()
if not stderr == "":
msg = "Error in making directories: %s\n" % stderr
logging.error(msg)
logging.debug("Executing command %s\n" % cmdArgs)
raise CreateWorkAreaException(msg)
return
def submit(self, jobs, info=None):
"""
_submit_
Submit jobs for one subscription
"""
successfulJobs = []
failedJobs = []
if len(jobs) == 0:
# Then was have nothing to do
return successfulJobs, failedJobs
schedd = htcondor.Schedd()
# Submit the jobs
for jobsReady in grouper(jobs, self.jobsPerSubmit):
cluster_ad = self.getClusterAd()
proc_ads = self.getProcAds(jobsReady)
logging.debug("Start: Submitting %d jobs using Condor Python SubmitMany" % len(proc_ads))
try:
clusterId = schedd.submitMany(cluster_ad, proc_ads)
except Exception as ex:
logging.error("SimpleCondorPlugin job submission failed.")
logging.error("Moving on the the next batch of jobs and/or cycle....")
logging.exception(ex)
condorErrorReport = Report()
condorErrorReport.addError("JobSubmit", 61202, "CondorError", str(ex))
for job in jobsReady:
job['fwjr'] = condorErrorReport
failedJobs.append(job)
else:
logging.debug("Finish: Submitting jobs using Condor Python SubmitMany")
for index,job in enumerate(jobsReady):
job['gridid'] = "%s.%s" % (clusterId, index)
job['status'] = 'Idle'
successfulJobs.append(job)
# We must return a list of jobs successfully submitted and a list of jobs failed
logging.info("Done submitting jobs for this cycle in SimpleCondorPlugin")
return successfulJobs, failedJobs
def archiveJobs(self):
"""
_archiveJobs_
archiveJobs will handle the master task of looking for finished jobs,
and running the code that cleans them out.
"""
doneList = self.findFinishedJobs()
logging.info("Found %i finished jobs to archive", len(doneList))
jobCounter = 0
for slicedList in grouper(doneList, 10000):
self.cleanWorkArea(slicedList)
successList = []
failList = []
killList = []
for job in slicedList:
if job["outcome"] == "success":
successList.append(job)
elif job["outcome"] == "killed":
killList.append(job)
else:
failList.append(job)
if self.uploadPublishInfo:
self.createAndUploadPublish(successList)
myThread = threading.currentThread()
myThread.transaction.begin()
self.changeState.propagate(successList, "cleanout", "success")
self.changeState.propagate(failList, "cleanout", "exhausted")
self.changeState.propagate(killList, "cleanout", "killed")
myThread.transaction.commit()
jobCounter += len(slicedList)
logging.info("Successfully archived %d jobs out of %d.",
jobCounter, len(doneList))
def algorithm(self, parameters):
"""
get information from wmbs, workqueue and local couch
"""
try:
logging.info("Getting not archived data info from FWRJ db...")
data = self.fwjrAPI.getFWJRByArchiveStatus('ready', limit=1000)['rows']
for slicedData in grouper(data, self.numDocsUploadPerCall):
jobIDs = []
archiverDocs = []
for job in slicedData:
doc = createArchiverDoc(job)
archiverDocs.append(doc)
jobIDs.append(job["id"])
response = self.wmarchiver.archiveData(archiverDocs)
# Partial success is not allowed either all the insert is successful of none is successful.
if response[0]['status'] == "ok" and len(response[0]['ids']) == len(jobIDs):
archiveIDs = response[0]['ids']
for docID in jobIDs:
self.fwjrAPI.updateArchiveUploadedStatus(docID)
logging.info("...successfully uploaded %d docs", len(jobIDs))
logging.debug("JobIDs uploaded: %s", jobIDs)
logging.debug("Archive IDs returned: %s", response[0]['ids'])
if len(set(archiveIDs)) == len(archiveIDs):
duplicateIDs = set([x for x in archiveIDs if archiveIDs.count(x) > 1])
logging.info("There are duplicate entry %s", duplicateIDs)
else:
logging.warning("Upload failed: %s: %s", response[0]['status'], response[0]['reason'])
logging.debug("failed JobIds %s", jobIDs)
except Exception as ex:
logging.error("Error occurred, will retry later:")
logging.error(str(ex))
logging.error("Trace back: \n%s" % traceback.format_exc())
def archiveJobs(self):
"""
_archiveJobs_
archiveJobs will handle the master task of looking for finished jobs,
and running the code that cleans them out.
