def __init__(self, config):
"""
Initialise class members
"""
BaseWorkerThread.__init__(self)
self.config = config
myThread = threading.currentThread()
self.daoFactory = DAOFactory(package = "WMCore.WMBS",
logger = myThread.logger,
dbinterface = myThread.dbi)
self.changeState = ChangeState(self.config)
self.maxRetries = self.config.ErrorHandler.maxRetries
if not isinstance(self.maxRetries, dict):
self.maxRetries = {'default' : self.maxRetries}
if 'default' not in self.maxRetries:
raise ErrorHandlerException('Max retries for the default job type must be specified')
self.maxProcessSize = getattr(self.config.ErrorHandler, 'maxProcessSize', 250)
self.exitCodes = getattr(self.config.ErrorHandler, 'failureExitCodes', [])
self.maxFailTime = getattr(self.config.ErrorHandler, 'maxFailTime', 32 * 3600)
self.readFWJR = getattr(self.config.ErrorHandler, 'readFWJR', False)
self.passCodes = getattr(self.config.ErrorHandler, 'passExitCodes', [])
self.getJobs = self.daoFactory(classname = "Jobs.GetAllJobs")
self.idLoad = self.daoFactory(classname = "Jobs.LoadFromIDWithType")
self.loadAction = self.daoFactory(classname = "Jobs.LoadForErrorHandler")
self.dataCollection = DataCollectionService(url = config.ACDC.couchurl,
database = config.ACDC.database)
return
def _getDBSBlock(self, match, wmspec):
"""Get DBS info for this block"""
blockName = match['Inputs'].keys()[0] #TODO: Allow more than one
if match['ACDC']:
acdcInfo = match['ACDC']
acdc = DataCollectionService(acdcInfo["server"], acdcInfo["database"])
collection = acdc.getDataCollection(acdcInfo['collection'])
splitedBlockName = ACDCBlock.splitBlockName(blockName)
fileLists = acdc.getChunkFiles(acdcInfo['collection'],
acdcInfo['fileset'],
splitedBlockName['Offset'],
splitedBlockName['NumOfFiles'],
user = wmspec.getOwner().get("name"),
group = wmspec.getOwner().get("group"))
block = {}
block["Files"] = fileLists
return blockName, block
else:
dbs = get_dbs(match['Dbs'])
if wmspec.getTask(match['TaskName']).parentProcessingFlag():
dbsBlockDict = dbs.getFileBlockWithParents(blockName)
else:
dbsBlockDict = dbs.getFileBlock(blockName)
return blockName, dbsBlockDict[blockName]
def setUp(self):
"""
setup for test.
"""
super(ErrorHandlerTest, self).setUp()
myThread = threading.currentThread()
self.testInit = TestInitCouchApp(__file__)
self.testInit.setLogging()
self.testInit.setDatabaseConnection()
self.testInit.setSchema(customModules=["WMCore.WMBS"],
useDefault=False)
self.testInit.setupCouch("errorhandler_t", "GroupUser", "ACDC")
self.testInit.setupCouch("errorhandler_t_jd/jobs", "JobDump")
self.testInit.setupCouch("errorhandler_t_jd/fwjrs", "FWJRDump")
self.daofactory = DAOFactory(package="WMCore.WMBS",
logger=myThread.logger,
dbinterface=myThread.dbi)
self.getJobs = self.daofactory(classname="Jobs.GetAllJobs")
self.setJobTime = self.daofactory(classname="Jobs.SetStateTime")
locationAction = self.daofactory(classname="Locations.New")
locationAction.execute(siteName="malpaquet", pnn="T2_CH_CERN")
self.testDir = self.testInit.generateWorkDir()
self.configFile = EmulatorSetup.setupWMAgentConfig()
self.nJobs = 10
self.dataCS = DataCollectionService(url=self.testInit.couchUrl,
database="errorhandler_t")
return
def algorithm(self, *args, **kwargs):
"""
_algorithm_
Split files into a number of lumis per job
Allow a flag to determine if we split files between jobs
"""
avgEventsPerJob = int(kwargs.get('events_per_job', 5000))
eventLimit = int(kwargs.get('max_events_per_lumi', 20000))
splitOnFile = bool(kwargs.get('halt_job_on_file_boundaries', True))
ignoreACDC = bool(kwargs.get('ignore_acdc_except', False))
collectionName = kwargs.get('collectionName', None)
splitOnRun = kwargs.get('splitOnRun', True)
getParents = kwargs.get('include_parents', False)
runWhitelist = kwargs.get('runWhitelist', [])
runs = kwargs.get('runs', None)
lumis = kwargs.get('lumis', None)
timePerEvent, sizePerEvent, memoryRequirement = \
self.getPerformanceParameters(kwargs.get('performance', {}))
deterministicPileup = kwargs.get('deterministicPileup', False)
eventsPerLumiInDataset = 0
if deterministicPileup and self.package == 'WMCore.WMBS':
getJobNumber = self.daoFactory(classname = "Jobs.GetNumberOfJobsPerWorkflow")
jobNumber = getJobNumber.execute(workflow = self.subscription.getWorkflow().id)
self.nJobs = jobNumber
goodRunList = {}
if runs and lumis:
goodRunList = buildLumiMask(runs, lumis)
# If we have runLumi info, we need to load it from couch
if collectionName:
try:
from WMCore.ACDC.DataCollectionService import DataCollectionService
couchURL = kwargs.get('couchURL')
couchDB = kwargs.get('couchDB')
filesetName = kwargs.get('filesetName')
collectionName = kwargs.get('collectionName')
owner = kwargs.get('owner')
group = kwargs.get('group')
logging.info('Creating jobs for ACDC fileset %s' % filesetName)
dcs = DataCollectionService(couchURL, couchDB)
goodRunList = dcs.getLumiWhitelist(collectionName, filesetName, owner, group)
except Exception, ex:
msg = "Exception while trying to load goodRunList\n"
if ignoreACDC:
msg += "Ditching goodRunList\n"
msg += str(ex)
msg += str(traceback.format_exc())
logging.error(msg)
goodRunList = {}
else:
msg += "Refusing to create any jobs.\n"
msg += str(ex)
msg += str(traceback.format_exc())
logging.error(msg)
return
def _getDBSBlock(self, match, wmspec):
"""Get DBS info for this block"""
blockName = match['Inputs'].keys()[0] #TODO: Allow more than one
if match['ACDC']:
acdcInfo = match['ACDC']
acdc = DataCollectionService(acdcInfo["server"], acdcInfo["database"])
collection = acdc.getDataCollection(acdcInfo['collection'])
splitedBlockName = ACDCBlock.splitBlockName(blockName)
fileLists = acdc.getChunkFiles(acdcInfo['collection'],
acdcInfo['fileset'],
splitedBlockName['Offset'],
splitedBlockName['NumOfFiles'],
user = wmspec.getOwner().get("name"),
group = wmspec.getOwner().get("group"))
block = {}
block["Files"] = fileLists
return blockName, block
else:
dbs = get_dbs(match['Dbs'])
if wmspec.getTask(match['TaskName']).parentProcessingFlag():
dbsBlockDict = dbs.getFileBlockWithParents(blockName)
else:
dbsBlockDict = dbs.getFileBlock(blockName)
if wmspec.locationDataSourceFlag():
blockInfo = dbsBlockDict[blockName]
seElements = []
for cmsSite in match['Inputs'].values()[0]: #TODO: Allow more than one
ses = self.SiteDB.cmsNametoSE(cmsSite)
seElements.extend(ses)
seElements = list(set(seElements))
blockInfo['StorageElements'] = seElements
return blockName, dbsBlockDict[blockName]
def validBlocks(self, task):
"""Return blocks that pass the input data restriction"""
validBlocks = []
# TODO take the chunk size from parameter
chunkSize = 200
acdcInfo = task.getInputACDC()
if not acdcInfo:
raise WorkQueueWMSpecError(
self.wmspec, 'No acdc section for %s' % task.getPathName())
acdc = DataCollectionService(acdcInfo["server"], acdcInfo["database"])
if self.data:
acdcBlockSplit = ACDCBlock.splitBlockName(self.data.keys()[0])
else:
#if self.data is not passed, assume the the data is input dataset
# from the spec
acdcBlockSplit = False
if acdcBlockSplit:
dbsBlock = {}
dbsBlock['Name'] = self.data.keys()[0]
block = acdc.getChunkInfo(
acdcInfo['collection'],
acdcBlockSplit['TaskName'],
acdcBlockSplit['Offset'],
acdcBlockSplit['NumOfFiles'],
user=self.wmspec.getOwner().get("name"),
group=self.wmspec.getOwner().get("group"))
dbsBlock['NumberOfFiles'] = block['files']
dbsBlock['NumberOfEvents'] = block['events']
dbsBlock['NumberOfLumis'] = block['lumis']
dbsBlock['ACDC'] = acdcInfo
dbsBlock["Sites"] = sitesFromStorageEelements(block["locations"])
validBlocks.append(dbsBlock)
else:
acdcBlocks = acdc.chunkFileset(
acdcInfo['collection'],
acdcInfo['fileset'],
chunkSize,
user=self.wmspec.getOwner().get("name"),
group=self.wmspec.getOwner().get("group"))
for block in acdcBlocks:
dbsBlock = {}
dbsBlock['Name'] = ACDCBlock.name(self.wmspec.name(),
acdcInfo["fileset"],
block['offset'],
block['files'])
dbsBlock['NumberOfFiles'] = block['files']
dbsBlock['NumberOfEvents'] = block['events']
dbsBlock['NumberOfLumis'] = block['lumis']
dbsBlock["Sites"] = sitesFromStorageEelements(
block["locations"])
dbsBlock['ACDC'] = acdcInfo
validBlocks.append(dbsBlock)
return validBlocks
def algorithm(self, *args, **kwargs):
"""
_algorithm_
Split files into a number of lumis per job
Allow a flag to determine if we split files between jobs
"""
myThread = threading.currentThread()
lumisPerJob = int(kwargs.get('lumis_per_job', 1))
splitOnFile = bool(kwargs.get('halt_job_on_file_boundaries', True))
ignoreACDC = bool(kwargs.get('ignore_acdc_except', False))
collectionName = kwargs.get('collectionName', None)
splitOnRun = kwargs.get('splitOnRun', True)
getParents = kwargs.get('include_parents', False)
runWhitelist = kwargs.get('runWhitelist', [])
runs = kwargs.get('runs', None)
lumis = kwargs.get('lumis', None)
timePerEvent, sizePerEvent, memoryRequirement = \
self.getPerformanceParameters(kwargs.get('performance', {}))
goodRunList = {}
if runs and lumis:
goodRunList = buildLumiMask(runs, lumis)
# If we have runLumi info, we need to load it from couch
if collectionName:
try:
from WMCore.ACDC.DataCollectionService import DataCollectionService
couchURL = kwargs.get('couchURL')
couchDB = kwargs.get('couchDB')
filesetName = kwargs.get('filesetName')
collectionName = kwargs.get('collectionName')
owner = kwargs.get('owner')
group = kwargs.get('group')
logging.info('Creating jobs for ACDC fileset %s' % filesetName)
dcs = DataCollectionService(couchURL, couchDB)
goodRunList = dcs.getLumiWhitelist(collectionName, filesetName, owner, group)
except Exception, ex:
msg = "Exception while trying to load goodRunList\n"
if ignoreACDC:
msg += "Ditching goodRunList\n"
msg += str(ex)
msg += str(traceback.format_exc())
logging.error(msg)
goodRunList = {}
else:
msg += "Refusing to create any jobs.\n"
msg += str(ex)
msg += str(traceback.format_exc())
logging.error(msg)
return
def algorithm(self, *args, **kwargs):
"""
_algorithm_
Split files into a number of lumis per job
Allow a flag to determine if we split files between jobs
"""
myThread = threading.currentThread()
lumisPerJob = int(kwargs.get('lumis_per_job', 1))
splitOnFile = bool(kwargs.get('halt_job_on_file_boundaries', True))
ignoreACDC = bool(kwargs.get('ignore_acdc_except', False))
collectionName = kwargs.get('collectionName', None)
splitOnRun = kwargs.get('splitOnRun', True)
getParents = kwargs.get('include_parents', False)
runWhitelist = kwargs.get('runWhitelist', [])
runs = kwargs.get('runs', None)
lumis = kwargs.get('lumis', None)
goodRunList = {}
if runs and lumis:
goodRunList = buildLumiMask(runs, lumis)
# If we have runLumi info, we need to load it from couch
if collectionName:
try:
from WMCore.ACDC.DataCollectionService import DataCollectionService
couchURL = kwargs.get('couchURL')
couchDB = kwargs.get('couchDB')
filesetName = kwargs.get('filesetName')
collectionName = kwargs.get('collectionName')
owner = kwargs.get('owner')
group = kwargs.get('group')
logging.info('Creating jobs for ACDC fileset %s' % filesetName)
dcs = DataCollectionService(couchURL, couchDB)
goodRunList = dcs.getLumiWhitelist(collectionName, filesetName,
owner, group)
except Exception, ex:
msg = "Exception while trying to load goodRunList\n"
if ignoreACDC:
msg += "Ditching goodRunList\n"
msg += str(ex)
msg += str(traceback.format_exc())
logging.error(msg)
goodRunList = {}
else:
msg += "Refusing to create any jobs.\n"
msg += str(ex)
msg += str(traceback.format_exc())
logging.error(msg)
return
def testFailedJobsUniqueWf(self):
"""
Performance test of failedJobs with all failed jobs belonging
to the same workflow and the same task name
"""
loadList = []
for i in range(1, 5000):
loadList.append(self.jobConfig('wf1', '/wf1/task1', i, 'lfn1'))
dcs = DataCollectionService(url=self.testInit.couchUrl, database="wmcore-acdc-datacollectionsvc")
dcs.failedJobs(loadList)
return
def testFailedJobsUniqueWf(self):
"""
Performance test of failedJobs with all failed jobs belonging
to the same workflow and the same task name
"""
loadList = []
for i in range(1, 5000):
loadList.append(self.jobConfig('wf1', '/wf1/task1', i, 'lfn1'))
dcs = DataCollectionService(url=self.testInit.couchUrl, database="wmcore-acdc-datacollectionsvc")
dcs.failedJobs(loadList)
return
def createResubmitSpec(self, serverUrl, couchDB):
"""
_createResubmitSpec_
Create a bogus resubmit workload.
"""
self.site = "cmssrm.fnal.gov"
workload = WMWorkloadHelper(WMWorkload("TestWorkload"))
reco = workload.newTask("reco")
workload.setOwnerDetails(name = "evansde77", group = "DMWM")
# first task uses the input dataset
reco.addInputDataset(primary = "PRIMARY", processed = "processed-v1", tier = "TIER1")
reco.data.input.splitting.algorithm = "File"
reco.setTaskType("Processing")
cmsRunReco = reco.makeStep("cmsRun1")
cmsRunReco.setStepType("CMSSW")
reco.applyTemplates()
cmsRunRecoHelper = cmsRunReco.getTypeHelper()
cmsRunRecoHelper.addOutputModule("outputRECO",
primaryDataset = "PRIMARY",
processedDataset = "processed-v2",
dataTier = "TIER2",
lfnBase = "/store/dunkindonuts",
mergedLFNBase = "/store/kfc")
dcs = DataCollectionService(url = serverUrl, database = couchDB)
def getJob(workload):
job = Job()
job["task"] = workload.getTask("reco").getPathName()
job["workflow"] = workload.name()
job["location"] = self.site
job["owner"] = "evansde77"
job["group"] = "DMWM"
return job
testFileA = WMFile(lfn = makeUUID(), size = 1024, events = 1024)
testFileA.setLocation([self.site])
testFileA.addRun(Run(1, 1, 2))
testFileB = WMFile(lfn = makeUUID(), size = 1024, events = 1024)
testFileB.setLocation([self.site])
testFileB.addRun(Run(1, 3, 4))
testJobA = getJob(workload)
testJobA.addFile(testFileA)
testJobA.addFile(testFileB)
dcs.failedJobs([testJobA])
topLevelTask = workload.getTopLevelTask()[0]
workload.truncate("Resubmit_TestWorkload", topLevelTask.getPathName(),
serverUrl, couchDB)
return workload
def validBlocks(self, task):
"""Return blocks that pass the input data restriction"""
validBlocks = []
# TODO take the chunk size from parameter
chunkSize = 200
acdcInfo = task.getInputACDC()
if not acdcInfo:
raise WorkQueueWMSpecError(self.wmspec, 'No acdc section for %s' % task.getPathName())
acdc = DataCollectionService(acdcInfo["server"], acdcInfo["database"])
if self.data:
acdcBlockSplit = ACDCBlock.splitBlockName(self.data.keys()[0])
else:
#if self.data is not passed, assume the the data is input dataset
# from the spec
acdcBlockSplit = False
if acdcBlockSplit:
dbsBlock = {}
dbsBlock['Name'] = self.data.keys()[0]
block = acdc.getChunkInfo(acdcInfo['collection'],
acdcBlockSplit['TaskName'],
acdcBlockSplit['Offset'],
acdcBlockSplit['NumOfFiles'],
user = self.wmspec.getOwner().get("name"),
group = self.wmspec.getOwner().get("group"))
dbsBlock['NumberOfFiles'] = block['files']
dbsBlock['NumberOfEvents'] = block['events']
dbsBlock['NumberOfLumis'] = block['lumis']
dbsBlock['ACDC'] = acdcInfo
dbsBlock["Sites"] = sitesFromStorageEelements(block["locations"])
validBlocks.append(dbsBlock)
else:
acdcBlocks = acdc.chunkFileset(acdcInfo['collection'],
acdcInfo['fileset'],
chunkSize,
user = self.wmspec.getOwner().get("name"),
group = self.wmspec.getOwner().get("group"))
for block in acdcBlocks:
dbsBlock = {}
dbsBlock['Name'] = ACDCBlock.name(self.wmspec.name(),
acdcInfo["fileset"],
block['offset'], block['files'])
dbsBlock['NumberOfFiles'] = block['files']
dbsBlock['NumberOfEvents'] = block['events']
dbsBlock['NumberOfLumis'] = block['lumis']
dbsBlock["Sites"] = sitesFromStorageEelements(block["locations"])
dbsBlock['ACDC'] = acdcInfo
validBlocks.append(dbsBlock)
return validBlocks
def main():
start = time.time()
# blockName = match['Inputs'].keys()[0]
blockName = "/acdc/vlimant_ACDC0_task_SUS-RunIIFall18wmLHEGS-00025__v1_T_190218_145226_481/:pdmvserv_task_SUS-RunIIFall18wmLHEGS-00025__v1_T_181211_005112_2222:SUS-RunIIFall18wmLHEGS-00025_0/0/31055"
# acdcInfo = match['ACDC']
acdcInfo = {"database": "acdcserver",
"fileset": "/pdmvserv_task_SUS-RunIIFall18wmLHEGS-00025__v1_T_181211_005112_2222/SUS-RunIIFall18wmLHEGS-00025_0",
"collection": "pdmvserv_task_SUS-RunIIFall18wmLHEGS-00025__v1_T_181211_005112_2222",
"server": "https://cmsweb.cern.ch/couchdb"}
acdc = DataCollectionService(acdcInfo["server"], acdcInfo["database"])
splitedBlockName = ACDCBlock.splitBlockName(blockName)
print("Splitted block name: %s" % splitedBlockName)
fileLists = acdc.getChunkFiles(acdcInfo['collection'],
acdcInfo['fileset'],
splitedBlockName['Offset'],
splitedBlockName['NumOfFiles'])
print("Retrieved %d unique files from the ACDCServer" % len(fileLists))
block = {}
block["Files"] = fileLists
wantedLumis = set([252052, 240646])
for f in fileLists:
for run in f['runs']:
maskDict = run.json()
lumisSet = set(maskDict['Lumis'].keys())
if wantedLumis.intersection(lumisSet):
print("File: %s with events: %s, contains these lumis: %s" % (f['lfn'], f['events'], wantedLumis.intersection(lumisSet)))
# with open("chunkfiles.json", 'w') as fo:
# json.dump(block, fo)
end = time.time()
print("Spent %s secs running so far" % (end - start))
sys.exit(1)
### Now doing the WMBSHelper stuff
reqUrl = "https://cmsweb.cern.ch/couchdb/reqmgr_workload_cache"
requestName = "vlimant_ACDC0_task_HIG-RunIIFall17wmLHEGS-01122__v1_T_180808_130708_5376"
wmspec = WMWorkloadHelper()
wmspec.loadSpecFromCouch(reqUrl, requestName)
taskName = "HIG-RunIIFall17DRPremix-00788_0"
mask = None
cacheDir = "/data/srv/wmagent/v1.1.14.patch6/install/wmagent/WorkQueueManager/cache"
# wmbsHelper = WMBSHelper(wmspec, match['TaskName'], blockName, mask, self.params['CacheDir'])
wmbsHelper = WMBSHelper(wmspec, taskName, blockName, mask, cacheDir)
sub, numFilesAdded = wmbsHelper.createSubscriptionAndAddFiles(block=block)
def validBlocks(self, task):
"""Return blocks that pass the input data restriction according
to the splitting algorithm"""
validBlocks = []
acdcInfo = task.getInputACDC()
if not acdcInfo:
raise WorkQueueWMSpecError(
self.wmspec, 'No acdc section for %s' % task.getPathName())
acdc = DataCollectionService(acdcInfo["server"], acdcInfo["database"])
if self.data:
acdcBlockSplit = ACDCBlock.splitBlockName(self.data.keys()[0])
else:
# if self.data is not passed, assume the the data is input dataset
# from the spec
acdcBlockSplit = False
if acdcBlockSplit:
dbsBlock = {}
dbsBlock['Name'] = self.data.keys()[0]
block = acdc.getChunkInfo(
acdcInfo['collection'],
acdcBlockSplit['TaskName'],
acdcBlockSplit['Offset'],
acdcBlockSplit['NumOfFiles'],
user=self.wmspec.getOwner().get("name"),
group=self.wmspec.getOwner().get("group"))
dbsBlock['NumberOfFiles'] = block['files']
dbsBlock['NumberOfEvents'] = block['events']
dbsBlock['NumberOfLumis'] = block['lumis']
dbsBlock['ACDC'] = acdcInfo
if task.getTrustSitelists():
dbsBlock["Sites"] = self.sites
else:
# TODO remove this line when all DBS origin_site_name is converted to PNN
block["locations"] = self.siteDB.checkAndConvertSENameToPNN(
block["locations"])
# upto this
dbsBlock["Sites"] = self.siteDB.PNNstoPSNs(block["locations"])
validBlocks.append(dbsBlock)
else:
if self.args['SplittingAlgo'] in self.unsupportedAlgos:
raise WorkQueueWMSpecError(
self.wmspec, 'ACDC is not supported for %s' %
self.args['SplittingAlgo'])
splittingFunc = self.defaultAlgo
if self.args['SplittingAlgo'] in self.algoMapping:
splittingFunc = self.algoMapping[self.args['SplittingAlgo']]
validBlocks = splittingFunc(acdc, acdcInfo, task)
return validBlocks
def testFailedJobsScrambledWf(self):
"""
Performance test of failedJobs where jobs belong to 10 different
workflows and 3 different tasks
"""
loadList = []
for i in range(1, 5000):
wfName = "wf%d" % (i % 10)
taskName = "/wf%d/task%d" % (i % 10, i % 3)
loadList.append(self.jobConfig(wfName, taskName, i, '/file/name/lfn1'))
dcs = DataCollectionService(url=self.testInit.couchUrl, database="wmcore-acdc-datacollectionsvc")
dcs.failedJobs(loadList)
return
def testFailedJobsScrambledWf(self):
"""
Performance test of failedJobs where jobs belong to 10 different
workflows and 3 different tasks
"""
loadList = []
for i in range(1, 5000):
wfName = "wf%d" % (i % 10)
taskName = "/wf%d/task%d" % (i % 10, i % 3)
loadList.append(self.jobConfig(wfName, taskName, i, '/file/name/lfn1'))
dcs = DataCollectionService(url=self.testInit.couchUrl, database="wmcore-acdc-datacollectionsvc")
dcs.failedJobs(loadList)
return
def main():
start = time.time()
# acdcInfo = match['ACDC']
acdcInfo = {
"database": "acdcserver",
"fileset":
"/pdmvserv_task_SUS-RunIIFall18wmLHEGS-00025__v1_T_181211_005112_2222/SUS-RunIIFall18wmLHEGS-00025_0",
"collection":
"pdmvserv_task_SUS-RunIIFall18wmLHEGS-00025__v1_T_181211_005112_2222",
"server": "https://cmsweb.cern.ch/couchdb"
}
dcs = DataCollectionService(acdcInfo["server"], acdcInfo["database"])
# acdcFileList = dcs.getProductionACDCInfo(acdcInfo['collection'], acdcInfo['fileset'])
files = dcs._getFilesetInfo(acdcInfo['collection'], acdcInfo['fileset'])
print("%s" % pformat(files[0]))
files = mergeFilesInfo(files)
acdcFileList = []
for value in files:
fileInfo = {
"lfn": value["lfn"],
"first_event": value["first_event"],
"lumis": value["runs"][0]["lumis"],
"events": value["events"]
}
acdcFileList.append(fileInfo)
#print("Data retrieved:\n%s" % pformat(acdcFileList))
print("Retrieved %d files from the ACDCServer" % len(acdcFileList))
listLumis = []
wantedLumis = set([252052, 240646])
for f in acdcFileList:
listLumis.extend(f['lumis'])
lumisSet = set(f['lumis'])
if wantedLumis.intersection(lumisSet):
print("File: %s with events: %s, contains these lumis: %s" %
(f['lfn'], f['events'], f['lumis']))
print("Total amount of lumis: %d, where unique are: %d" %
(len(listLumis), len(set(listLumis))))
# with open("chunkfiles.json", 'w') as fo:
# json.dump(block, fo)
end = time.time()
print("Spent %s secs running so far" % (end - start))
sys.exit(1)
def __init__(self, config):
"""
Initialise class members
"""
BaseWorkerThread.__init__(self)
self.config = config
myThread = threading.currentThread()
self.daoFactory = DAOFactory(package = "WMCore.WMBS",
logger = myThread.logger,
dbinterface = myThread.dbi)
self.changeState = ChangeState(self.config)
self.maxRetries = self.config.ErrorHandler.maxRetries
self.maxProcessSize = getattr(self.config.ErrorHandler, 'maxProcessSize', 250)
self.getJobs = self.daoFactory(classname = "Jobs.GetAllJobs")
self.dataCollection = DataCollectionService(url = config.ACDC.couchurl,
database = config.ACDC.database)
# initialize the alert framework (if available - config.Alert present)
# self.sendAlert will be then be available
self.initAlerts(compName = "ErrorHandler")
return
def setUp(self):
"""
setup for test.
"""
myThread = threading.currentThread()
self.testInit = TestInitCouchApp(__file__)
self.testInit.setLogging()
self.testInit.setDatabaseConnection(destroyAllDatabase = True)
self.testInit.setSchema(customModules = ["WMCore.WMBS"],
useDefault = False)
self.testInit.setupCouch("errorhandler_t", "GroupUser", "ACDC")
self.testInit.setupCouch("errorhandler_t_jd/jobs", "JobDump")
self.testInit.setupCouch("errorhandler_t_jd/fwjrs", "FWJRDump")
self.daofactory = DAOFactory(package = "WMCore.WMBS",
logger = myThread.logger,
dbinterface = myThread.dbi)
self.getJobs = self.daofactory(classname = "Jobs.GetAllJobs")
self.setJobTime = self.daofactory(classname = "Jobs.SetStateTime")
locationAction = self.daofactory(classname = "Locations.New")
locationAction.execute(siteName = "malpaquet", seName = "malpaquet")
self.testDir = self.testInit.generateWorkDir()
self.nJobs = 10
self.dataCS = DataCollectionService(url = self.testInit.couchUrl,
database = "errorhandler_t")
return
def __init__(self, config):
"""
Initialise class members
"""
BaseWorkerThread.__init__(self)
self.config = config
myThread = threading.currentThread()
self.daoFactory = DAOFactory(package="WMCore.WMBS", logger=myThread.logger, dbinterface=myThread.dbi)
self.changeState = ChangeState(self.config)
self.maxRetries = self.config.ErrorHandler.maxRetries
if not isinstance(self.maxRetries, dict):
self.maxRetries = {"default": self.maxRetries}
if "default" not in self.maxRetries:
raise ErrorHandlerException("Max retries for the default job type must be specified")
self.maxProcessSize = getattr(self.config.ErrorHandler, "maxProcessSize", 250)
self.exitCodes = getattr(self.config.ErrorHandler, "failureExitCodes", [])
self.maxFailTime = getattr(self.config.ErrorHandler, "maxFailTime", 32 * 3600)
self.readFWJR = getattr(self.config.ErrorHandler, "readFWJR", False)
self.passCodes = getattr(self.config.ErrorHandler, "passExitCodes", [])
self.getJobs = self.daoFactory(classname="Jobs.GetAllJobs")
self.idLoad = self.daoFactory(classname="Jobs.LoadFromIDWithType")
self.loadAction = self.daoFactory(classname="Jobs.LoadForErrorHandler")
self.dataCollection = DataCollectionService(url=config.ACDC.couchurl, database=config.ACDC.database)
return
def validBlocks(self, task):
"""Return blocks that pass the input data restriction according
to the splitting algorithm"""
validBlocks = []
acdcInfo = task.getInputACDC()
if not acdcInfo:
raise WorkQueueWMSpecError(self.wmspec, 'No acdc section for %s' % task.getPathName())
acdc = DataCollectionService(acdcInfo["server"], acdcInfo["database"])
if self.data:
acdcBlockSplit = ACDCBlock.splitBlockName(self.data.keys()[0])
else:
# if self.data is not passed, assume the the data is input dataset
# from the spec
acdcBlockSplit = False
if acdcBlockSplit:
dbsBlock = {}
dbsBlock['Name'] = self.data.keys()[0]
block = acdc.getChunkInfo(acdcInfo['collection'],
acdcBlockSplit['TaskName'],
acdcBlockSplit['Offset'],
acdcBlockSplit['NumOfFiles'],
user = self.wmspec.getOwner().get("name"),
group = self.wmspec.getOwner().get("group"))
dbsBlock['NumberOfFiles'] = block['files']
dbsBlock['NumberOfEvents'] = block['events']
dbsBlock['NumberOfLumis'] = block['lumis']
dbsBlock['ACDC'] = acdcInfo
if task.inputLocationFlag():
dbsBlock["Sites"] = self.sites
else:
#TODO remove this line when all DBS origin_site_name is converted to PNN
block["locations"] = self.siteDB.checkAndConvertSENameToPNN(block["locations"])
#upto this
dbsBlock["Sites"] = self.siteDB.PNNstoPSNs(block["locations"])
validBlocks.append(dbsBlock)
else:
if self.args['SplittingAlgo'] in self.unsupportedAlgos:
raise WorkQueueWMSpecError(self.wmspec, 'ACDC is not supported for %s' % self.args['SplittingAlgo'])
splittingFunc = self.defaultAlgo
if self.args['SplittingAlgo'] in self.algoMapping:
splittingFunc = self.algoMapping[self.args['SplittingAlgo']]
validBlocks = splittingFunc(acdc, acdcInfo, task)
return validBlocks
def __init__(self, config):
"""
Initialise class members
"""
BaseWorkerThread.__init__(self)
self.config = config
myThread = threading.currentThread()
self.daoFactory = DAOFactory(package="WMCore.WMBS",
logger=myThread.logger,
dbinterface=myThread.dbi)
self.changeState = ChangeState(self.config)
self.maxRetries = self.config.ErrorHandler.maxRetries
if type(self.maxRetries) != dict:
self.maxRetries = {'default': self.maxRetries}
if 'default' not in self.maxRetries:
raise ErrorHandlerException(
'Max retries for the default job type must be specified')
self.maxProcessSize = getattr(self.config.ErrorHandler,
'maxProcessSize', 250)
self.exitCodes = getattr(self.config.ErrorHandler, 'failureExitCodes',
[])
self.maxFailTime = getattr(self.config.ErrorHandler, 'maxFailTime',
32 * 3600)
self.readFWJR = getattr(self.config.ErrorHandler, 'readFWJR', False)
self.passCodes = getattr(self.config.ErrorHandler, 'passExitCodes', [])
self.getJobs = self.daoFactory(classname="Jobs.GetAllJobs")
self.idLoad = self.daoFactory(classname="Jobs.LoadFromIDWithType")
self.loadAction = self.daoFactory(classname="Jobs.LoadForErrorHandler")
self.dataCollection = DataCollectionService(
url=config.ACDC.couchurl, database=config.ACDC.database)
# initialize the alert framework (if available - config.Alert present)
# self.sendAlert will be then be available
self.initAlerts(compName="ErrorHandler")
# Some exit codes imply an immediate failure, non-configurable
self.exitCodes.extend(WMJobPermanentSystemErrors)
return
def lumiListFromACDC(couchURL=None, couchDB=None, filesetName=None, collectionName=None):
"""
This is not implemented yet
:return:
"""
from WMCore.ACDC.DataCollectionService import DataCollectionService
goodRunList = None
try:
logging.info('Creating jobs for ACDC fileset %s', filesetName)
dcs = DataCollectionService(couchURL, couchDB)
goodRunList = dcs.getLumilistWhitelist(collectionName, filesetName)
except Exception as ex:
msg = "Exception while trying to load goodRunList. "
msg += "Refusing to create any jobs.\nDetails: %s" % ex.__str__()
logging.exception(msg)
return goodRunList
def lumiListFromACDC(couchURL=None, couchDB=None, filesetName=None, collectionName=None):
"""
This is not implemented yet
:return:
"""
from WMCore.ACDC.DataCollectionService import DataCollectionService
goodRunList = None
try:
logging.info('Creating jobs for ACDC fileset %s', filesetName)
dcs = DataCollectionService(couchURL, couchDB)
goodRunList = dcs.getLumilistWhitelist(collectionName, filesetName)
except Exception as ex:
msg = "Exception while trying to load goodRunList. "
msg += "Refusing to create any jobs.\nDetails: %s" % str(ex)
logging.exception(msg)
return goodRunList
def validBlocks(self, task):
"""Return blocks that pass the input data restriction according