"""
doneList = self.findFinishedJobs()
logging.info("Found %i finished jobs to archive", len(doneList))
jobCounter = 0
for slicedList in grouper(doneList, 10000):
self.cleanWorkArea(slicedList)
successList = []
failList = []
killList = []
for job in slicedList:
if job["outcome"] == "success":
successList.append(job)
elif job["outcome"] == "killed":
killList.append(job)
else:
failList.append(job)
if self.uploadPublishInfo:
self.createAndUploadPublish(successList)
myThread = threading.currentThread()
myThread.transaction.begin()
self.changeState.propagate(successList, "cleanout", "success")
self.changeState.propagate(failList, "cleanout", "exhausted")
self.changeState.propagate(killList, "cleanout", "killed")
myThread.transaction.commit()
jobCounter += len(slicedList)
logging.info("Successfully archived %d jobs out of %d.", jobCounter, len(doneList))
def execute(self, emulator=None):
"""
_execute_
"""
scramCommand = self.step.application.setup.scramCommand
scramArch = self.step.application.setup.scramArch
cmsswVersion = self.step.application.setup.cmsswVersion
# Are we using emulators again?
if (emulator != None):
return emulator.emulate(self.step, self.job)
overrides = {}
if hasattr(self.step, 'override'):
overrides = self.step.override.dictionary_()
# Set wait to over an hour
waitTime = overrides.get(
'waitTime', 3600 + (self.step.retryDelay * self.step.retryCount))
# hardcode CERN Castor T0_CH_CERN_MSS stageout parameters
castorStageOutParams = {}
castorStageOutParams['command'] = overrides.get('command', "xrdcp")
castorStageOutParams['option'] = overrides.get('option',
"--cerncastor")
castorStageOutParams['phedex-node'] = overrides.get(
'phedex-node', "T2_CH_CERN")
castorStageOutParams['lfn-prefix'] = overrides.get(
'lfn-prefix', "root://castorcms.cern.ch//castor/cern.ch/cms")
# hardcode CERN EOS T2_CH_CERN stageout parameters
eosStageOutParams = {}
eosStageOutParams['command'] = overrides.get('command', "xrdcp")
eosStageOutParams['option'] = overrides.get('option', "")
eosStageOutParams['phedex-node'] = overrides.get(
'phedex-node', "T2_CH_CERN")
eosStageOutParams['lfn-prefix'] = overrides.get(
'lfn-prefix', "root://eoscms.cern.ch//eos/cms")
# are we using the new stageout method ?
useNewStageOutCode = False
if getattr(self.step, 'newStageout', False) or \
('newStageOut' in overrides and overrides.get('newStageOut')):
useNewStageOutCode = True
try:
castorStageOutMgr = StageOutMgr(**castorStageOutParams)
eosStageOutMgr = StageOutMgr(**eosStageOutParams)
stageInMgr = StageInMgr()
deleteMgr = DeleteMgr()
except Exception as ex:
msg = "Unable to load StageOut/Delete Impl: %s" % str(ex)
logging.error(msg)
raise WMExecutionFailure(60312, "MgrImplementationError", msg)
# prepare output tar file
taskName = self.report.getTaskName().split('/')[-1]
host = socket.gethostname().split('.')[0]
tarName = '%s-%s-%s-%i-logs.tar' % (
self.report.data.workload, taskName, host, self.job["counter"])
tarLocation = os.path.join(self.stepSpace.location, tarName)
# check if the cmsswVersion supports edmCopyUtil (min CMSSW_8_X)
result = re.match("CMSSW_([0-9]+)_([0-9]+)_([0-9]+).*", cmsswVersion)
useEdmCopyUtil = False
if result:
try:
if int(result.group(1)) >= 8:
useEdmCopyUtil = True
except ValueError:
pass
# setup Scram needed to run edmCopyUtil
if useEdmCopyUtil:
scram = Scram(
command=scramCommand,
version=cmsswVersion,
initialise=self.step.application.setup.softwareEnvironment,
directory=self.step.builder.workingDir,
architecture=scramArch,
)
logging.info("Running scram")
try:
projectOutcome = scram.project()
except Exception as ex:
msg = "Exception raised while running scram.\n"
msg += str(ex)
logging.critical("Error running SCRAM")
logging.critical(msg)
raise WMExecutionFailure(50513, "ScramSetupFailure", msg)
if projectOutcome > 0:
msg = scram.diagnostic()
logging.critical("Error running SCRAM")
logging.critical(msg)
raise WMExecutionFailure(50513, "ScramSetupFailure", msg)
runtimeOutcome = scram.runtime()
if runtimeOutcome > 0:
msg = scram.diagnostic()