to the splitting algorithm"""
validBlocks = []
acdcInfo = task.getInputACDC()
if not acdcInfo:
raise WorkQueueWMSpecError(
self.wmspec, 'No acdc section for %s' % task.getPathName())
acdc = DataCollectionService(acdcInfo["server"], acdcInfo["database"])
if self.data:
acdcBlockSplit = ACDCBlock.splitBlockName(next(iter(self.data)))
else:
# if self.data is not passed, assume the the data is input dataset
# from the spec
acdcBlockSplit = False
if acdcBlockSplit:
dbsBlock = {}
dbsBlock['Name'] = next(iter(self.data))
block = acdc.getChunkInfo(acdcInfo['collection'],
acdcBlockSplit['TaskName'],
acdcBlockSplit['Offset'],
acdcBlockSplit['NumOfFiles'])
dbsBlock['NumberOfFiles'] = block['files']
dbsBlock['NumberOfEvents'] = block['events']
dbsBlock['NumberOfLumis'] = block['lumis']
dbsBlock['ACDC'] = acdcInfo
if task.getTrustSitelists().get('trustlists'):
dbsBlock["Sites"] = self.sites
else:
dbsBlock["Sites"] = self.cric.PNNstoPSNs(block["locations"])
validBlocks.append(dbsBlock)
else:
if self.args['SplittingAlgo'] in self.unsupportedAlgos:
raise WorkQueueWMSpecError(
self.wmspec, 'ACDC is not supported for %s' %
self.args['SplittingAlgo'])
splittingFunc = self.defaultAlgo
if self.args['SplittingAlgo'] in self.algoMapping:
splittingFunc = self.algoMapping[self.args['SplittingAlgo']]
validBlocks = splittingFunc(acdc, acdcInfo, task)
return validBlocks
def validBlocks(self, task):
"""Return blocks that pass the input data restriction according
to the splitting algorithm"""
validBlocks = []
acdcInfo = task.getInputACDC()
if not acdcInfo:
raise WorkQueueWMSpecError(self.wmspec, 'No acdc section for %s' % task.getPathName())
acdc = DataCollectionService(acdcInfo["server"], acdcInfo["database"])
if self.data:
acdcBlockSplit = ACDCBlock.splitBlockName(self.data.keys()[0])
else:
# if self.data is not passed, assume the the data is input dataset
# from the spec
acdcBlockSplit = False
if acdcBlockSplit:
dbsBlock = {}
dbsBlock['Name'] = self.data.keys()[0]
block = acdc.getChunkInfo(acdcInfo['collection'],
acdcBlockSplit['TaskName'],
acdcBlockSplit['Offset'],
acdcBlockSplit['NumOfFiles'])
dbsBlock['NumberOfFiles'] = block['files']
dbsBlock['NumberOfEvents'] = block['events']
dbsBlock['NumberOfLumis'] = block['lumis']
dbsBlock['ACDC'] = acdcInfo
if task.getTrustSitelists().get('trustlists'):
dbsBlock["Sites"] = self.sites
else:
dbsBlock["Sites"] = self.siteDB.PNNstoPSNs(block["locations"])
validBlocks.append(dbsBlock)
else:
if self.args['SplittingAlgo'] in self.unsupportedAlgos:
raise WorkQueueWMSpecError(self.wmspec, 'ACDC is not supported for %s' % self.args['SplittingAlgo'])
splittingFunc = self.defaultAlgo
if self.args['SplittingAlgo'] in self.algoMapping:
splittingFunc = self.algoMapping[self.args['SplittingAlgo']]
validBlocks = splittingFunc(acdc, acdcInfo, task)
return validBlocks
def main():
start = time.time()
# acdcInfo = match['ACDC']
acdcInfo = {"database": "acdcserver",
"fileset": "/pdmvserv_task_SUS-RunIIFall18wmLHEGS-00025__v1_T_181211_005112_2222/SUS-RunIIFall18wmLHEGS-00025_0",
"collection": "pdmvserv_task_SUS-RunIIFall18wmLHEGS-00025__v1_T_181211_005112_2222",
"server": "https://cmsweb.cern.ch/couchdb"}
dcs = DataCollectionService(acdcInfo["server"], acdcInfo["database"])
# acdcFileList = dcs.getProductionACDCInfo(acdcInfo['collection'], acdcInfo['fileset'])
files = dcs._getFilesetInfo(acdcInfo['collection'], acdcInfo['fileset'])
print("%s" % pformat(files[0]))
files = mergeFilesInfo(files)
acdcFileList = []
for value in files:
fileInfo = {"lfn": value["lfn"],
"first_event": value["first_event"],
"lumis": value["runs"][0]["lumis"],
"events": value["events"]}
acdcFileList.append(fileInfo)
#print("Data retrieved:\n%s" % pformat(acdcFileList))
print("Retrieved %d files from the ACDCServer" % len(acdcFileList))
listLumis = []
wantedLumis = set([252052, 240646])
for f in acdcFileList:
listLumis.extend(f['lumis'])
lumisSet = set(f['lumis'])
if wantedLumis.intersection(lumisSet):
print("File: %s with events: %s, contains these lumis: %s" % (f['lfn'], f['events'], f['lumis']))
print("Total amount of lumis: %d, where unique are: %d" % (len(listLumis), len(set(listLumis))))
# with open("chunkfiles.json", 'w') as fo:
# json.dump(block, fo)
end = time.time()
print("Spent %s secs running so far" % (end - start))
sys.exit(1)
def test06UploadACDC(self):
# get previous request we can piggyback on
for request in reversed(self.__class__.reqmgr.getRequest()):
request = request['WMCore.RequestManager.DataStructs.Request.Request']['RequestName']
if 'RequestCancellation_t' in request:
self.__class__.requestParams['OriginalRequestName'] = request
break
else:
raise nose.SkipTest("no suitable request in reqmgr to resubmit")
self.__class__.requestParams['InitialTaskPath'] = self.__class__.requestParams['InitialTaskPath'] % self.__class__.requestParams['OriginalRequestName']
self.__class__.requestParams['ACDCServer'] = self.__class__.endpoint + '/couchdb'
# create and upload acdc
service = DataCollectionService(url=self.__class__.endpoint + '/couchdb', database = 'wmagent_acdc')
service.createCollection(self.__class__.requestParams['OriginalRequestName'], 'integration', 'DMWM')
with open(os.path.join(getTestBase(), '..', 'data', 'ACDC', 'linacre_ACDC_ReReco13JulCosmics_120809_130020_117_120823_200309_5735.json')) as infile:
acdc_json = infile.read().replace('linacre_ACDC_ReReco13JulCosmics_120809_130020_117_120823_200309_5735',
self.__class__.requestParams['OriginalRequestName'])
acdc_json = loads(acdc_json)
acdc_database = Database('wmagent_acdc', self.__class__.endpoint + '/couchdb')
acdc_database.commit(acdc_json)
def lumiListFromACDC(couchURL=None, couchDB=None, filesetName=None, collectionName=None, ignoreACDC=False):
"""
This is not implemented yet
:return:
"""
try:
logging.info('Creating jobs for ACDC fileset %s', filesetName)
dcs = DataCollectionService(couchURL, couchDB)
goodRunList = dcs.getLumilistWhitelist(collectionName, filesetName)
except Exception as ex:
msg = "Exception while trying to load goodRunList\n"
if ignoreACDC: # Logic can go in main function?
msg += "Ditching goodRunList\n" + str(ex) + str(traceback.format_exc())
logging.error(msg)
goodRunList = {}
else:
msg += "Refusing to create any jobs.\n" + str(ex) + str(traceback.format_exc())
logging.error(msg)
return None # An error condtion - check
return goodRunList
def test06UploadACDC(self):
# get previous request we can piggyback on
for request in reversed(self.__class__.reqmgr.getRequest()):
request = request[
'WMCore.RequestManager.DataStructs.Request.Request'][
'RequestName']
if 'RequestCancellation_t' in request:
self.__class__.requestParams['OriginalRequestName'] = request
break
else:
raise nose.SkipTest("no suitable request in reqmgr to resubmit")
self.__class__.requestParams[
'InitialTaskPath'] = self.__class__.requestParams[
'InitialTaskPath'] % self.__class__.requestParams[
'OriginalRequestName']
self.__class__.requestParams[
'ACDCServer'] = self.__class__.endpoint + '/couchdb'
# create and upload acdc
service = DataCollectionService(url=self.__class__.endpoint +
'/couchdb',
database='wmagent_acdc')
service.createCollection(
self.__class__.requestParams['OriginalRequestName'], 'integration',
'DMWM')
with open(
os.path.join(
getTestBase(), '..', 'data', 'ACDC',
'linacre_ACDC_ReReco13JulCosmics_120809_130020_117_120823_200309_5735.json'
)) as infile:
acdc_json = infile.read().replace(
'linacre_ACDC_ReReco13JulCosmics_120809_130020_117_120823_200309_5735',
self.__class__.requestParams['OriginalRequestName'])
acdc_json = loads(acdc_json)
acdc_database = Database('wmagent_acdc',
self.__class__.endpoint + '/couchdb')
acdc_database.commit(acdc_json)
def __init__(self, config):
"""
Initialise class members
"""
BaseWorkerThread.__init__(self)
self.config = config
myThread = threading.currentThread()
self.daoFactory = DAOFactory(package="WMCore.WMBS",
logger=myThread.logger,
dbinterface=myThread.dbi)
self.changeState = ChangeState(self.config)
if hasattr(self.config, "Tier0Feeder"):
self.reqAuxDB = None
self.maxRetries = self.config.ErrorHandler.maxRetries
else:
self.reqAuxDB = ReqMgrAux(self.config.General.ReqMgr2ServiceURL)
self.exitCodesNoRetry = []
self.maxProcessSize = getattr(self.config.ErrorHandler,
'maxProcessSize', 250)
self.maxFailTime = getattr(self.config.ErrorHandler, 'maxFailTime',
32 * 3600)
self.readFWJR = getattr(self.config.ErrorHandler, 'readFWJR', False)
self.passCodes = getattr(self.config.ErrorHandler, 'passExitCodes', [])
self.getJobs = self.daoFactory(classname="Jobs.GetAllJobs")
self.idLoad = self.daoFactory(classname="Jobs.LoadFromIDWithType")
self.loadAction = self.daoFactory(classname="Jobs.LoadForErrorHandler")
self.dataCollection = DataCollectionService(
url=config.ACDC.couchurl, database=config.ACDC.database)
self.setupComponentParam()
return
def algorithm(self, *args, **kwargs):
"""
_algorithm_
An event base splitting algorithm. All available files are split into a
set number of events per job.
"""
eventsPerJob = int(kwargs.get("events_per_job", 100))
eventsPerLumi = int(kwargs.get("events_per_lumi", eventsPerJob))
getParents = kwargs.get("include_parents", False)
lheInput = kwargs.get("lheInputFiles", False)
collectionName = kwargs.get('collectionName', None)
timePerEvent, sizePerEvent, memoryRequirement = \
self.getPerformanceParameters(kwargs.get('performance', {}))
acdcFileList = []
# If we have runLumi info, we need to load it from couch
if collectionName:
try:
from WMCore.ACDC.DataCollectionService import DataCollectionService
couchURL = kwargs.get('couchURL')
couchDB = kwargs.get('couchDB')
filesetName = kwargs.get('filesetName')
collectionName = kwargs.get('collectionName')
owner = kwargs.get('owner')
group = kwargs.get('group')
logging.info('Creating jobs for ACDC fileset %s' % filesetName)
dcs = DataCollectionService(couchURL, couchDB)
acdcFileList = dcs.getProductionACDCInfo(
collectionName, filesetName, owner, group)
except Exception, ex:
msg = "Exception while trying to load goodRunList\n"
msg += "Refusing to create any jobs.\n"
msg += str(ex)
msg += str(traceback.format_exc())
logging.error(msg)
return
def _getDBSBlock(self, match, wmspec):
"""Get DBS info for this block"""
blockName = match['Inputs'].keys()[0] #TODO: Allow more than one
if match['ACDC']:
acdcInfo = match['ACDC']
acdc = DataCollectionService(acdcInfo["server"],
acdcInfo["database"])
collection = acdc.getDataCollection(acdcInfo['collection'])
splitedBlockName = ACDCBlock.splitBlockName(blockName)
fileLists = acdc.getChunkFiles(
acdcInfo['collection'],
acdcInfo['fileset'],
splitedBlockName['Offset'],
splitedBlockName['NumOfFiles'],
user=wmspec.getOwner().get("name"),
group=wmspec.getOwner().get("group"))
block = {}
block["Files"] = fileLists
return blockName, block
else:
dbs = get_dbs(match['Dbs'])
if wmspec.getTask(match['TaskName']).parentProcessingFlag():
dbsBlockDict = dbs.getFileBlockWithParents(blockName)
else:
dbsBlockDict = dbs.getFileBlock(blockName)
if wmspec.locationDataSourceFlag():
blockInfo = dbsBlockDict[blockName]
seElements = []
for cmsSite in match['Inputs'].values(
)[0]: #TODO: Allow more than one
ses = self.SiteDB.cmsNametoSE(cmsSite)
seElements.extend(ses)
seElements = list(set(seElements))
blockInfo['StorageElements'] = seElements
return blockName, dbsBlockDict[blockName]
def algorithm(self, *args, **kwargs):
"""
_algorithm_
An event base splitting algorithm. All available files are split into a
set number of events per job.
"""
eventsPerJob = int(kwargs.get("events_per_job", 100))
eventsPerLumi = int(kwargs.get("events_per_lumi", eventsPerJob))
getParents = kwargs.get("include_parents", False)
lheInput = kwargs.get("lheInputFiles", False)
collectionName = kwargs.get('collectionName', None)
timePerEvent, sizePerEvent, memoryRequirement = \
self.getPerformanceParameters(kwargs.get('performance', {}))
acdcFileList = []
# If we have runLumi info, we need to load it from couch
if collectionName:
try:
from WMCore.ACDC.DataCollectionService import DataCollectionService
couchURL = kwargs.get('couchURL')
couchDB = kwargs.get('couchDB')
filesetName = kwargs.get('filesetName')
collectionName = kwargs.get('collectionName')
owner = kwargs.get('owner')
group = kwargs.get('group')
logging.info('Creating jobs for ACDC fileset %s' % filesetName)
dcs = DataCollectionService(couchURL, couchDB)
acdcFileList = dcs.getProductionACDCInfo(collectionName, filesetName, owner, group)
except Exception, ex:
msg = "Exception while trying to load goodRunList\n"
msg += "Refusing to create any jobs.\n"
msg += str(ex)
msg += str(traceback.format_exc())
logging.error(msg)
return
def __init__(self, config):
"""
Initialise class members
"""
BaseWorkerThread.__init__(self)
self.config = config
myThread = threading.currentThread()
self.daoFactory = DAOFactory(package = "WMCore.WMBS",
logger = myThread.logger,
dbinterface = myThread.dbi)
self.changeState = ChangeState(self.config)
self.maxRetries = self.config.ErrorHandler.maxRetries
if type(self.maxRetries) != dict:
self.maxRetries = {'default' : self.maxRetries}
if 'default' not in self.maxRetries:
raise ErrorHandlerException('Max retries for the default job type must be specified')
self.maxProcessSize = getattr(self.config.ErrorHandler, 'maxProcessSize', 250)
self.exitCodes = getattr(self.config.ErrorHandler, 'failureExitCodes', [])
self.maxFailTime = getattr(self.config.ErrorHandler, 'maxFailTime', 32 * 3600)
self.readFWJR = getattr(self.config.ErrorHandler, 'readFWJR', False)
self.passCodes = getattr(self.config.ErrorHandler, 'passExitCodes', [])
self.getJobs = self.daoFactory(classname = "Jobs.GetAllJobs")
self.idLoad = self.daoFactory(classname = "Jobs.LoadFromIDWithType")
self.loadAction = self.daoFactory(classname = "Jobs.LoadForErrorHandler")
self.dataCollection = DataCollectionService(url = config.ACDC.couchurl,
database = config.ACDC.database)
# initialize the alert framework (if available - config.Alert present)
# self.sendAlert will be then be available
self.initAlerts(compName = "ErrorHandler")
# Some exit codes imply an immediate failure, non-configurable
self.exitCodes.extend(WMJobPermanentSystemErrors)
return
def __init__(self, config):
"""
Initialise class members
"""
BaseWorkerThread.__init__(self)
self.config = config
myThread = threading.currentThread()
self.daoFactory = DAOFactory(package="WMCore.WMBS",
logger=myThread.logger,
dbinterface=myThread.dbi)
self.changeState = ChangeState(self.config)
if hasattr(self.config, "Tier0Feeder"):
self.reqAuxDB = None
self.maxRetries = self.config.ErrorHandler.maxRetries
else:
self.reqAuxDB = ReqMgrAux(self.config.General.ReqMgr2ServiceURL)
self.maxRetries = self.reqAuxDB.getWMAgentConfig(self.config.Agent.hostName).get("MaxRetries")
if not isinstance(self.maxRetries, dict):
self.maxRetries = {'default': self.maxRetries}
if 'default' not in self.maxRetries:
raise ErrorHandlerException('Max retries for the default job type must be specified')
self.exitCodesNoRetry = []
self.maxProcessSize = getattr(self.config.ErrorHandler, 'maxProcessSize', 250)
self.maxFailTime = getattr(self.config.ErrorHandler, 'maxFailTime', 32 * 3600)
self.readFWJR = getattr(self.config.ErrorHandler, 'readFWJR', False)
self.passCodes = getattr(self.config.ErrorHandler, 'passExitCodes', [])
self.getJobs = self.daoFactory(classname="Jobs.GetAllJobs")
self.idLoad = self.daoFactory(classname="Jobs.LoadFromIDWithType")
self.loadAction = self.daoFactory(classname="Jobs.LoadForErrorHandler")
self.dataCollection = DataCollectionService(url=config.ACDC.couchurl,
database=config.ACDC.database)
return
def __init__(self, config):
"""
Initialise class members
"""
BaseWorkerThread.__init__(self)
self.config = config
myThread = threading.currentThread()
self.daoFactory = DAOFactory(package="WMCore.WMBS",
logger=myThread.logger,
dbinterface=myThread.dbi)
self.changeState = ChangeState(self.config)
if hasattr(self.config, "Tier0Feeder"):
self.reqAuxDB = None
self.maxRetries = self.config.ErrorHandler.maxRetries
else:
self.reqAuxDB = ReqMgrAux(self.config.General.ReqMgr2ServiceURL)
self.exitCodesNoRetry = []
self.maxProcessSize = getattr(self.config.ErrorHandler, 'maxProcessSize', 250)
self.maxFailTime = getattr(self.config.ErrorHandler, 'maxFailTime', 32 * 3600)
self.readFWJR = getattr(self.config.ErrorHandler, 'readFWJR', False)
self.passCodes = getattr(self.config.ErrorHandler, 'passExitCodes', [])
self.getJobs = self.daoFactory(classname="Jobs.GetAllJobs")
self.idLoad = self.daoFactory(classname="Jobs.LoadFromIDWithType")
self.loadAction = self.daoFactory(classname="Jobs.LoadForErrorHandler")
self.dataCollection = DataCollectionService(url=config.ACDC.couchurl,
database=config.ACDC.database)
self.setupComponentParam()
return
class AccountantWorker(WMConnectionBase):
"""
Class that actually does the work of parsing FWJRs for the Accountant
Run through ProcessPool
"""
def __init__(self, config):
"""
__init__
Create all DAO objects that are used by this class.
"""
WMConnectionBase.__init__(self, "WMCore.WMBS")
myThread = threading.currentThread()
self.dbsDaoFactory = DAOFactory(package="WMComponent.DBS3Buffer",
logger=myThread.logger,
dbinterface=myThread.dbi)
self.getOutputMapAction = self.daofactory(
classname="Jobs.GetOutputMap")
self.bulkAddToFilesetAction = self.daofactory(
classname="Fileset.BulkAddByLFN")
self.bulkParentageAction = self.daofactory(
classname="Files.AddBulkParentage")
self.getJobTypeAction = self.daofactory(classname="Jobs.GetType")
self.getParentInfoAction = self.daofactory(
classname="Files.GetParentInfo")
self.setParentageByJob = self.daofactory(
classname="Files.SetParentageByJob")
self.setParentageByMergeJob = self.daofactory(
classname="Files.SetParentageByMergeJob")
self.setFileRunLumi = self.daofactory(classname="Files.AddRunLumi")
self.setFileLocation = self.daofactory(
classname="Files.SetLocationByLFN")
self.setFileAddChecksum = self.daofactory(
classname="Files.AddChecksumByLFN")
self.addFileAction = self.daofactory(classname="Files.Add")
self.jobCompleteInput = self.daofactory(classname="Jobs.CompleteInput")
self.setBulkOutcome = self.daofactory(classname="Jobs.SetOutcomeBulk")
self.getWorkflowSpec = self.daofactory(
classname="Workflow.GetSpecAndNameFromTask")
self.getJobInfoByID = self.daofactory(classname="Jobs.LoadFromID")
self.getFullJobInfo = self.daofactory(
classname="Jobs.LoadForErrorHandler")
self.getJobTaskNameAction = self.daofactory(
classname="Jobs.GetFWJRTaskName")
self.pnn_to_psn = self.daofactory(
classname="Locations.GetPNNtoPSNMapping").execute()
self.dbsStatusAction = self.dbsDaoFactory(
classname="DBSBufferFiles.SetStatus")
self.dbsParentStatusAction = self.dbsDaoFactory(
classname="DBSBufferFiles.GetParentStatus")
self.dbsChildrenAction = self.dbsDaoFactory(
classname="DBSBufferFiles.GetChildren")
self.dbsCreateFiles = self.dbsDaoFactory(
classname="DBSBufferFiles.Add")
self.dbsSetLocation = self.dbsDaoFactory(
classname="DBSBufferFiles.SetLocationByLFN")
self.dbsInsertLocation = self.dbsDaoFactory(
classname="DBSBufferFiles.AddLocation")
self.dbsSetChecksum = self.dbsDaoFactory(
classname="DBSBufferFiles.AddChecksumByLFN")
self.dbsSetRunLumi = self.dbsDaoFactory(
classname="DBSBufferFiles.AddRunLumi")
self.dbsGetWorkflow = self.dbsDaoFactory(classname="ListWorkflow")
self.dbsLFNHeritage = self.dbsDaoFactory(
classname="DBSBufferFiles.BulkHeritageParent")
self.stateChanger = ChangeState(config)
# Decide whether or not to attach jobReport to returned value
self.returnJobReport = getattr(config.JobAccountant,
'returnReportFromWorker', False)
# Store location for the specs for DBS
self.specDir = getattr(config.JobAccountant, 'specDir', None)
# maximum RAW EDM size for Repack output before data is put into Error dataset and skips PromptReco
self.maxAllowedRepackOutputSize = getattr(
config.JobAccountant, 'maxAllowedRepackOutputSize',
12 * 1024 * 1024 * 1024)
# ACDC service
self.dataCollection = DataCollectionService(
url=config.ACDC.couchurl, database=config.ACDC.database)
jobDBurl = sanitizeURL(config.JobStateMachine.couchurl)['url']
jobDBName = config.JobStateMachine.couchDBName
jobCouchdb = CouchServer(jobDBurl)
self.fwjrCouchDB = jobCouchdb.connectDatabase("%s/fwjrs" % jobDBName)
self.localWMStats = WMStatsWriter(config.TaskArchiver.localWMStatsURL,
appName="WMStatsAgent")
# Hold data for later commital
self.dbsFilesToCreate = []
self.wmbsFilesToBuild = []
self.wmbsMergeFilesToBuild = []
self.fileLocation = None
self.mergedOutputFiles = []
self.listOfJobsToSave = []
self.listOfJobsToFail = []
self.filesetAssoc = []
self.parentageBinds = []
self.parentageBindsForMerge = []
self.jobsWithSkippedFiles = {}
self.count = 0
self.datasetAlgoID = collections.deque(maxlen=1000)
self.datasetAlgoPaths = collections.deque(maxlen=1000)
self.dbsLocations = set()
self.workflowIDs = collections.deque(maxlen=1000)
self.workflowPaths = collections.deque(maxlen=1000)
self.phedex = PhEDEx()
self.locLists = self.phedex.getNodeMap()
return
def reset(self):
"""
_reset_
Reset all global vars between runs.
"""
self.dbsFilesToCreate = []
self.wmbsFilesToBuild = []
self.wmbsMergeFilesToBuild = []
self.fileLocation = None
self.mergedOutputFiles = []
self.listOfJobsToSave = []
self.listOfJobsToFail = []
self.filesetAssoc = []
self.parentageBinds = []
self.parentageBindsForMerge = []
self.jobsWithSkippedFiles = {}
gc.collect()
return
def loadJobReport(self, parameters):
"""
_loadJobReport_
Given a framework job report on disk, load it and return a
FwkJobReport instance. If there is any problem loading or parsing the
framework job report return None.
"""
# The jobReportPath may be prefixed with "file://" which needs to be
# removed so it doesn't confuse the FwkJobReport() parser.
jobReportPath = parameters.get("fwjr_path", None)
if not jobReportPath:
logging.error("Bad FwkJobReport Path: %s" % jobReportPath)
return self.createMissingFWKJR(parameters, 99999,
"FWJR path is empty")
jobReportPath = jobReportPath.replace("file://", "")
if not os.path.exists(jobReportPath):
logging.error("Bad FwkJobReport Path: %s" % jobReportPath)
return self.createMissingFWKJR(
parameters, 99999,
'Cannot find file in jobReport path: %s' % jobReportPath)
if os.path.getsize(jobReportPath) == 0:
logging.error("Empty FwkJobReport: %s" % jobReportPath)
return self.createMissingFWKJR(
parameters, 99998, 'jobReport of size 0: %s ' % jobReportPath)
jobReport = Report()
try:
jobReport.load(jobReportPath)
except Exception as ex:
msg = "Error loading jobReport %s\n" % jobReportPath
msg += str(ex)
logging.error(msg)
logging.debug("Failing job: %s\n" % parameters)
return self.createMissingFWKJR(parameters, 99997,
'Cannot load jobReport')
if len(jobReport.listSteps()) == 0:
logging.error("FwkJobReport with no steps: %s" % jobReportPath)
return self.createMissingFWKJR(
parameters, 99997,
'jobReport with no steps: %s ' % jobReportPath)
return jobReport
def isTaskExistInFWJR(self, jobReport, jobStatus):
"""
If taskName is not available in the FWJR, then tries to
recover it getting data from the SQL database.
"""
if not jobReport.getTaskName():
logging.warning(
"Trying to recover a corrupted FWJR for a %s job with job id %s"
% (jobStatus, jobReport.getJobID()))
jobInfo = self.getJobTaskNameAction.execute(
jobId=jobReport.getJobID(),
conn=self.getDBConn(),
transaction=self.existingTransaction())
jobReport.setTaskName(jobInfo['taskName'])
jobReport.save(jobInfo['fwjr_path'])
if not jobReport.getTaskName():
msg = "Report to developers. Failed to recover corrupted fwjr for %s job id %s" % (
jobStatus, jobReport.getJobID())
raise AccountantWorkerException(msg)
else:
logging.info(
"TaskName '%s' successfully recovered and added to fwjr id %s."
% (jobReport.getTaskName(), jobReport.getJobID()))
return
def __call__(self, parameters):
"""
__call__
Handle a completed job. The parameters dictionary will contain the job
ID and the path to the framework job report.
"""
returnList = []
self.reset()
for job in parameters:
logging.info("Handling %s" % job["fwjr_path"])
# Load the job and set the ID
fwkJobReport = self.loadJobReport(job)
fwkJobReport.setJobID(job['id'])
jobSuccess = self.handleJob(jobID=job["id"],
fwkJobReport=fwkJobReport)
if self.returnJobReport:
returnList.append({
'id': job["id"],
'jobSuccess': jobSuccess,
'jobReport': fwkJobReport
})
else:
returnList.append({'id': job["id"], 'jobSuccess': jobSuccess})
self.count += 1
self.beginTransaction()
# Now things done at the end of the job
# Do what we can with WMBS files
self.handleWMBSFiles(self.wmbsFilesToBuild, self.parentageBinds)
# handle merge files separately since parentage need to set
# separately to support robust merge
self.handleWMBSFiles(self.wmbsMergeFilesToBuild,
self.parentageBindsForMerge)
# Create DBSBufferFiles
self.createFilesInDBSBuffer()
# Handle filesetAssoc
if len(self.filesetAssoc) > 0:
self.bulkAddToFilesetAction.execute(
binds=self.filesetAssoc,
conn=self.getDBConn(),
transaction=self.existingTransaction())
# Move successful jobs to successful
if len(self.listOfJobsToSave) > 0:
idList = [x['id'] for x in self.listOfJobsToSave]
outcomeBinds = [{
'jobid': x['id'],
'outcome': x['outcome']
} for x in self.listOfJobsToSave]
self.setBulkOutcome.execute(binds=outcomeBinds,
conn=self.getDBConn(),
transaction=self.existingTransaction())
self.jobCompleteInput.execute(
id=idList,
lfnsToSkip=self.jobsWithSkippedFiles,
conn=self.getDBConn(),
transaction=self.existingTransaction())
self.stateChanger.propagate(self.listOfJobsToSave, "success",
"complete")
# If we have failed jobs, fail them
if len(self.listOfJobsToFail) > 0:
outcomeBinds = [{
'jobid': x['id'],
'outcome': x['outcome']
} for x in self.listOfJobsToFail]
self.setBulkOutcome.execute(binds=outcomeBinds,
conn=self.getDBConn(),
transaction=self.existingTransaction())
self.stateChanger.propagate(self.listOfJobsToFail, "jobfailed",
"complete")
# Arrange WMBS parentage
if len(self.parentageBinds) > 0:
self.setParentageByJob.execute(
binds=self.parentageBinds,
conn=self.getDBConn(),
transaction=self.existingTransaction())
if len(self.parentageBindsForMerge) > 0:
self.setParentageByMergeJob.execute(
binds=self.parentageBindsForMerge,
conn=self.getDBConn(),
transaction=self.existingTransaction())
# Straighten out DBS Parentage
if len(self.mergedOutputFiles) > 0:
self.handleDBSBufferParentage()
if len(self.jobsWithSkippedFiles) > 0:
self.handleSkippedFiles()
self.commitTransaction(existingTransaction=False)
return returnList
def outputFilesetsForJob(self, outputMap, merged, moduleLabel):
"""
_outputFilesetsForJob_
Determine if the file should be placed in any other fileset. Note that
this will not return the JobGroup output fileset as all jobs will have
their output placed there.
"""
if moduleLabel not in outputMap:
logging.info("Output module label missing from output map.")
return []
outputFilesets = []
for outputFileset in outputMap[moduleLabel]:
if merged == False and outputFileset["output_fileset"] != None:
outputFilesets.append(outputFileset["output_fileset"])
else:
if outputFileset["merged_output_fileset"] != None:
outputFilesets.append(
outputFileset["merged_output_fileset"])
return outputFilesets
def addFileToDBS(self, jobReportFile, task, errorDataset=False):
"""
_addFileToDBS_
Add a file that was output from a job to the DBS buffer.
"""
datasetInfo = jobReportFile["dataset"]
dbsFile = DBSBufferFile(lfn=jobReportFile["lfn"],
size=jobReportFile["size"],
events=jobReportFile["events"],
checksums=jobReportFile["checksums"],
status="NOTUPLOADED")
dbsFile.setAlgorithm(appName=datasetInfo["applicationName"],
appVer=datasetInfo["applicationVersion"],
appFam=jobReportFile["module_label"],
psetHash="GIBBERISH",
configContent=jobReportFile.get('configURL'))
if errorDataset:
dbsFile.setDatasetPath(
"/%s/%s/%s" %
(datasetInfo["primaryDataset"] + "-Error",
datasetInfo["processedDataset"], datasetInfo["dataTier"]))
else:
dbsFile.setDatasetPath(
"/%s/%s/%s" %
(datasetInfo["primaryDataset"],
datasetInfo["processedDataset"], datasetInfo["dataTier"]))
dbsFile.setValidStatus(
validStatus=jobReportFile.get("validStatus", None))
dbsFile.setProcessingVer(ver=jobReportFile.get('processingVer', None))
dbsFile.setAcquisitionEra(
era=jobReportFile.get('acquisitionEra', None))
dbsFile.setGlobalTag(globalTag=jobReportFile.get('globalTag', None))
#TODO need to find where to get the prep id
dbsFile.setPrepID(prep_id=jobReportFile.get('prep_id', None))
dbsFile['task'] = task
for run in jobReportFile["runs"]:
newRun = Run(runNumber=run.run)
newRun.extend(run.lumis)
dbsFile.addRun(newRun)
dbsFile.setLocation(pnn=list(jobReportFile["locations"])[0],
immediateSave=False)
self.dbsFilesToCreate.append(dbsFile)
return
def findDBSParents(self, lfn):
"""
_findDBSParents_
Find the parent of the file in DBS
This is meant to be called recursively
"""
parentsInfo = self.getParentInfoAction.execute(
[lfn],
conn=self.getDBConn(),
transaction=self.existingTransaction())
newParents = set()
for parentInfo in parentsInfo:
# This will catch straight to merge files that do not have redneck
# parents. We will mark the straight to merge file from the job
# as a child of the merged parent.
if int(parentInfo["merged"]) == 1:
newParents.add(parentInfo["lfn"])
elif parentInfo['gpmerged'] == None:
continue
# Handle the files that result from merge jobs that aren't redneck
# children. We have to setup parentage and then check on whether or
# not this file has any redneck children and update their parentage
# information.
elif int(parentInfo["gpmerged"]) == 1:
newParents.add(parentInfo["gplfn"])
# If that didn't work, we've reached the great-grandparents
# And we have to work via recursion
else:
parentSet = self.findDBSParents(lfn=parentInfo['gplfn'])
for parent in parentSet:
newParents.add(parent)
return newParents
def addFileToWMBS(self, jobType, fwjrFile, jobMask, task, jobID=None):
"""
_addFileToWMBS_
Add a file that was produced in a job to WMBS.