logging.critical("Error running SCRAM")
logging.critical(msg)
raise WMExecutionFailure(50513, "ScramSetupFailure", msg)
# iterate through input files
localLogs = []
deleteLogArchives = []
if useEdmCopyUtil:
numberOfFilesPerCopy = 10
else:
numberOfFilesPerCopy = 1
for logs in grouper(self.job["input_files"], numberOfFilesPerCopy):
copyCommand = "env X509_USER_PROXY=%s edmCopyUtil" % os.environ.get(
'X509_USER_PROXY', None)
for log in logs:
copyCommand += " %s" % log['lfn']
copyCommand += " %s" % self.step.builder.workingDir
# give up after timeout of 1 minute per input file
signal.signal(signal.SIGALRM, alarmHandler)
signal.alarm(60 * numberOfFilesPerCopy)
filesCopied = False
try:
if useEdmCopyUtil:
logging.info("Running edmCopyUtil")
retval = scram(copyCommand)
if retval == 0:
filesCopied = True
else:
logging.info("Running stageIn")
for log in logs:
fileInfo = {"LFN": log['lfn']}
logArchive = stageInMgr(**fileInfo)
if logArchive:
filesCopied = True
except Alarm:
logging.error(
"Indefinite hang during edmCopyUtil/stageIn of logArchives"
)
except StageOutFailure:
logging.error("Unable to stageIn logArchives")
except Exception:
raise
signal.alarm(0)
if filesCopied:
for log in logs:
localLogs.append(
os.path.join(self.step.builder.workingDir,
os.path.basename(log['lfn'])))
deleteLogArchives.append(log)
self.report.addInputFile(sourceName="logArchives",
lfn=log['lfn'])
else:
logging.error("Unable to copy logArchives to local disk")
if useEdmCopyUtil:
with open('scramOutput.log', 'r') as f:
logging.error("Scram output: %s", f.read())
for log in logs:
self.report.addSkippedFile(log['lfn'], None)
# create tarfile if any logArchive copied in
if localLogs:
tarFile = tarfile.open(tarLocation, 'w:')
for log in localLogs:
path = log.split('/')
tarFile.add(name=log,
arcname=os.path.join(path[-3], path[-2], path[-1]))
os.remove(log)
tarFile.close()
else:
msg = "Unable to copy any logArchives to local disk"
logging.error(msg)
raise WMExecutionFailure(60312, "LogCollectError", msg)
# now staging out the LogCollect tarfile
logging.info("Staging out LogCollect tarfile to Castor and EOS")
now = datetime.datetime.now()
lfn = "/store/logs/prod/%i/%.2i/%s/%s/%s" % (
now.year, now.month, "WMAgent", self.report.data.workload,
os.path.basename(tarLocation))
tarInfo = {'LFN': lfn, 'PFN': tarLocation, 'PNN': None, 'GUID': None}
# perform mandatory stage out to CERN Castor
signal.signal(signal.SIGALRM, alarmHandler)
signal.alarm(waitTime)
try:
castorStageOutMgr(tarInfo)
except Alarm:
msg = "Indefinite hang during stageOut of LogCollect to Castor"
logging.error(msg)
raise WMExecutionFailure(60409, "LogCollectTimeout", msg)
except Exception as ex:
msg = "Unable to stageOut LogCollect to Castor:\n"
msg += str(ex)
logging.error(msg)
raise WMExecutionFailure(60408, "LogCollectStageOutError", msg)
signal.alarm(0)
# add to job report
self.report.addOutputFile(outputModule="LogCollect", file=tarInfo)
outputRef = getattr(self.report.data, self.stepName)
outputRef.output.pfn = tarInfo['PFN']
outputRef.output.location = tarInfo['PNN']
outputRef.output.lfn = tarInfo['LFN']
tarInfo = {'LFN': lfn, 'PFN': tarLocation, 'PNN': None, 'GUID': None}
# then, perform best effort stage out to CERN EOS
signal.signal(signal.SIGALRM, alarmHandler)
signal.alarm(waitTime)
try:
eosStageOutMgr(tarInfo)
except Alarm:
logging.error(
"Indefinite hang during stageOut of LogCollect to EOS")
except Exception as ex:
logging.error("Unable to stageOut LogCollect to EOS:\n", ex)
signal.alarm(0)
# we got this far, delete input
for log in deleteLogArchives:
# give up after timeout of 1 minutes
signal.signal(signal.SIGALRM, alarmHandler)
signal.alarm(60)
try:
fileToDelete = {
'LFN': log['lfn'],
'PFN': None,
'PNN': None,
'StageOutCommand': None
}
deleteMgr(fileToDelete=fileToDelete)
except Alarm:
logging.error("Indefinite hang during delete of logArchive")
except Exception as ex:
logging.error("Unable to delete logArchive: %s", ex)
signal.alarm(0)
return
def execute(self, emulator = None):
"""
_execute_
"""