"""
fwjrFile["first_event"] = jobMask["FirstEvent"]
if fwjrFile["first_event"] == None:
fwjrFile["first_event"] = 0
if jobType == "Merge" and fwjrFile["module_label"] != "logArchive":
setattr(fwjrFile["fileRef"], 'merged', True)
fwjrFile["merged"] = True
wmbsFile = self.createFileFromDataStructsFile(file=fwjrFile,
jobID=jobID)
if jobType == "Merge":
self.wmbsMergeFilesToBuild.append(wmbsFile)
else:
self.wmbsFilesToBuild.append(wmbsFile)
if fwjrFile["merged"]:
self.addFileToDBS(
fwjrFile, task, jobType == "Repack"
and fwjrFile["size"] > self.maxAllowedRepackOutputSize)
return wmbsFile
def _mapLocation(self, fwkJobReport):
for file in fwkJobReport.getAllFileRefs():
if file and hasattr(file, 'location'):
file.location = self.phedex.getBestNodeName(
file.location, self.locLists)
def handleJob(self, jobID, fwkJobReport):
"""
_handleJob_
Figure out if a job was successful or not, handle it appropriately
(parse FWJR, update WMBS) and return the success status as a boolean
"""
jobSuccess = fwkJobReport.taskSuccessful()
outputMap = self.getOutputMapAction.execute(
jobID=jobID,
conn=self.getDBConn(),
transaction=self.existingTransaction())
jobType = self.getJobTypeAction.execute(
jobID=jobID,
conn=self.getDBConn(),
transaction=self.existingTransaction())
if jobSuccess:
fileList = fwkJobReport.getAllFiles()
# consistency check comparing outputMap to fileList
# they should match except for some limited special cases
outputModules = set([])
for fwjrFile in fileList:
outputModules.add(fwjrFile['outputModule'])
if set(outputMap.keys()) == outputModules:
pass
elif jobType == "LogCollect" and len(
outputMap.keys()) == 0 and outputModules == set(
['LogCollect']):
pass
elif jobType == "Merge" and set(outputMap.keys()) == set([
'Merged', 'MergedError', 'logArchive'
]) and outputModules == set(['Merged', 'logArchive']):
pass
elif jobType == "Merge" and set(outputMap.keys()) == set([
'Merged', 'MergedError', 'logArchive'
]) and outputModules == set(['MergedError', 'logArchive']):
pass
elif jobType == "Express" and set(
outputMap.keys()).difference(outputModules) == set(
['write_RAW']):
pass
else:
failJob = True
if jobType in ["Processing", "Production"]:
cmsRunSteps = 0
for step in fwkJobReport.listSteps():
if step.startswith("cmsRun"):
cmsRunSteps += 1
if cmsRunSteps > 1:
failJob = False
if failJob:
jobSuccess = False
logging.error(
"Job %d , list of expected outputModules does not match job report, failing job",
jobID)
logging.debug("Job %d , expected outputModules %s", jobID,
sorted(outputMap.keys()))
logging.debug("Job %d , fwjr outputModules %s", jobID,
sorted(outputModules))
fileList = fwkJobReport.getAllFilesFromStep(
step='logArch1')
else:
logging.debug(
"Job %d , list of expected outputModules does not match job report, accepted for multi-step CMSSW job",
jobID)
else:
fileList = fwkJobReport.getAllFilesFromStep(step='logArch1')
if jobSuccess:
logging.info("Job %d , handle successful job", jobID)
else:
logging.warning("Job %d , bad jobReport, failing job", jobID)
# make sure the task name is present in FWJR (recover from WMBS if needed)
if len(fileList) > 0:
if jobSuccess:
self.isTaskExistInFWJR(fwkJobReport, "success")
else:
self.isTaskExistInFWJR(fwkJobReport, "failed")
# special check for LogCollect jobs
skipLogCollect = False
if jobSuccess and jobType == "LogCollect":
for fwjrFile in fileList:
try:
# this assumes there is only one file for LogCollect jobs, not sure what happend if that changes
self.associateLogCollectToParentJobsInWMStats(
fwkJobReport, fwjrFile["lfn"],
fwkJobReport.getTaskName())
except Exception as ex:
skipLogCollect = True
logging.error(
"Error occurred: associating log collect location, will try again\n %s"
% str(ex))
break
# now handle the job (unless the special LogCollect check failed)
if not skipLogCollect:
wmbsJob = Job(id=jobID)
wmbsJob.load()
outputID = wmbsJob.loadOutputID()
wmbsJob.getMask()
wmbsJob["fwjr"] = fwkJobReport
if jobSuccess:
wmbsJob["outcome"] = "success"
else:
wmbsJob["outcome"] = "failure"
for fwjrFile in fileList:
logging.debug("Job %d , register output %s", jobID,
fwjrFile["lfn"])
wmbsFile = self.addFileToWMBS(jobType,
fwjrFile,
wmbsJob["mask"],
jobID=jobID,
task=fwkJobReport.getTaskName())
merged = fwjrFile['merged']
moduleLabel = fwjrFile["module_label"]
if merged:
self.mergedOutputFiles.append(wmbsFile)
self.filesetAssoc.append({
"lfn": wmbsFile["lfn"],
"fileset": outputID
})
# LogCollect jobs have no output fileset
if jobType == "LogCollect":
pass
# Repack jobs that wrote too large merged output skip output filesets
elif jobType == "Repack" and merged and wmbsFile[
"size"] > self.maxAllowedRepackOutputSize:
pass
else:
outputFilesets = self.outputFilesetsForJob(
outputMap, merged, moduleLabel)
for outputFileset in outputFilesets:
self.filesetAssoc.append({
"lfn": wmbsFile["lfn"],
"fileset": outputFileset
})
# Check if the job had any skipped files, put them in ACDC containers
# We assume full file processing (no job masks)
if jobSuccess:
skippedFiles = fwkJobReport.getAllSkippedFiles()
if skippedFiles and jobType not in ['LogCollect', 'Cleanup']:
self.jobsWithSkippedFiles[jobID] = skippedFiles
# Only save once job is done, and we're sure we made it through okay
self._mapLocation(wmbsJob['fwjr'])
if jobSuccess:
self.listOfJobsToSave.append(wmbsJob)
else:
self.listOfJobsToFail.append(wmbsJob)
return jobSuccess
def associateLogCollectToParentJobsInWMStats(self, fwkJobReport,
logAchiveLFN, task):
"""
_associateLogCollectToParentJobsInWMStats_
Associate a logArchive output to its parent job
"""
inputFileList = fwkJobReport.getAllInputFiles()
requestName = task.split('/')[1]
keys = []
for inputFile in inputFileList:
keys.append([requestName, inputFile["lfn"]])
resultRows = self.fwjrCouchDB.loadView(
"FWJRDump",
'jobsByOutputLFN',
options={"stale": "update_after"},
keys=keys)['rows']
if len(resultRows) > 0:
#get data from wmbs
parentWMBSJobIDs = []
for row in resultRows:
parentWMBSJobIDs.append({"jobid": row["value"]})
#update Job doc in wmstats
results = self.getJobInfoByID.execute(parentWMBSJobIDs)
parentJobNames = []
if isinstance(results, list):
for jobInfo in results:
parentJobNames.append(jobInfo['name'])
else:
parentJobNames.append(results['name'])
self.localWMStats.updateLogArchiveLFN(parentJobNames, logAchiveLFN)
else:
#TODO: if the couch db is consistent with DB this should be removed (checking resultRow > 0)
#It need to be failed and retried.
logging.error(
"job report is missing for updating log archive mapping\n Input file list\n %s"
% inputFileList)
return
def createMissingFWKJR(self,
parameters,
errorCode=999,
errorDescription='Failure of unknown type'):
"""
_createMissingFWJR_
Create a missing FWJR if the report can't be found by the code in the
path location.
"""
report = Report()
report.addError("cmsRun1", 84, errorCode, errorDescription)
report.data.cmsRun1.status = "Failed"
return report
def createFilesInDBSBuffer(self):
"""
_createFilesInDBSBuffer_
It does the actual job of creating things in DBSBuffer
WARNING: This assumes all files in a job have the same final location
"""
if len(self.dbsFilesToCreate) == 0:
# Whoops, nothing to do!
return
dbsFileTuples = []
dbsFileLoc = []
dbsCksumBinds = []
runLumiBinds = []
selfChecksums = None
jobLocations = set()
for dbsFile in self.dbsFilesToCreate:
# Append a tuple in the format specified by DBSBufferFiles.Add
# Also run insertDatasetAlgo
assocID = None
datasetAlgoPath = '%s:%s:%s:%s:%s:%s:%s:%s' % (
dbsFile['datasetPath'], dbsFile["appName"], dbsFile["appVer"],
dbsFile["appFam"], dbsFile["psetHash"],
dbsFile['processingVer'], dbsFile['acquisitionEra'],
dbsFile['globalTag'])
# First, check if this is in the cache
if datasetAlgoPath in self.datasetAlgoPaths:
for da in self.datasetAlgoID:
if da['datasetAlgoPath'] == datasetAlgoPath:
assocID = da['assocID']
break
if not assocID:
# Then we have to get it ourselves
try:
assocID = dbsFile.insertDatasetAlgo()
self.datasetAlgoPaths.append(datasetAlgoPath)
self.datasetAlgoID.append({
'datasetAlgoPath': datasetAlgoPath,
'assocID': assocID
})
except WMException:
raise
except Exception as ex:
msg = "Unhandled exception while inserting datasetAlgo: %s\n" % datasetAlgoPath
msg += str(ex)
logging.error(msg)
raise AccountantWorkerException(msg)
# Associate the workflow to the file using the taskPath and the requestName
# TODO: debug why it happens and then drop/recover these cases automatically
taskPath = dbsFile.get('task')
if not taskPath:
msg = "Can't do workflow association, report this error to a developer.\n"
msg += "DbsFile : %s" % str(dbsFile)
raise AccountantWorkerException(msg)
workflowName = taskPath.split('/')[1]
workflowPath = '%s:%s' % (workflowName, taskPath)
if workflowPath in self.workflowPaths:
for wf in self.workflowIDs:
if wf['workflowPath'] == workflowPath:
workflowID = wf['workflowID']
break
else:
result = self.dbsGetWorkflow.execute(
workflowName,
taskPath,
conn=self.getDBConn(),
transaction=self.existingTransaction())
workflowID = result['id']
self.workflowPaths.append(workflowPath)
self.workflowIDs.append({
'workflowPath': workflowPath,
'workflowID': workflowID
})
lfn = dbsFile['lfn']
selfChecksums = dbsFile['checksums']
jobLocation = dbsFile.getLocations()[0]
jobLocations.add(jobLocation)
dbsFileTuples.append((lfn, dbsFile['size'], dbsFile['events'],
assocID, dbsFile['status'], workflowID))
dbsFileLoc.append({'lfn': lfn, 'pnn': jobLocation})
if dbsFile['runs']:
runLumiBinds.append({'lfn': lfn, 'runs': dbsFile['runs']})
if selfChecksums:
# If we have checksums we have to create a bind
# For each different checksum
for entry in selfChecksums.keys():
dbsCksumBinds.append({
'lfn': lfn,
'cksum': selfChecksums[entry],
'cktype': entry
})
try:
diffLocation = jobLocations.difference(self.dbsLocations)
for jobLocation in diffLocation:
self.dbsInsertLocation.execute(
siteName=jobLocation,
conn=self.getDBConn(),
transaction=self.existingTransaction())
self.dbsLocations.add(jobLocation)
self.dbsCreateFiles.execute(files=dbsFileTuples,
conn=self.getDBConn(),
transaction=self.existingTransaction())
self.dbsSetLocation.execute(binds=dbsFileLoc,
conn=self.getDBConn(),
transaction=self.existingTransaction())
self.dbsSetChecksum.execute(bulkList=dbsCksumBinds,
conn=self.getDBConn(),
transaction=self.existingTransaction())
if len(runLumiBinds) > 0:
self.dbsSetRunLumi.execute(
file=runLumiBinds,
conn=self.getDBConn(),
transaction=self.existingTransaction())
except WMException:
raise
except Exception as ex:
msg = "Got exception while inserting files into DBSBuffer!\n"
msg += str(ex)
logging.error(msg)
logging.debug("Listing binds:")
logging.debug("jobLocation: %s\n" % jobLocation)
logging.debug("dbsFiles: %s\n" % dbsFileTuples)
logging.debug("dbsFileLoc: %s\n" % dbsFileLoc)
logging.debug("Checksum binds: %s\n" % dbsCksumBinds)
logging.debug("RunLumi binds: %s\n" % runLumiBinds)
raise AccountantWorkerException(msg)
# Now that we've created those files, clear the list
self.dbsFilesToCreate = []
return
def handleWMBSFiles(self, wmbsFilesToBuild, parentageBinds):
"""
_handleWMBSFiles_
Do what can be done in bulk in bulk
"""
if len(wmbsFilesToBuild) == 0:
# Nothing to do
return
runLumiBinds = []
fileCksumBinds = []
fileLocations = []
fileCreate = []
for wmbsFile in wmbsFilesToBuild:
lfn = wmbsFile['lfn']
if lfn == None:
continue
selfChecksums = wmbsFile['checksums']
# by jobType add to different parentage relation
# if it is the merge job, don't include the parentage on failed input files.
# otherwise parentage is set for all input files.
parentageBinds.append({'child': lfn, 'jobid': wmbsFile['jid']})
if wmbsFile['runs']:
runLumiBinds.append({'lfn': lfn, 'runs': wmbsFile['runs']})
if len(wmbsFile.getLocations()) > 0:
outpnn = wmbsFile.getLocations()[0]
if self.pnn_to_psn.get(outpnn, None):
fileLocations.append({'lfn': lfn, 'location': outpnn})
else:
msg = "PNN doesn't exist in wmbs_location_sename table: %s (investigate)" % outpnn
logging.error(msg)
raise AccountantWorkerException(msg)
if selfChecksums:
# If we have checksums we have to create a bind
# For each different checksum
for entry in selfChecksums.keys():
fileCksumBinds.append({
'lfn': lfn,
'cksum': selfChecksums[entry],
'cktype': entry
})
fileCreate.append([
lfn, wmbsFile['size'], wmbsFile['events'], None,
wmbsFile["first_event"], wmbsFile['merged']
])
if len(fileCreate) == 0:
return
try:
self.addFileAction.execute(files=fileCreate,
conn=self.getDBConn(),
transaction=self.existingTransaction())
if runLumiBinds:
self.setFileRunLumi.execute(
file=runLumiBinds,
conn=self.getDBConn(),
transaction=self.existingTransaction())
self.setFileAddChecksum.execute(
bulkList=fileCksumBinds,
conn=self.getDBConn(),
transaction=self.existingTransaction())
self.setFileLocation.execute(
lfn=fileLocations,
location=self.fileLocation,
conn=self.getDBConn(),
transaction=self.existingTransaction())
except WMException:
raise
except Exception as ex:
msg = "Error while adding files to WMBS!\n"
msg += str(ex)
logging.error(msg)
logging.debug("Printing binds: \n")
logging.debug("FileCreate binds: %s\n" % fileCreate)
logging.debug("Runlumi binds: %s\n" % runLumiBinds)
logging.debug("Checksum binds: %s\n" % fileCksumBinds)
logging.debug("FileLocation binds: %s\n" % fileLocations)
raise AccountantWorkerException(msg)
# Clear out finished files
wmbsFilesToBuild = []
return
def createFileFromDataStructsFile(self, file, jobID):
"""
_createFileFromDataStructsFile_
This function will create a WMBS File given a DataStructs file
"""
wmbsFile = File()
wmbsFile.update(file)
if isinstance(file["locations"], set):
pnn = list(file["locations"])[0]
elif isinstance(file["locations"], list):
if len(file['locations']) > 1:
logging.error(
"Have more then one location for a file in job %i" %
(jobID))
logging.error("Choosing location %s" % (file['locations'][0]))
pnn = file["locations"][0]
else:
pnn = file["locations"]
wmbsFile["locations"] = set()
if pnn != None:
wmbsFile.setLocation(pnn=pnn, immediateSave=False)
wmbsFile['jid'] = jobID
return wmbsFile
def handleDBSBufferParentage(self):
"""
_handleDBSBufferParentage_
Handle all the DBSBuffer Parentage in bulk if you can
"""
outputLFNs = [f['lfn'] for f in self.mergedOutputFiles]
bindList = []
for lfn in outputLFNs:
newParents = self.findDBSParents(lfn=lfn)
for parentLFN in newParents:
bindList.append({'child': lfn, 'parent': parentLFN})
# Now all the parents should exist
# Commit them to DBSBuffer
logging.info("About to commit all DBSBuffer Heritage information")
logging.info(len(bindList))
if len(bindList) > 0:
try:
self.dbsLFNHeritage.execute(
binds=bindList,
conn=self.getDBConn(),
transaction=self.existingTransaction())
except WMException:
raise
except Exception as ex:
msg = "Error while trying to handle the DBS LFN heritage\n"
msg += str(ex)
msg += "BindList: %s" % bindList
logging.error(msg)
raise AccountantWorkerException(msg)
return
def handleSkippedFiles(self):
"""
_handleSkippedFiles_
Handle all the skipped files in bulk,
the way it handles the skipped files
imposes an important restriction:
Skipped files should have been processed by a single job
in the task and no job mask exists in it.
This is suitable for jobs using ParentlessMergeBySize/FileBased/MinFileBased
splitting algorithms.
Here ACDC records and created and the file are moved
to wmbs_sub_files_failed from completed.
"""
jobList = self.getFullJobInfo.execute(
[{
'jobid': x
} for x in self.jobsWithSkippedFiles.keys()],
fileSelection=self.jobsWithSkippedFiles,
conn=self.getDBConn(),
transaction=self.existingTransaction())
self.dataCollection.failedJobs(jobList, useMask=False)
return
class ErrorHandlerPoller(BaseWorkerThread):
"""
Polls for Error Conditions, handles them
"""
def __init__(self, config):
"""
Initialise class members
"""
BaseWorkerThread.__init__(self)
self.config = config
myThread = threading.currentThread()
self.daoFactory = DAOFactory(package = "WMCore.WMBS",
logger = myThread.logger,
dbinterface = myThread.dbi)
self.changeState = ChangeState(self.config)
self.maxRetries = self.config.ErrorHandler.maxRetries
if type(self.maxRetries) != dict:
self.maxRetries = {'default' : self.maxRetries}
if 'default' not in self.maxRetries:
raise ErrorHandlerException('Max retries for the default job type must be specified')
self.maxProcessSize = getattr(self.config.ErrorHandler, 'maxProcessSize', 250)
self.exitCodes = getattr(self.config.ErrorHandler, 'failureExitCodes', [])
self.maxFailTime = getattr(self.config.ErrorHandler, 'maxFailTime', 32 * 3600)
self.readFWJR = getattr(self.config.ErrorHandler, 'readFWJR', False)
self.passCodes = getattr(self.config.ErrorHandler, 'passExitCodes', [])
self.getJobs = self.daoFactory(classname = "Jobs.GetAllJobs")
self.idLoad = self.daoFactory(classname = "Jobs.LoadFromIDWithType")
self.loadAction = self.daoFactory(classname = "Jobs.LoadForErrorHandler")
self.dataCollection = DataCollectionService(url = config.ACDC.couchurl,
database = config.ACDC.database)
# initialize the alert framework (if available - config.Alert present)
# self.sendAlert will be then be available
self.initAlerts(compName = "ErrorHandler")
return
def setup(self, parameters = None):
"""
Load DB objects required for queries
"""
# For now, does nothing
return
def terminate(self, params):
"""
_terminate_
Do one pass, then commit suicide
"""
logging.debug("terminating. doing one more pass before we die")
self.algorithm(params)
def exhaustJobs(self, jobList):
"""
_exhaustJobs_
Actually do the jobs exhaustion
"""
self.changeState.propagate(jobList, 'exhausted', 'retrydone')
# Remove all the files in the exhausted jobs.
logging.debug("About to fail input files for exhausted jobs")
for job in jobList:
job.failInputFiles()
# Do not build ACDC for utilitarian job types
jobList = [ job for job in jobList if job['type'] not in ['LogCollect','Cleanup'] ]
self.handleACDC(jobList)
return
def processRetries(self, jobList, state):
"""
_processRetries_
Actually do the retries
"""
logging.info("Processing retries for %d failed jobs of type %sfailed" % (len(jobList), state))
retrydoneJobs = []
cooloffJobs = []
passJobs = []
# Retries < max retry count
for job in jobList:
allowedRetries = self.maxRetries.get(job['type'], self.maxRetries['default'])
# Retries < allowed max retry count
if job['retry_count'] < allowedRetries and state != 'create':
cooloffJobs.append(job)
# Check if Retries >= allowed max retry count
elif job['retry_count'] >= allowedRetries or state == 'create':
retrydoneJobs.append(job)
msg = "Stopping retries for job %d" % job['id']
logging.debug(msg)
self.sendAlert(4, msg = msg)
logging.debug("JobInfo: %s" % job)
if self.readFWJR:
# Then we have to check each FWJR for exit status
cooloffJobs, passJobs, retrydoneFWJRJobs = self.readFWJRForErrors(cooloffJobs)
retrydoneJobs.extend(retrydoneFWJRJobs)
# Now to actually do something.
logging.debug("About to propagate jobs")
if len(retrydoneJobs) > 0:
self.changeState.propagate(retrydoneJobs, 'retrydone',
'%sfailed' % state, updatesummary = True)
if len(cooloffJobs) > 0:
self.changeState.propagate(cooloffJobs, '%scooloff' % state,
'%sfailed' % state, updatesummary = True)
if len(passJobs) > 0:
# Overwrite the transition states and move directly to created
self.changeState.propagate(passJobs, 'created', 'new')
return
def handleACDC(self, jobList):
"""
_handleACDC_
Do the ACDC creation and hope it works
"""
idList = [x['id'] for x in jobList]
logging.info("Starting to build ACDC with %i jobs" % len(idList))
logging.info("This operation will take some time...")
loadList = self.loadJobsFromListFull(idList)
for job in loadList:
job.getMask()
self.dataCollection.failedJobs(loadList)
return
def readFWJRForErrors(self, jobList):
"""
_readFWJRForErrors_
Check the FWJRs of the failed jobs
and determine those that can be retried
and which must be retried without going through cooloff.
Returns a triplet with cooloff, passed and exhausted jobs.
"""
cooloffJobs = []
passJobs = []
exhaustJobs = []
for job in jobList:
report = Report()
reportPath = job['fwjr_path']
if reportPath is None:
logging.error("No FWJR in job %i, ErrorHandler can't process it.\n Passing it to cooloff." % job['id'])
cooloffJobs.append(job)
continue
if not os.path.isfile(reportPath):
logging.error("Failed to find FWJR for job %i in location %s.\n Passing it to cooloff." % (job['id'], reportPath))
cooloffJobs.append(job)
continue
try:
report.load(reportPath)
# First let's check the time conditions
times = report.getFirstStartLastStop()
startTime = None
stopTime = None
if times is not None:
startTime = times['startTime']
stopTime = times['stopTime']
if startTime == None or stopTime == None:
# We have no information to make a decision, keep going.
logging.debug("No start, stop times for steps for job %i" % job['id'])
elif stopTime - startTime > self.maxFailTime:
msg = "Job %i exhausted after running on node for %i seconds" % (job['id'], stopTime - startTime)
logging.debug(msg)
exhaustJobs.append(job)
continue
if len([x for x in report.getExitCodes() if x in self.exitCodes]):
msg = "Job %i exhausted due to a bad exit code (%s)" % (job['id'], str(report.getExitCodes()))
logging.error(msg)
self.sendAlert(4, msg = msg)
exhaustJobs.append(job)
continue
if len([x for x in report.getExitCodes() if x in self.passCodes]):
msg = "Job %i restarted immediately due to an exit code (%s)" % (job['id'], str(report.getExitCodes()))
passJobs.append(job)
continue
cooloffJobs.append(job)
except Exception as ex:
logging.warning("Exception while trying to check jobs for failures!")
logging.warning(str(ex))
logging.warning("Ignoring and sending job to cooloff")
cooloffJobs.append(job)
return cooloffJobs, passJobs, exhaustJobs
def handleRetryDoneJobs(self, jobList):
"""
_handleRetryDoneJobs_
"""
myThread = threading.currentThread()
logging.debug("About to process %d retry done jobs" % len(jobList))
myThread.transaction.begin()
self.exhaustJobs(jobList)
myThread.transaction.commit()
return
def handleFailedJobs(self, jobList, state):
"""
_handleFailedJobs_
"""
myThread = threading.currentThread()
logging.debug("About to process %d failures" % len(jobList))
myThread.transaction.begin()
self.processRetries(jobList, state)
myThread.transaction.commit()
return
def handleErrors(self):
"""
Queries DB for all watched filesets, if matching filesets become
available, create the subscriptions
"""
# Run over created, submitted and executed job failures
failure_states = [ 'create', 'submit', 'job' ]
for state in failure_states:
idList = self.getJobs.execute(state = "%sfailed" % state)
logging.info("Found %d failed jobs in state %sfailed" % (len(idList), state))
while len(idList) > 0:
tmpList = idList[:self.maxProcessSize]
idList = idList[self.maxProcessSize:]
jobList = self.loadJobsFromList(tmpList)
self.handleFailedJobs(jobList, state)
# Run over jobs done with retries
idList = self.getJobs.execute(state = 'retrydone')
logging.info("Found %d jobs done with all retries" % len(idList))
while len(idList) > 0:
tmpList = idList[:self.maxProcessSize]
idList = idList[self.maxProcessSize:]
jobList = self.loadJobsFromList(tmpList)
self.handleRetryDoneJobs(jobList)
return
def loadJobsFromList(self, idList):
"""
_loadJobsFromList_
Load jobs in bulk
"""
binds = []
for jobID in idList:
binds.append({"jobid": jobID})
results = self.idLoad.execute(jobID = binds)
# You have to have a list
if type(results) == dict:
results = [results]
listOfJobs = []
for entry in results:
# One job per entry
tmpJob = Job(id = entry['id'])
tmpJob.update(entry)
listOfJobs.append(tmpJob)
return listOfJobs
def loadJobsFromListFull(self, idList):
"""
_loadJobsFromList_
Load jobs in bulk.
Include the full metadata.
"""
binds = []
for jobID in idList:
binds.append({"jobid": jobID})
results = self.loadAction.execute(jobID = binds)
# You have to have a list
if type(results) == dict:
results = [results]
listOfJobs = []
for entry in results:
# One job per entry
tmpJob = Job(id = entry['id'])
tmpJob.update(entry)
listOfJobs.append(tmpJob)
return listOfJobs
def algorithm(self, parameters = None):
"""
Performs the handleErrors method, looking for each type of failure
And deal with it as desired.
"""
logging.debug("Running error handling algorithm")
myThread = threading.currentThread()
try:
self.handleErrors()
except WMException as ex:
try:
myThread.transaction.rollback()
except:
pass
raise
except CouchConnectionError as ex:
msg = "Caught CouchConnectionError exception in ErrorHandler\n"
msg += "transactions postponed until the next polling cycle\n"
msg += str(ex)
logging.exception(msg)
except Exception as ex:
msg = "Caught exception in ErrorHandler\n"
msg += str(ex)
msg += str(traceback.format_exc())
msg += "\n\n"
logging.error(msg)
self.sendAlert(6, msg = msg)
if getattr(myThread, 'transaction', None) != None \
and getattr(myThread.transaction, 'transaction', None) != None:
myThread.transaction.rollback()
raise ErrorHandlerException(msg)
def algorithm(self, *args, **kwargs):
"""
_algorithm_
Split files into a number of lumis per job
Allow a flag to determine if we split files between jobs
"""
avgEventsPerJob = int(kwargs.get('events_per_job', 5000))
jobLimit = int(kwargs.get('job_limit', 0))
jobTimeLimit = int(
kwargs.get('job_time_limit', self.defaultJobTimeLimit))
totalEvents = int(kwargs.get('total_events', 0))
splitOnFile = bool(kwargs.get('halt_job_on_file_boundaries', False))
self.collectionName = kwargs.get('collectionName', None)
splitOnRun = kwargs.get('splitOnRun', True)
getParents = kwargs.get('include_parents', False)
runWhitelist = kwargs.get('runWhitelist', [])
runs = kwargs.get('runs', None)
lumis = kwargs.get('lumis', None)
applyLumiCorrection = bool(kwargs.get('applyLumiCorrection', False))
deterministicPileup = kwargs.get('deterministicPileup', False)
timePerEvent, sizePerEvent, memoryRequirement = \
self.getPerformanceParameters(kwargs.get('performance', {}))
eventsPerLumiInDataset = 0
if avgEventsPerJob <= 0:
msg = "events_per_job parameter must be positive. Its value is: %d" % avgEventsPerJob
raise RuntimeError(msg)
if self.package == 'WMCore.WMBS':
self.loadRunLumi = self.daoFactory(
classname="Files.GetBulkRunLumi")
if deterministicPileup:
getJobNumber = self.daoFactory(
classname="Jobs.GetNumberOfJobsPerWorkflow")
self.nJobs = getJobNumber.execute(
workflow=self.subscription.getWorkflow().id)
logging.info(
'Creating jobs in DeterministicPileup mode for %s',
self.subscription.workflowName())
goodRunList = {}
if runs and lumis:
goodRunList = buildLumiMask(runs, lumis)
# If we have runLumi info, we need to load it from couch
if self.collectionName:
try:
from WMCore.ACDC.DataCollectionService import DataCollectionService
couchURL = kwargs.get('couchURL')
couchDB = kwargs.get('couchDB')
filesetName = kwargs.get('filesetName')
logging.info('Creating jobs for ACDC fileset %s', filesetName)
dcs = DataCollectionService(couchURL, couchDB)
goodRunList = dcs.getLumiWhitelist(self.collectionName,
filesetName)
except Exception as ex:
msg = "Exception while trying to load goodRunList. "
msg += "Refusing to create any jobs.\nDetails: %s" % str(ex)
logging.exception(msg)
return
lDict = self.getFilesSortedByLocation(avgEventsPerJob)
if not lDict:
logging.info(
"There are not enough events/files to be splitted. Trying again next cycle"
)
return
locationDict = {}
for key in lDict.keys():
newlist = []
# First we need to load the data
if self.loadRunLumi:
fileLumis = self.loadRunLumi.execute(files=lDict[key])
if not fileLumis:
logging.warning(
"Empty fileLumis dict for workflow %s, subs %s.",
self.subscription.workflowName(),
self.subscription['id'])
for f in lDict[key]:
lumiDict = fileLumis.get(f['id'], {})
for run in lumiDict.keys():
f.addRun(run=Run(run, *lumiDict[run]))
for f in lDict[key]:
if len(f['runs']) == 0:
continue
f['runs'] = sorted(f['runs'])
f['lumiCount'] = 0
for run in f['runs']:
run.lumis.sort()
f['lumiCount'] += len(run.lumis)
f['lowestRun'] = f['runs'][0]
# Do average event per lumi calculation
if f['lumiCount']:
f['avgEvtsPerLumi'] = round(
float(f['events']) / f['lumiCount'])
if deterministicPileup:
# We assume that all lumis are equal in the dataset
eventsPerLumiInDataset = f['avgEvtsPerLumi']
else:
# No lumis in the file, ignore it
continue
newlist.append(f)
locationDict[key] = sorted(newlist,
key=operator.itemgetter('lowestRun'))
totalJobs = 0
lastLumi = None
firstLumi = None
lastRun = None
lumisInJob = 0
totalAvgEventCount = 0
currentJobAvgEventCount = 0
stopTask = False
self.lumiChecker = LumiChecker(applyLumiCorrection)
for location in locationDict:
# For each location, we need a new jobGroup
self.newGroup()
stopJob = True
for f in locationDict[location]:
if getParents:
parentLFNs = self.findParent(lfn=f['lfn'])
for lfn in parentLFNs:
parent = File(lfn=lfn)
f['parents'].add(parent)
lumisInJobInFile = 0
updateSplitOnJobStop = False
failNextJob = False
# If estimated job time is higher the job time limit (condor limit)
# and it's only one lumi then ditch that lumi
timePerLumi = f['avgEvtsPerLumi'] * timePerEvent
if timePerLumi > jobTimeLimit and f['lumiCount'] == 1:
failNextJob = True
stopJob = True
lumisPerJob = 1
elif splitOnFile:
# Then we have to split on every boundary
stopJob = True
# Check the average number of events per lumi in this file
# Adapt the lumis per job to match the target conditions
if f['avgEvtsPerLumi']:
# If there are events in the file
ratio = float(avgEventsPerJob) / f['avgEvtsPerLumi']
lumisPerJob = max(int(math.floor(ratio)), 1)
else:
# Zero event file, then the ratio goes to infinity. Computers don't like that
lumisPerJob = f['lumiCount']
else:
# Analyze how many events does this job already has
# Check how many we want as target, include as many lumi sections as possible
updateSplitOnJobStop = True
eventsRemaining = max(
avgEventsPerJob - currentJobAvgEventCount, 0)
if f['avgEvtsPerLumi']:
lumisAllowed = int(
math.floor(
float(eventsRemaining) / f['avgEvtsPerLumi']))
else:
lumisAllowed = f['lumiCount']
lumisPerJob = max(lumisInJob + lumisAllowed, 1)
for run in f['runs']:
if not isGoodRun(goodRunList=goodRunList, run=run.run):
# Then skip this one
continue
if len(runWhitelist) > 0 and not run.run in runWhitelist:
# Skip due to run whitelist
continue
firstLumi = None
if splitOnRun and run.run != lastRun:
# Then we need to kill this job and get a new one
stopJob = True
# Now loop over the lumis
for lumi in run:
if (not isGoodLumi(goodRunList, run=run.run, lumi=lumi)
or self.lumiChecker.isSplitLumi(
run.run, lumi, f)):
# Kill the chain of good lumis
# Skip this lumi
if firstLumi != None and firstLumi != lumi:
self.currentJob['mask'].addRunAndLumis(
run=run.run, lumis=[firstLumi, lastLumi])
eventsAdded = ((lastLumi - firstLumi + 1) *
f['avgEvtsPerLumi'])
runAddedTime = eventsAdded * timePerEvent
runAddedSize = eventsAdded * sizePerEvent
self.currentJob.addResourceEstimates(
jobTime=runAddedTime, disk=runAddedSize)
firstLumi = None
lastLumi = None
continue
# You have to kill the lumi chain if they're not continuous
if lastLumi and not lumi == lastLumi + 1:
self.currentJob['mask'].addRunAndLumis(
run=run.run, lumis=[firstLumi, lastLumi])
eventsAdded = ((lastLumi - firstLumi + 1) *
f['avgEvtsPerLumi'])
runAddedTime = eventsAdded * timePerEvent
runAddedSize = eventsAdded * sizePerEvent
self.currentJob.addResourceEstimates(
jobTime=runAddedTime, disk=runAddedSize)
firstLumi = None
lastLumi = None
if firstLumi is None:
# Set the first lumi in the run
firstLumi = lumi
# If we're full, end the job
if lumisInJob == lumisPerJob:
stopJob = True
# Actually do the new job creation
if stopJob:
if firstLumi != None and lastLumi != None and lastRun != None:
self.currentJob['mask'].addRunAndLumis(
run=lastRun, lumis=[firstLumi, lastLumi])
eventsAdded = ((lastLumi - firstLumi + 1) *
f['avgEvtsPerLumi'])
runAddedTime = eventsAdded * timePerEvent
runAddedSize = eventsAdded * sizePerEvent
self.currentJob.addResourceEstimates(
jobTime=runAddedTime, disk=runAddedSize)
msg = None
if failNextJob:
msg = "File %s has a single lumi %s, in run %s " % (
f['lfn'], lumi, run.run)
msg += "with too many events %d and it woud take %d sec to run" \
% (f['events'], timePerLumi)
self.lumiChecker.closeJob(self.currentJob)
self.newJob(name=self.getJobName(),
failedJob=failNextJob,
failedReason=msg)
if deterministicPileup:
skipEvents = (
self.nJobs -
1) * lumisPerJob * eventsPerLumiInDataset
self.currentJob.addBaggageParameter(
"skipPileupEvents", skipEvents)
self.currentJob.addResourceEstimates(
memory=memoryRequirement)
failNextJob = False
firstLumi = lumi
lumisInJob = 0
lumisInJobInFile = 0
currentJobAvgEventCount = 0
totalJobs += 1
if jobLimit and totalJobs > jobLimit:
msg = "Job limit of {0} jobs exceeded.".format(
jobLimit)
raise RuntimeError(msg)
# Add the file to new jobs
self.currentJob.addFile(f)
if updateSplitOnJobStop:
# Then we were carrying from a previous file
# Reset calculations for this file
updateSplitOnJobStop = False
if f['avgEvtsPerLumi']:
ratio = float(
avgEventsPerJob) / f['avgEvtsPerLumi']
lumisPerJob = max(int(math.floor(ratio)),
1)
else:
lumisPerJob = f['lumiCount']
lumisInJob += 1
lumisInJobInFile += 1
lastLumi = lumi
stopJob = False
lastRun = run.run
totalAvgEventCount += f['avgEvtsPerLumi']
if self.currentJob and not f in self.currentJob[
'input_files']:
self.currentJob.addFile(f)