scramCommand = self.step.application.setup.scramCommand
scramArch = self.step.application.setup.scramArch
cmsswVersion = self.step.application.setup.cmsswVersion
# Are we using emulators again?
if (emulator != None):
return emulator.emulate( self.step, self.job )
overrides = {}
if hasattr(self.step, 'override'):
overrides = self.step.override.dictionary_()
# Set wait to over an hour
waitTime = overrides.get('waitTime', 3600 + (self.step.retryDelay * self.step.retryCount))
# Pull out StageOutMgr Overrides
# switch between old stageOut behavior and new, fancy stage out behavior
useNewStageOutCode = False
if 'newStageOut' in overrides and overrides.get('newStageOut'):
useNewStageOutCode = True
# hardcode CERN Castor T0_CH_CERN_MSS stageout parameters
castorStageOutParams = {}
castorStageOutParams['command'] = overrides.get('command', "srmv2-lcg")
castorStageOutParams['option'] = overrides.get('option', "")
castorStageOutParams['se-name'] = overrides.get('se-name', "srm-cms.cern.ch")
castorStageOutParams['phedex-node'] = overrides.get('phedex-node', "T2_CH_CERN")
castorStageOutParams['lfn-prefix'] = overrides.get('lfn-prefix', "srm://srm-cms.cern.ch:8443/srm/managerv2?SFN=/castor/cern.ch/cms")
# hardcode CERN EOS T2_CH_CERN stageout parameters
eosStageOutParams = {}
eosStageOutParams['command'] = overrides.get('command', "srmv2-lcg")
eosStageOutParams['option'] = overrides.get('option', "")
eosStageOutParams['se-name'] = overrides.get('se-name', "srm-eoscms.cern.ch")
eosStageOutParams['phedex-node'] = overrides.get('phedex-node', "T2_CH_CERN")
eosStageOutParams['lfn-prefix'] = overrides.get('lfn-prefix', "srm://srm-eoscms.cern.ch:8443/srm/v2/server?SFN=/eos/cms")
# are we using the new stageout method ?
useNewStageOutCode = False
if getattr(self.step, 'newStageout', False) or \
('newStageOut' in overrides and overrides.get('newStageOut')):
useNewStageOutCode = True
try:
if useNewStageOutCode:
# is this even working ???
#logging.info("LOGCOLLECT IS USING NEW STAGEOUT CODE")
#stageOutMgr = StageOutMgr(retryPauseTime = self.step.retryDelay,
# numberOfRetries = self.step.retryCount,
# **overrides)
#stageInMgr = StageInMgr(retryPauseTime = 0,
# numberOfRetries = 0)
#deleteMgr = DeleteMgr(retryPauseTime = 0,
# numberOfRetries = 0)
castorStageOutMgr = StageOutMgr(**castorStageOutParams)
eosStageOutMgr = StageOutMgr(**eosStageOutParams)
stageInMgr = StageInMgr()
deleteMgr = DeleteMgr()
else:
castorStageOutMgr = StageOutMgr(**castorStageOutParams)
eosStageOutMgr = StageOutMgr(**eosStageOutParams)
stageInMgr = StageInMgr()
deleteMgr = DeleteMgr()
except Exception as ex:
msg = "Unable to load StageOut/Delete Impl: %s" % str(ex)
logging.error(msg)
raise WMExecutionFailure(60312, "MgrImplementationError", msg)
# prepare output tar file
taskName = self.report.getTaskName().split('/')[-1]
host = socket.gethostname().split('.')[0]
tarName = '%s-%s-%s-%i-logs.tar' % (self.report.data.workload, taskName, host , self.job["counter"])
tarLocation = os.path.join(self.stepSpace.location, tarName)
# check if the cmsswVersion supports edmCopyUtil (min CMSSW_8_X)
result = re.match("CMSSW_([0-9]+)_([0-9]+)_([0-9]+).*", cmsswVersion)
useEdmCopyUtil = False
if result:
try:
if int(result.group(1)) >= 8:
useEdmCopyUtil = True
except ValueError:
pass
# setup Scram needed to run edmCopyUtil
if useEdmCopyUtil:
scram = Scram(
command=scramCommand,
version=cmsswVersion,
initialise=self.step.application.setup.softwareEnvironment,
directory=self.step.builder.workingDir,
architecture=scramArch,
)
logging.info("Running scram")
try:
projectOutcome = scram.project()
except Exception as ex:
msg = "Exception raised while running scram.\n"
msg += str(ex)
logging.critical("Error running SCRAM")
logging.critical(msg)
raise WMExecutionFailure(50513, "ScramSetupFailure", msg)
if projectOutcome > 0:
msg = scram.diagnostic()
logging.critical("Error running SCRAM")
logging.critical(msg)
raise WMExecutionFailure(50513, "ScramSetupFailure", msg)
runtimeOutcome = scram.runtime()
if runtimeOutcome > 0:
msg = scram.diagnostic()
logging.critical("Error running SCRAM")
logging.critical(msg)
raise WMExecutionFailure(50513, "ScramSetupFailure", msg)
# iterate through input files
localLogs = []
deleteLogArchives = []
if useEdmCopyUtil:
numberOfFilesPerCopy = 10
else:
numberOfFilesPerCopy = 1
for logs in grouper(self.job["input_files"], numberOfFilesPerCopy):
copyCommand = "env X509_USER_PROXY=%s edmCopyUtil" % os.environ.get('X509_USER_PROXY', None)
for log in logs:
copyCommand += " %s" % log['lfn']
copyCommand += " %s" % self.step.builder.workingDir
# give up after timeout of 1 minute per input file
signal.signal(signal.SIGALRM, alarmHandler)
signal.alarm(60 * numberOfFilesPerCopy)
filesCopied = False
try:
if useEdmCopyUtil:
logging.info("Running edmCopyUtil")
retval = scram(copyCommand)
if retval == 0:
filesCopied = True
else:
logging.info("Running stageIn")
for log in logs:
fileInfo = {"LFN": log['lfn']}
logArchive = stageInMgr(**fileInfo)
if logArchive:
filesCopied = True
except Alarm:
logging.error("Indefinite hang during edmCopyUtil/stageIn of logArchives")
except StageOutFailure:
logging.error("Unable to stageIn logArchives")
except Exception:
raise
signal.alarm(0)
if filesCopied:
for log in logs:
localLogs.append(os.path.join(self.step.builder.workingDir, os.path.basename(log['lfn'])))
deleteLogArchives.append(log)
self.report.addInputFile(sourceName = "logArchives", lfn = log['lfn'])
else:
logging.error("Unable to copy logArchives to local disk")
if useEdmCopyUtil:
with open('scramOutput.log', 'r') as f:
logging.error("Scram output: %s", f.read())
for log in logs:
self.report.addSkippedFile(log['lfn'], None)
# create tarfile if any logArchive copied in
if localLogs:
tarFile = tarfile.open(tarLocation, 'w:')
for log in localLogs:
path = log.split('/')
tarFile.add(name = log,
arcname = os.path.join(path[-3],
path[-2],
path[-1]))
os.remove(log)
tarFile.close()
else:
msg = "Unable to copy any logArchives to local disk"
logging.error(msg)
raise WMExecutionFailure(60312, "LogCollectError", msg)
# now staging out the LogCollect tarfile
logging.info("Staging out LogCollect tarfile to Castor and EOS")
now = datetime.datetime.now()
lfn = "/store/logs/prod/%i/%.2i/%s/%s/%s" % (now.year, now.month, "WMAgent",
self.report.data.workload,
os.path.basename(tarLocation))
tarInfo = {'LFN' : lfn,
'PFN' : tarLocation,
'SEName' : None,
'PNN' : None,
'GUID' : None}
# perform mandatory stage out to CERN Castor
signal.signal(signal.SIGALRM, alarmHandler)
signal.alarm(waitTime)
try:
castorStageOutMgr(tarInfo)
except Alarm:
msg = "Indefinite hang during stageOut of LogCollect to Castor"
logging.error(msg)
raise WMExecutionFailure(60409, "LogCollectTimeout", msg)
except Exception as ex:
msg = "Unable to stageOut LogCollect to Castor:\n"
msg += str(ex)
logging.error(msg)
raise WMExecutionFailure(60408, "LogCollectStageOutError", msg)
signal.alarm(0)
# add to job report
self.report.addOutputFile(outputModule = "LogCollect", file = tarInfo)
outputRef = getattr(self.report.data, self.stepName)
outputRef.output.pfn = tarInfo['PFN']
outputRef.output.location = tarInfo['PNN']
outputRef.output.lfn = tarInfo['LFN']
tarInfo = {'LFN' : lfn,
'PFN' : tarLocation,
'SEName' : None,
'PNN' : None,
'GUID' : None}
# then, perform best effort stage out to CERN EOS
signal.signal(signal.SIGALRM, alarmHandler)
signal.alarm(waitTime)
try:
eosStageOutMgr(tarInfo)
except Alarm:
logging.error("Indefinite hang during stageOut of LogCollect to EOS")
except Exception as ex:
logging.error("Unable to stageOut LogCollect to EOS:\n", ex)
signal.alarm(0)
# we got this far, delete input
for log in deleteLogArchives:
# give up after timeout of 1 minutes
signal.signal(signal.SIGALRM, alarmHandler)
signal.alarm(60)
try:
fileToDelete = {'LFN': log['lfn'],
'PFN': None,
'SEName': None,
'PNN': None,
'StageOutCommand': None}
deleteMgr(fileToDelete = fileToDelete)
except Alarm:
logging.error("Indefinite hang during delete of logArchive")
except Exception as ex:
logging.error("Unable to delete logArchive: %s", ex)
signal.alarm(0)
return
def submit(self, jobs, info=None):
"""
_submit_
Submit jobs for one subscription
"""
# If we're here, then we have submitter components
self.scriptFile = self.config.JobSubmitter.submitScript
self.submitDir = self.config.JobSubmitter.submitDir
timeout = getattr(self.config.JobSubmitter, 'getTimeout', 400)
successfulJobs = []
failedJobs = []
jdlFiles = []
if len(jobs) == 0:
# Then was have nothing to do
return successfulJobs, failedJobs
if len(self.pool) == 0:
# Starting things up
# This is obviously a submit API
logging.info("Starting up PyCondorPlugin worker pool")
self.inputQueue = multiprocessing.Queue()
self.result = multiprocessing.Queue()
for x in range(self.nProcess):
p = multiprocessing.Process(target=submitWorker,
args=(self.inputQueue, self.result, timeout))
p.start()
self.pool.append(p)
if not os.path.exists(self.submitDir):
os.makedirs(self.submitDir)
# Submit the jobs
nSubmits = 0
queueError = False
for jobsReady in grouper(jobs, self.jobsPerWorker):
if queueError:
# If the queue has failed, then we must not process any more jobs this cycle.
break
idList = [x['id'] for x in jobsReady]
jdlList = self.makeSubmit(jobList=jobsReady)
if not jdlList:
# Then we got nothing
logging.error("No JDL file made!")
return {'NoResult': [0]}
jdlFile = "%s/submit_%i_%i.jdl" % (self.submitDir, os.getpid(), idList[0])
with open(jdlFile, 'w') as handle:
handle.writelines(jdlList)
jdlFiles.append(jdlFile)
# Now submit them
logging.info("About to submit %i jobs", len(jobsReady))
if self.glexecPath:
command = 'CS=`which condor_submit`; '
if self.glexecWrapScript:
command += 'export GLEXEC_ENV=`%s 2>/dev/null`; ' % self.glexecWrapScript
command += 'export GLEXEC_CLIENT_CERT=%s; ' % self.glexecProxyFile
command += 'export GLEXEC_SOURCE_PROXY=%s; ' % self.glexecProxyFile
command += 'export X509_USER_PROXY=%s; ' % self.glexecProxyFile
command += 'export GLEXEC_TARGET_PROXY=%s; ' % self.jdlProxyFile
if self.glexecUnwrapScript:
command += '%s %s -- $CS %s' % (self.glexecPath, self.glexecUnwrapScript, jdlFile)
else:
command += '%s $CS %s' % (self.glexecPath, jdlFile)
else:
command = "condor_submit %s" % jdlFile
try:
self.inputQueue.put({'command': command, 'idList': idList})
except AssertionError as ex:
msg = "Critical error: input pipeline probably closed.\n"
msg += str(ex)
msg += "Error Procedure: Something critical has happened in the worker process\n"
msg += "We will now proceed to pull all useful data from the queue (if it exists)\n"
msg += "Then refresh the worker pool\n"
logging.error(msg)
queueError = True
break
nSubmits += 1
# Now we should have sent all jobs to be submitted
# Going to do the rest of it now
for dummy in range(nSubmits):
try:
res = self.result.get(block=True, timeout=timeout)
except Queue.Empty:
# If the queue was empty go to the next submit
# Those jobs have vanished
logging.error("Queue.Empty error received!")
logging.error("This could indicate a critical condor error!")
logging.error("However, no information of any use was obtained due to process failure.")
logging.error("Either process failed, or process timed out after %s seconds.", timeout)
continue
except AssertionError as ex:
msg = "Found Assertion error while retrieving output from worker process.\n"
msg += str(ex)
msg += "This indicates something critical happened to a worker process"
msg += "We will recover what jobs we know were submitted, and resubmit the rest"
msg += "Refreshing worker pool at end of loop"
logging.error(msg)
continue
try:
dummyOut = res['stdout']
error = res['stderr']
idList = res['idList']
exitCode = res['exitCode']
except KeyError as ex:
msg = "Error in finding key from result pipe\n"
msg += "Something has gone critically wrong in the worker\n"
try:
msg += "Result: %s\n" % str(res)
except:
pass
msg += str(ex)
logging.error(msg)
continue
if not exitCode == 0:
logging.error("Condor returned non-zero. Printing out command stderr")
logging.error(error)
errorCheck, errorMsg = parseError(error=error)
logging.error("Processing failed jobs and proceeding to the next jobs.")
logging.error("Do not restart component.")