# We stop here if there are more total events than requested.
if totalEvents > 0 and totalAvgEventCount >= totalEvents:
stopTask = True
break
if firstLumi != None and lastLumi != None:
# Add this run to the mask
self.currentJob['mask'].addRunAndLumis(
run=run.run, lumis=[firstLumi, lastLumi])
eventsAdded = ((lastLumi - firstLumi + 1) *
f['avgEvtsPerLumi'])
runAddedTime = eventsAdded * timePerEvent
runAddedSize = eventsAdded * sizePerEvent
self.currentJob.addResourceEstimates(
jobTime=runAddedTime, disk=runAddedSize)
firstLumi = None
lastLumi = None
if stopTask:
break
if not splitOnFile:
currentJobAvgEventCount += f[
'avgEvtsPerLumi'] * lumisInJobInFile
if stopTask:
break
if stopTask:
break
self.lumiChecker.closeJob(self.currentJob)
self.lumiChecker.fixInputFiles()
return
def algorithm(self, *args, **kwargs):
"""
_algorithm_
Split files into a number of lumis per job
Allow a flag to determine if we split files between jobs
"""
lumisPerJob = int(kwargs.get('lumis_per_job', 1))
totalLumis = int(kwargs.get('total_lumis', 0))
splitOnFile = bool(kwargs.get('halt_job_on_file_boundaries', False))
self.collectionName = kwargs.get('collectionName', None)
splitOnRun = kwargs.get('splitOnRun', True)
getParents = kwargs.get('include_parents', False)
runWhitelist = kwargs.get('runWhitelist', [])
runs = kwargs.get('runs', None)
lumis = kwargs.get('lumis', None)
deterministicPileup = kwargs.get('deterministicPileup', False)
applyLumiCorrection = bool(kwargs.get('applyLumiCorrection', False))
eventsPerLumiInDataset = 0
if self.package == 'WMCore.WMBS':
self.loadRunLumi = self.daoFactory(
classname="Files.GetBulkRunLumi")
if deterministicPileup:
getJobNumber = self.daoFactory(
classname="Jobs.GetNumberOfJobsPerWorkflow")
self.nJobs = getJobNumber.execute(
workflow=self.subscription.getWorkflow().id)
timePerEvent, sizePerEvent, memoryRequirement = \
self.getPerformanceParameters(kwargs.get('performance', {}))
goodRunList = {}
if runs and lumis:
goodRunList = buildLumiMask(runs, lumis)
# If we have runLumi info, we need to load it from couch
if self.collectionName:
try:
from WMCore.ACDC.DataCollectionService import DataCollectionService
couchURL = kwargs.get('couchURL')
couchDB = kwargs.get('couchDB')
filesetName = kwargs.get('filesetName')
logging.info('Creating jobs for ACDC fileset %s', filesetName)
dcs = DataCollectionService(couchURL, couchDB)
goodRunList = dcs.getLumiWhitelist(self.collectionName,
filesetName)
except Exception as ex:
msg = "Exception while trying to load goodRunList. "
msg += "Refusing to create any jobs.\nDetails: %s" % str(ex)
logging.exception(msg)
return
lDict = self.getFilesSortedByLocation(lumisPerJob)
if not lDict:
logging.info(
"There are not enough lumis/files to be splitted. Trying again next cycle"
)
return
locationDict = {}
for key in lDict.keys():
newlist = []
for f in lDict[key]:
# if hasattr(f, 'loadData'):
# f.loadData()
if len(f['runs']) == 0:
continue
f['lumiCount'] = 0
f['runs'] = sorted(f['runs'])
for run in f['runs']:
run.lumis.sort()
f['lumiCount'] += len(run.lumis)
f['lowestRun'] = f['runs'][0]
# Do average event per lumi calculation
if f['lumiCount']:
f['avgEvtsPerLumi'] = round(
float(f['events']) / f['lumiCount'])
if deterministicPileup:
# We assume that all lumis are equal in the dataset
eventsPerLumiInDataset = f['avgEvtsPerLumi']
else:
# No lumis in the file, ignore it
continue
newlist.append(f)
locationDict[key] = sorted(newlist,
key=operator.itemgetter('lowestRun'))
# Split files into jobs with each job containing
# EXACTLY lumisPerJob number of lumis (except for maybe the last one)
totalJobs = 0
lastLumi = None
firstLumi = None
stopJob = True
stopTask = False
lastRun = None
lumisInJob = 0
lumisInTask = 0
self.lumiChecker = LumiChecker(applyLumiCorrection)
for location in locationDict.keys():
# For each location, we need a new jobGroup
self.newGroup()
stopJob = True
for f in locationDict[location]:
if getParents:
parentLFNs = self.findParent(lfn=f['lfn'])
for lfn in parentLFNs:
parent = File(lfn=lfn)
f['parents'].add(parent)
if splitOnFile:
# Then we have to split on every boundary
stopJob = True
for run in f['runs']:
if not isGoodRun(goodRunList=goodRunList, run=run.run):
# Then skip this one
continue
if len(runWhitelist) > 0 and not run.run in runWhitelist:
# Skip due to run whitelist
continue
firstLumi = None
if splitOnRun and run.run != lastRun:
# Then we need to kill this job and get a new one
stopJob = True
# Now loop over the lumis
for lumi in run:
# splitLumi checks if the lumi is split across jobs
if (not isGoodLumi(goodRunList, run=run.run, lumi=lumi)
or self.lumiChecker.isSplitLumi(
run.run, lumi, f)):
# Kill the chain of good lumis
# Skip this lumi
if firstLumi != None and firstLumi != lumi:
self.currentJob['mask'].addRunAndLumis(
run=run.run, lumis=[firstLumi, lastLumi])
addedEvents = ((lastLumi - firstLumi + 1) *
f['avgEvtsPerLumi'])
runAddedTime = addedEvents * timePerEvent
runAddedSize = addedEvents * sizePerEvent
self.currentJob.addResourceEstimates(
jobTime=runAddedTime, disk=runAddedSize)
firstLumi = None
lastLumi = None
continue
# You have to kill the lumi chain if they're not continuous
if lastLumi and not lumi == lastLumi + 1:
self.currentJob['mask'].addRunAndLumis(
run=run.run, lumis=[firstLumi, lastLumi])
addedEvents = ((lastLumi - firstLumi + 1) *
f['avgEvtsPerLumi'])
runAddedTime = addedEvents * timePerEvent
runAddedSize = addedEvents * sizePerEvent
self.currentJob.addResourceEstimates(
jobTime=runAddedTime, disk=runAddedSize)
firstLumi = None
lastLumi = None
if firstLumi is None:
# Set the first lumi in the run
firstLumi = lumi
# If we're full, end the job
if lumisInJob == lumisPerJob:
stopJob = True
# Actually do the new job creation
if stopJob:
if firstLumi != None and lastLumi != None and lastRun != None:
self.currentJob['mask'].addRunAndLumis(
run=lastRun, lumis=[firstLumi, lastLumi])
addedEvents = ((lastLumi - firstLumi + 1) *
f['avgEvtsPerLumi'])
runAddedTime = addedEvents * timePerEvent
runAddedSize = addedEvents * sizePerEvent
self.currentJob.addResourceEstimates(
jobTime=runAddedTime, disk=runAddedSize)
# before creating a new job add the lumis of the current one to the checker
self.lumiChecker.closeJob(self.currentJob)
self.newJob(name=self.getJobName())
self.currentJob.addResourceEstimates(
memory=memoryRequirement)
if deterministicPileup:
skipEvents = (
self.nJobs -
1) * lumisPerJob * eventsPerLumiInDataset
self.currentJob.addBaggageParameter(
"skipPileupEvents", skipEvents)
firstLumi = lumi
lumisInJob = 0
totalJobs += 1
# Add the file to new jobs
self.currentJob.addFile(f)
lumisInJob += 1
lumisInTask += 1
lastLumi = lumi
stopJob = False
lastRun = run.run
if self.currentJob and not f in self.currentJob[
'input_files']:
self.currentJob.addFile(f)
if totalLumis > 0 and lumisInTask >= totalLumis:
stopTask = True
break
if firstLumi != None and lastLumi != None:
# Add this run to the mask
self.currentJob['mask'].addRunAndLumis(
run=run.run, lumis=[firstLumi, lastLumi])
addedEvents = ((lastLumi - firstLumi + 1) *
f['avgEvtsPerLumi'])
runAddedTime = addedEvents * timePerEvent
runAddedSize = addedEvents * sizePerEvent
self.currentJob.addResourceEstimates(
jobTime=runAddedTime, disk=runAddedSize)
firstLumi = None
lastLumi = None
if stopTask:
break
if stopTask:
break
if stopTask:
break
self.lumiChecker.closeJob(self.currentJob)
self.lumiChecker.fixInputFiles()
return
class ErrorHandlerTest(unittest.TestCase):
"""
TestCase for TestErrorHandler module
"""
def setUp(self):
"""
setup for test.
"""
myThread = threading.currentThread()
self.testInit = TestInitCouchApp(__file__)
self.testInit.setLogging()
self.testInit.setDatabaseConnection(destroyAllDatabase = True)
self.testInit.setSchema(customModules = ["WMCore.WMBS"],
useDefault = False)
self.testInit.setupCouch("errorhandler_t", "GroupUser", "ACDC")
self.testInit.setupCouch("errorhandler_t_jd/jobs", "JobDump")
self.testInit.setupCouch("errorhandler_t_jd/fwjrs", "FWJRDump")
self.daofactory = DAOFactory(package = "WMCore.WMBS",
logger = myThread.logger,
dbinterface = myThread.dbi)
self.getJobs = self.daofactory(classname = "Jobs.GetAllJobs")
self.setJobTime = self.daofactory(classname = "Jobs.SetStateTime")
locationAction = self.daofactory(classname = "Locations.New")
locationAction.execute(siteName = "malpaquet", seName = "malpaquet")
self.testDir = self.testInit.generateWorkDir()
self.nJobs = 10
self.dataCS = DataCollectionService(url = self.testInit.couchUrl,
database = "errorhandler_t")
return
def tearDown(self):
"""
Database deletion
"""
self.testInit.clearDatabase()
self.testInit.delWorkDir()
self.testInit.tearDownCouch()
return
def getConfig(self):
"""
_getConfig_
"""
config = Configuration()
# First the general stuff
config.section_("General")
config.General.workDir = os.getenv("TESTDIR", self.testDir)
config.section_("CoreDatabase")
config.CoreDatabase.connectUrl = os.getenv("DATABASE")
config.CoreDatabase.socket = os.getenv("DBSOCK")
config.component_("ErrorHandler")
# The log level of the component.
config.ErrorHandler.logLevel = 'DEBUG'
# The namespace of the component
config.ErrorHandler.namespace = 'WMComponent.ErrorHandler.ErrorHandler'
# maximum number of threads we want to deal
# with messages per pool.
config.ErrorHandler.maxThreads = 30
# maximum number of retries we want for job
config.ErrorHandler.maxRetries = 5
# The poll interval at which to look for failed jobs
config.ErrorHandler.pollInterval = 60
# JobStateMachine
config.component_('JobStateMachine')
config.JobStateMachine.couchurl = os.getenv('COUCHURL', None)
config.JobStateMachine.couchDBName = "errorhandler_t_jd"
config.section_('ACDC')
config.ACDC.couchurl = self.testInit.couchUrl
config.ACDC.database = "errorhandler_t"
return config
def createWorkload(self, workloadName = 'Test', emulator = True):
"""
_createTestWorkload_
Creates a test workload for us to run on, hold the basic necessities.
"""
workload = testWorkload("Tier1ReReco")
rereco = workload.getTask("ReReco")
# Add RequestManager stuff
workload.data.request.section_('schema')
workload.data.request.schema.Requestor = 'nobody'
workload.data.request.schema.Group = 'testers'
taskMaker = TaskMaker(workload, os.path.join(self.testDir, 'workloadTest'))
taskMaker.skipSubscription = True
taskMaker.processWorkload()
return workload
def createTestJobGroup(self, nJobs = 10, retry_count = 1,
workloadPath = 'test', fwjrPath = None,
workloadName = makeUUID()):
"""
Creates a group of several jobs
"""
myThread = threading.currentThread()
myThread.transaction.begin()
testWorkflow = Workflow(spec = workloadPath, owner = "cmsdataops", group = "cmsdataops",
name = workloadName, task="/TestWorkload/ReReco")
testWorkflow.create()
testWMBSFileset = Fileset(name = "TestFileset")
testWMBSFileset.create()
testSubscription = Subscription(fileset = testWMBSFileset,
workflow = testWorkflow)
testSubscription.create()
testJobGroup = JobGroup(subscription = testSubscription)
testJobGroup.create()
testFile0 = File(lfn = "/this/is/a/parent", size = 1024, events = 10)
testFile0.addRun(Run(10, *[12312]))
testFile0.setLocation('malpaquet')
testFileA = File(lfn = "/this/is/a/lfnA", size = 1024, events = 10,
first_event = 88, last_event = 99)
testFileA.addRun(Run(10, *[12312, 12313]))
testFileA.setLocation('malpaquet')
testFileB = File(lfn = "/this/is/a/lfnB", size = 1024, events = 10,
first_event = 88, last_event = 99)
testFileB.addRun(Run(10, *[12314, 12315, 12316]))
testFileB.setLocation('malpaquet')
testFile0.create()
testFileA.create()
testFileB.create()
testFileA.addParent(lfn = "/this/is/a/parent")
testFileB.addParent(lfn = "/this/is/a/parent")
for i in range(0, nJobs):
testJob = Job(name = makeUUID())
testJob['retry_count'] = retry_count
testJob['retry_max'] = 10
testJob['mask'].addRunAndLumis(run = 10, lumis = [12312])
testJob['mask'].addRunAndLumis(run = 10, lumis = [12314, 12316])
testJob['mask']['FirstEvent'] = 100
testJob['cache_dir'] = os.path.join(self.testDir, testJob['name'])
testJob['fwjr_path'] = fwjrPath
os.mkdir(testJob['cache_dir'])
testJobGroup.add(testJob)
testJob.create(group = testJobGroup)
testJob.addFile(testFileA)
testJob.addFile(testFileB)
testJob.save()
testJobGroup.commit()
testSubscription.acquireFiles(files = [testFileA, testFileB])
testSubscription.save()
myThread.transaction.commit()
return testJobGroup
def testA_Create(self):
"""
WMComponent_t.ErrorHandler_t.ErrorHandler_t:testCreate()
Mimics creation of component and test jobs failed in create stage.