else:
errorCheck = None
if errorCheck:
self.errorCount += 1
condorErrorReport = Report()
condorErrorReport.addError("JobSubmit", 61202, "CondorError", errorMsg)
for jobID in idList:
for job in jobs:
if job.get('id', None) == jobID:
job['fwjr'] = condorErrorReport
failedJobs.append(job)
break
else:
if self.errorCount > 0:
self.errorCount -= 1
for jobID in idList:
for job in jobs:
if job.get('id', None) == jobID:
successfulJobs.append(job)
break
# If we get a lot of errors in a row it's probably time to
# report this to the operators.
if self.errorCount > self.errorThreshold:
try:
msg = "Exceeded errorThreshold while submitting to condor. Check condor status."
logging.error(msg)
logging.error("Reporting to Alert system and continuing to process jobs")
from WMCore.Alerts import API as alertAPI
preAlert, sender = alertAPI.setUpAlertsMessaging(self,
compName="BossAirPyCondorPlugin")
sendAlert = alertAPI.getSendAlert(sender=sender,
preAlert=preAlert)
sendAlert(6, msg=msg)
sender.unregister()
self.errorCount = 0
except:
# There's nothing we can really do here
pass
# Remove JDL files unless commanded otherwise
if self.deleteJDLFiles:
for f in jdlFiles:
os.remove(f)
# When we're finished, clean up the queue workers in order
# to free up memory (in the midst of the process, the forked
# memory space shouldn't be touched, so it should still be
# shared, but after this point any action by the Submitter will
# result in memory duplication).
logging.info("Purging worker pool to clean up memory")
self.close()
# We must return a list of jobs successfully submitted,
# and a list of jobs failed
logging.info("Done submitting jobs for this cycle in PyCondorPlugin")
return successfulJobs, failedJobs
def listFilesInBlockWithParents(self,
fileBlockName,
lumis=True,
validFileOnly=1):
"""
_listFilesInBlockWithParents_
Get a list of files in the named fileblock including
the parents of that file.
TODO: lumis can be false when lumi splitting is not required
However WMBSHelper expect file['LumiList'] to get the run number
so for now it will be always true.
"""
if not self.blockExists(fileBlockName):
msg = "DBSReader.listFilesInBlockWithParents(%s): No matching data"
raise DBSReaderError(msg % fileBlockName)
try:
#TODO: shoud we get only valid block for this?
files = self.dbs.listFileParents(block_name=fileBlockName)
fileDetails = self.listFilesInBlock(fileBlockName, lumis,
validFileOnly)
except dbsClientException as ex:
msg = "Error in "
msg += "DBSReader.listFilesInBlockWithParents(%s)\n" % (
fileBlockName, )
msg += "%s\n" % formatEx3(ex)
raise DBSReaderError(msg)
childByParents = defaultdict(list)
for f in files:
# Probably a child can have more than 1 parent file
for fp in f['parent_logical_file_name']:
childByParents[fp].append(f['logical_file_name'])
parentsLFNs = childByParents.keys()
parentFilesDetail = []
#TODO: slicing parentLFNs util DBS api is handling that.
#Remove slicing if DBS api handles
for pLFNs in grouper(parentsLFNs, 50):
parentFilesDetail.extend(
self.dbs.listFileArray(logical_file_name=pLFNs, detail=True))
if lumis:
parentLumis = self._getLumiList(lfns=parentsLFNs)
parentsByLFN = defaultdict(list)
for pf in parentFilesDetail:
parentLFN = pf['logical_file_name']
dbsFile = remapDBS3Keys(pf, stringify=True)
if lumis:
dbsFile["LumiList"] = parentLumis[parentLFN]
for childLFN in childByParents[parentLFN]:
parentsByLFN[childLFN].append(dbsFile)
for fileInfo in fileDetails:
fileInfo["ParentList"] = parentsByLFN[
fileInfo['logical_file_name']]
return fileDetails
def submit(self, jobs, info=None):
"""
_submit_
Submit jobs for one subscription
"""
# If we're here, then we have submitter components
self.scriptFile = self.config.JobSubmitter.submitScript
self.submitDir = self.config.JobSubmitter.submitDir
timeout = getattr(self.config.JobSubmitter, 'getTimeout', 400)
successfulJobs = []
failedJobs = []
jdlFiles = []
if len(jobs) == 0:
# Then was have nothing to do
return successfulJobs, failedJobs
if len(self.pool) == 0:
# Starting things up
# This is obviously a submit API
logging.info("Starting up PyCondorPlugin worker pool")
self.inputQueue = multiprocessing.Queue()
self.result = multiprocessing.Queue()
for x in range(self.nProcess):
p = multiprocessing.Process(target=submitWorker,
args=(self.inputQueue, self.result,
timeout))
p.start()
self.pool.append(p)
if not os.path.exists(self.submitDir):
os.makedirs(self.submitDir)
# Submit the jobs
nSubmits = 0
queueError = False
for jobsReady in grouper(jobs, self.jobsPerWorker):
if queueError:
# If the queue has failed, then we must not process any more jobs this cycle.
break
idList = [x['id'] for x in jobsReady]
jdlList = self.makeSubmit(jobList=jobsReady)
if not jdlList:
# Then we got nothing
logging.error("No JDL file made!")
return {'NoResult': [0]}
jdlFile = "%s/submit_%i_%i.jdl" % (self.submitDir, os.getpid(),
idList[0])
with open(jdlFile, 'w') as handle:
handle.writelines(jdlList)
jdlFiles.append(jdlFile)
# Now submit them
logging.info("About to submit %i jobs", len(jobsReady))
if self.glexecPath:
command = 'CS=`which condor_submit`; '
if self.glexecWrapScript:
command += 'export GLEXEC_ENV=`%s 2>/dev/null`; ' % self.glexecWrapScript
command += 'export GLEXEC_CLIENT_CERT=%s; ' % self.glexecProxyFile
command += 'export GLEXEC_SOURCE_PROXY=%s; ' % self.glexecProxyFile
command += 'export X509_USER_PROXY=%s; ' % self.glexecProxyFile
command += 'export GLEXEC_TARGET_PROXY=%s; ' % self.jdlProxyFile
if self.glexecUnwrapScript:
command += '%s %s -- $CS %s' % (
self.glexecPath, self.glexecUnwrapScript, jdlFile)
else:
command += '%s $CS %s' % (self.glexecPath, jdlFile)
else:
command = "condor_submit %s" % jdlFile
try:
self.inputQueue.put({'command': command, 'idList': idList})
except AssertionError as ex:
msg = "Critical error: input pipeline probably closed.\n"
msg += str(ex)
msg += "Error Procedure: Something critical has happened in the worker process\n"
msg += "We will now proceed to pull all useful data from the queue (if it exists)\n"
msg += "Then refresh the worker pool\n"
logging.error(msg)
queueError = True
break
nSubmits += 1
# Now we should have sent all jobs to be submitted
# Going to do the rest of it now
for dummy in range(nSubmits):
try:
res = self.result.get(block=True, timeout=timeout)
except Queue.Empty:
# If the queue was empty go to the next submit
# Those jobs have vanished
logging.error("Queue.Empty error received!")
logging.error("This could indicate a critical condor error!")
logging.error(
"However, no information of any use was obtained due to process failure."
)
logging.error(
"Either process failed, or process timed out after %s seconds.",
timeout)
continue
except AssertionError as ex:
msg = "Found Assertion error while retrieving output from worker process.\n"
msg += str(ex)
msg += "This indicates something critical happened to a worker process"
msg += "We will recover what jobs we know were submitted, and resubmit the rest"
msg += "Refreshing worker pool at end of loop"
logging.error(msg)
continue
try:
dummyOut = res['stdout']
error = res['stderr']
idList = res['idList']
exitCode = res['exitCode']
except KeyError as ex:
msg = "Error in finding key from result pipe\n"
msg += "Something has gone critically wrong in the worker\n"
try:
msg += "Result: %s\n" % str(res)
except:
pass
msg += str(ex)
logging.error(msg)
continue
if not exitCode == 0:
logging.error(
"Condor returned non-zero. Printing out command stderr")
logging.error(error)
errorCheck, errorMsg = parseError(error=error)
logging.error(
"Processing failed jobs and proceeding to the next jobs.")
logging.error("Do not restart component.")
else:
errorCheck = None
if errorCheck:
self.errorCount += 1
condorErrorReport = Report()
condorErrorReport.addError("JobSubmit", 61202, "CondorError",
errorMsg)
for jobID in idList:
for job in jobs:
if job.get('id', None) == jobID:
job['fwjr'] = condorErrorReport
failedJobs.append(job)
break
else:
if self.errorCount > 0:
self.errorCount -= 1
for jobID in idList:
for job in jobs:
if job.get('id', None) == jobID:
successfulJobs.append(job)
break
# If we get a lot of errors in a row it's probably time to
# stop the component
if self.errorCount > self.errorThreshold:
msg = "Exceeded errorThreshold while submitting to condor. Check condor status."
logging.error(msg)
raise BossAirPluginException(msg)
# Remove JDL files unless commanded otherwise
if self.deleteJDLFiles:
for f in jdlFiles:
os.remove(f)
# When we're finished, clean up the queue workers in order
# to free up memory (in the midst of the process, the forked
# memory space shouldn't be touched, so it should still be
# shared, but after this point any action by the Submitter will
# result in memory duplication).
logging.info("Purging worker pool to clean up memory")
self.close()
# We must return a list of jobs successfully submitted,
# and a list of jobs failed
logging.info("Done submitting jobs for this cycle in PyCondorPlugin")
return successfulJobs, failedJobs