"""
workloadName = 'TestWorkload'
workload = self.createWorkload(workloadName = workloadName)
workloadPath = os.path.join(self.testDir, 'workloadTest', 'TestWorkload',
'WMSandbox', 'WMWorkload.pkl')
testJobGroup = self.createTestJobGroup(nJobs = self.nJobs,
workloadPath = workloadPath,
workloadName = workloadName)
config = self.getConfig()
changer = ChangeState(config)
changer.propagate(testJobGroup.jobs, 'createfailed', 'new')
idList = self.getJobs.execute(state = 'CreateFailed')
self.assertEqual(len(idList), self.nJobs)
testErrorHandler = ErrorHandlerPoller(config)
testErrorHandler.setup(None)
testErrorHandler.algorithm(None)
idList = self.getJobs.execute(state = 'CreateFailed')
self.assertEqual(len(idList), 0)
idList = self.getJobs.execute(state = 'CreateCooloff')
self.assertEqual(len(idList), self.nJobs)
changer.propagate(testJobGroup.jobs, 'new', 'CreateCooloff')
changer.propagate(testJobGroup.jobs, 'createfailed', 'new')
# Now exhaust them
for job in testJobGroup.jobs:
job['retry_count'] = 6
job.save()
testErrorHandler.algorithm(None)
idList = self.getJobs.execute(state = 'Exhausted')
self.assertEqual(len(idList), self.nJobs)
# Check that it showed up in ACDC
collection = self.dataCS.getDataCollection(workloadName)
# Now look at what's inside
self.assertTrue(len(collection['filesets']) > 0)
for fileset in collection["filesets"]:
counter = 0
for f in fileset.listFiles():
counter += 1
self.assertTrue(f['lfn'] in ["/this/is/a/lfnA", "/this/is/a/lfnB"])
self.assertEqual(f['events'], 10)
self.assertEqual(f['size'], 1024)
self.assertEqual(f['parents'], [u'/this/is/a/parent'])
self.assertTrue(f['runs'][0]['lumis'] in [[12312], [12314, 12315, 12316]],
"Unknown lumi %s" % f['runs'][0]['lumis'])
self.assertTrue(f['merged'], 1)
self.assertTrue(f['first_event'], 88)
self.assertTrue(f['last_event'], 99)
self.assertEqual(counter, 20)
return
def testB_Submit(self):
"""
WMComponent_t.ErrorHandler_t.ErrorHandler_t:testSubmit()
Mimics creation of component and test jobs failed in submit stage.
"""
workloadName = 'TestWorkload'
workload = self.createWorkload(workloadName = workloadName)
workloadPath = os.path.join(self.testDir, 'workloadTest', 'TestWorkload',
'WMSandbox', 'WMWorkload.pkl')
testJobGroup = self.createTestJobGroup(nJobs = self.nJobs,
workloadPath = workloadPath)
config = self.getConfig()
changer = ChangeState(config)
changer.propagate(testJobGroup.jobs, 'created', 'new')
changer.propagate(testJobGroup.jobs, 'submitfailed', 'created')
idList = self.getJobs.execute(state = 'SubmitFailed')
self.assertEqual(len(idList), self.nJobs)
testErrorHandler = ErrorHandlerPoller(config)
testErrorHandler.setup(None)
testErrorHandler.algorithm(None)
idList = self.getJobs.execute(state = 'SubmitFailed')
self.assertEqual(len(idList), 0)
idList = self.getJobs.execute(state = 'SubmitCooloff')
self.assertEqual(len(idList), self.nJobs)
return
def testC_Jobs(self):
"""
WMComponent_t.ErrorHandler_t.ErrorHandler_t.testJobs()
Mimics creation of component and test jobs failed in execute stage.
"""
workloadName = 'TestWorkload'
workload = self.createWorkload(workloadName = workloadName)
workloadPath = os.path.join(self.testDir, 'workloadTest', 'TestWorkload',
'WMSandbox', 'WMWorkload.pkl')
testJobGroup = self.createTestJobGroup(nJobs = self.nJobs,
workloadPath = workloadPath)
config = self.getConfig()
changer = ChangeState(config)
changer.propagate(testJobGroup.jobs, 'created', 'new')
changer.propagate(testJobGroup.jobs, 'executing', 'created')
changer.propagate(testJobGroup.jobs, 'complete', 'executing')
changer.propagate(testJobGroup.jobs, 'jobfailed', 'complete')
idList = self.getJobs.execute(state = 'JobFailed')
self.assertEqual(len(idList), self.nJobs)
testErrorHandler = ErrorHandlerPoller(config)
testErrorHandler.setup(None)
testErrorHandler.algorithm(None)
idList = self.getJobs.execute(state = 'JobFailed')
self.assertEqual(len(idList), 0)
idList = self.getJobs.execute(state = 'JobCooloff')
self.assertEqual(len(idList), self.nJobs)
return
def testD_Exhausted(self):
"""
_testExhausted_
Test that the system can exhaust jobs correctly
"""
workloadName = 'TestWorkload'
workload = self.createWorkload(workloadName = workloadName)
workloadPath = os.path.join(self.testDir, 'workloadTest', 'TestWorkload',
'WMSandbox', 'WMWorkload.pkl')
testJobGroup = self.createTestJobGroup(nJobs = self.nJobs, retry_count = 5,
workloadPath = workloadPath)
config = self.getConfig()
config.ErrorHandler.maxRetries = 1
changer = ChangeState(config)
changer.propagate(testJobGroup.jobs, 'created', 'new')
changer.propagate(testJobGroup.jobs, 'executing', 'created')
changer.propagate(testJobGroup.jobs, 'complete', 'executing')
changer.propagate(testJobGroup.jobs, 'jobfailed', 'complete')
testSubscription = Subscription(id = 1) # You should only have one
testSubscription.load()
testSubscription.loadData()
# Do we have files to start with?
self.assertEqual(len(testSubscription.filesOfStatus("Acquired")), 2)
testErrorHandler = ErrorHandlerPoller(config)
testErrorHandler.setup(None)
testErrorHandler.algorithm(None)
idList = self.getJobs.execute(state = 'JobFailed')
self.assertEqual(len(idList), 0)
idList = self.getJobs.execute(state = 'JobCooloff')
self.assertEqual(len(idList), 0)
idList = self.getJobs.execute(state = 'Exhausted')
self.assertEqual(len(idList), self.nJobs)
# Did we fail the files?
self.assertEqual(len(testSubscription.filesOfStatus("Acquired")), 0)
self.assertEqual(len(testSubscription.filesOfStatus("Failed")), 2)
def testE_FailJobs(self):
"""
_FailJobs_
Test our ability to fail jobs based on the information in the FWJR
"""
workloadName = 'TestWorkload'
workload = self.createWorkload(workloadName = workloadName)
workloadPath = os.path.join(self.testDir, 'workloadTest', 'TestWorkload',
'WMSandbox', 'WMWorkload.pkl')
fwjrPath = os.path.join(WMCore.WMBase.getTestBase(),
"WMComponent_t/JobAccountant_t",
"fwjrs/badBackfillJobReport.pkl")
testJobGroup = self.createTestJobGroup(nJobs = self.nJobs,
workloadPath = workloadPath,
fwjrPath = fwjrPath)
config = self.getConfig()
config.ErrorHandler.readFWJR = True
config.ErrorHandler.failureExitCodes = [8020]
changer = ChangeState(config)
changer.propagate(testJobGroup.jobs, 'created', 'new')
changer.propagate(testJobGroup.jobs, 'executing', 'created')
changer.propagate(testJobGroup.jobs, 'complete', 'executing')
changer.propagate(testJobGroup.jobs, 'jobfailed', 'complete')
testErrorHandler = ErrorHandlerPoller(config)
testErrorHandler.setup(None)
testErrorHandler.algorithm(None)
# This should exhaust all jobs due to exit code
idList = self.getJobs.execute(state = 'JobFailed')
self.assertEqual(len(idList), 0)
idList = self.getJobs.execute(state = 'JobCooloff')
self.assertEqual(len(idList), 0)
idList = self.getJobs.execute(state = 'Exhausted')
self.assertEqual(len(idList), self.nJobs)
config.ErrorHandler.failureExitCodes = []
config.ErrorHandler.maxFailTime = -10
testErrorHandler2 = ErrorHandlerPoller(config)
changer.propagate(testJobGroup.jobs, 'created', 'new')
changer.propagate(testJobGroup.jobs, 'executing', 'created')
changer.propagate(testJobGroup.jobs, 'complete', 'executing')
changer.propagate(testJobGroup.jobs, 'jobfailed', 'complete')
testErrorHandler2.algorithm(None)
# This should exhaust all jobs due to timeout
idList = self.getJobs.execute(state = 'JobFailed')
self.assertEqual(len(idList), 0)
idList = self.getJobs.execute(state = 'JobCooloff')
self.assertEqual(len(idList), 0)
idList = self.getJobs.execute(state = 'Exhausted')
self.assertEqual(len(idList), self.nJobs)
config.ErrorHandler.maxFailTime = 24 * 3600
config.ErrorHandler.passExitCodes = [8020]
testErrorHandler3 = ErrorHandlerPoller(config)
changer.propagate(testJobGroup.jobs, 'created', 'new')
changer.propagate(testJobGroup.jobs, 'executing', 'created')
changer.propagate(testJobGroup.jobs, 'complete', 'executing')
changer.propagate(testJobGroup.jobs, 'jobfailed', 'complete')
testErrorHandler3.algorithm(None)
idList = self.getJobs.execute(state = 'Created')
self.assertEqual(len(idList), self.nJobs)
return
def testZ_Profile(self):
"""
_testProfile_
Do a full profile of the poller
"""
return
import cProfile, pstats
nJobs = 1000
testJobGroup = self.createTestJobGroup(nJobs = nJobs)
config = self.getConfig()
changer = ChangeState(config)
changer.propagate(testJobGroup.jobs, 'createfailed', 'new')
idList = self.getJobs.execute(state = 'CreateFailed')
self.assertEqual(len(idList), nJobs)
testErrorHandler = ErrorHandlerPoller(config)
testErrorHandler.setup(None)
startTime = time.time()
#cProfile.runctx("testErrorHandler.algorithm()", globals(), locals(), filename = "profStats.stat")
testErrorHandler.algorithm()
stopTime = time.time()
idList = self.getJobs.execute(state = 'CreateFailed')
self.assertEqual(len(idList), 0)
idList = self.getJobs.execute(state = 'CreateCooloff')
self.assertEqual(len(idList), nJobs)
print("Took %f seconds to run polling algo" % (stopTime - startTime))
p = pstats.Stats('profStats.stat')
p.sort_stats('cumulative')
p.print_stats(0.2)
return
def algorithm(self, *args, **kwargs):
"""
_algorithm_
Split files into a number of lumis per job
Allow a flag to determine if we split files between jobs
"""
avgEventsPerJob = int(kwargs.get('events_per_job', 5000))
eventLimit = int(kwargs.get('max_events_per_lumi', 20000))
totalEvents = int(kwargs.get('total_events', 0))
splitOnFile = bool(kwargs.get('halt_job_on_file_boundaries', True))
ignoreACDC = bool(kwargs.get('ignore_acdc_except', False))
collectionName = kwargs.get('collectionName', None)
splitOnRun = kwargs.get('splitOnRun', True)
getParents = kwargs.get('include_parents', False)
runWhitelist = kwargs.get('runWhitelist', [])
runs = kwargs.get('runs', None)
lumis = kwargs.get('lumis', None)
timePerEvent, sizePerEvent, memoryRequirement = \
self.getPerformanceParameters(kwargs.get('performance', {}))
deterministicPileup = kwargs.get('deterministicPileup', False)
eventsPerLumiInDataset = 0
if deterministicPileup and self.package == 'WMCore.WMBS':
getJobNumber = self.daoFactory(classname = "Jobs.GetNumberOfJobsPerWorkflow")
jobNumber = getJobNumber.execute(workflow = self.subscription.getWorkflow().id)
self.nJobs = jobNumber
goodRunList = {}
if runs and lumis:
goodRunList = buildLumiMask(runs, lumis)
# If we have runLumi info, we need to load it from couch
if collectionName:
try:
from WMCore.ACDC.DataCollectionService import DataCollectionService
couchURL = kwargs.get('couchURL')
couchDB = kwargs.get('couchDB')
filesetName = kwargs.get('filesetName')
collectionName = kwargs.get('collectionName')
owner = kwargs.get('owner')
group = kwargs.get('group')
logging.info('Creating jobs for ACDC fileset %s' % filesetName)
dcs = DataCollectionService(couchURL, couchDB)
goodRunList = dcs.getLumiWhitelist(collectionName, filesetName, owner, group)
except Exception as ex:
msg = "Exception while trying to load goodRunList\n"
if ignoreACDC:
msg += "Ditching goodRunList\n"
msg += str(ex)
msg += str(traceback.format_exc())
logging.error(msg)
goodRunList = {}
else:
msg += "Refusing to create any jobs.\n"
msg += str(ex)
msg += str(traceback.format_exc())
logging.error(msg)
return
lDict = self.sortByLocation()
locationDict = {}
# First we need to load the data
if self.package == 'WMCore.WMBS':
loadRunLumi = self.daoFactory(classname = "Files.GetBulkRunLumi")
for key in lDict.keys():
newlist = []
# First we need to load the data
if self.package == 'WMCore.WMBS':
fileLumis = loadRunLumi.execute(files = lDict[key])
for f in lDict[key]:
lumiDict = fileLumis.get(f['id'], {})
for run in lumiDict.keys():
f.addRun(run = Run(run, *lumiDict[run]))
for f in lDict[key]:
if len(f['runs']) == 0:
continue
f['runs'] = sorted(f['runs'])
f['lumiCount'] = 0
for run in f['runs']:
run.lumis.sort()
f['lumiCount'] += len(run.lumis)
f['lowestRun'] = f['runs'][0]
#Do average event per lumi calculation
if f['lumiCount']:
f['avgEvtsPerLumi'] = round(float(f['events'])/f['lumiCount'])
if deterministicPileup:
# We assume that all lumis are equal in the dataset
eventsPerLumiInDataset = f['avgEvtsPerLumi']
else:
#No lumis in the file, ignore it
continue
newlist.append(f)
locationDict[key] = sorted(newlist, key=operator.itemgetter('lowestRun'))
totalJobs = 0
lastLumi = None
firstLumi = None
lastRun = None
lumisInJob = 0
totalAvgEventCount = 0
currentJobAvgEventCount = 0
stopTask = False
for location in locationDict:
# For each location, we need a new jobGroup
self.newGroup()
stopJob = True
for f in locationDict[location]:
if getParents:
parentLFNs = self.findParent(lfn = f['lfn'])
for lfn in parentLFNs:
parent = File(lfn = lfn)
f['parents'].add(parent)
lumisInJobInFile = 0
updateSplitOnJobStop = False
failNextJob = False
#If the number of events per lumi is higher than the limit
#and it's only one lumi then ditch that lumi
if f['avgEvtsPerLumi'] > eventLimit and f['lumiCount'] == 1:
failNextJob = True
stopJob = True
lumisPerJob = 1
elif splitOnFile:
# Then we have to split on every boundary
stopJob = True
#Check the average number of events per lumi in this file
#Adapt the lumis per job to match the target conditions
if f['avgEvtsPerLumi']:
#If there are events in the file
ratio = float(avgEventsPerJob) / f['avgEvtsPerLumi']
lumisPerJob = max(int(math.floor(ratio)), 1)
else:
#Zero event file, then the ratio goes to infinity. Computers don't like that
lumisPerJob = f['lumiCount']
else:
#Analyze how many events does this job already has
#Check how many we want as target, include as many lumi sections as possible
updateSplitOnJobStop = True
eventsRemaining = max(avgEventsPerJob - currentJobAvgEventCount, 0)
if f['avgEvtsPerLumi']:
lumisAllowed = int(math.floor(float(eventsRemaining) / f['avgEvtsPerLumi']))
else:
lumisAllowed = f['lumiCount']
lumisPerJob = max(lumisInJob + lumisAllowed, 1)
for run in f['runs']:
if not isGoodRun(goodRunList = goodRunList, run = run.run):
# Then skip this one
continue
if len(runWhitelist) > 0 and not run.run in runWhitelist:
# Skip due to run whitelist
continue
firstLumi = None
if splitOnRun and run.run != lastRun:
# Then we need to kill this job and get a new one
stopJob = True
# Now loop over the lumis
for lumi in run:
if not isGoodLumi(goodRunList, run = run.run, lumi = lumi):
# Kill the chain of good lumis
# Skip this lumi
if firstLumi != None and firstLumi != lumi:
self.currentJob['mask'].addRunAndLumis(run = run.run,
lumis = [firstLumi, lastLumi])
eventsAdded = ((lastLumi - firstLumi + 1) * f['avgEvtsPerLumi'])
runAddedTime = eventsAdded * timePerEvent
runAddedSize = eventsAdded * sizePerEvent
self.currentJob.addResourceEstimates(jobTime = runAddedTime, disk = runAddedSize)
firstLumi = None
lastLumi = None
continue
# You have to kill the lumi chain if they're not continuous
if lastLumi and not lumi == lastLumi + 1:
self.currentJob['mask'].addRunAndLumis(run = run.run,
lumis = [firstLumi, lastLumi])
eventsAdded = ((lastLumi - firstLumi + 1) * f['avgEvtsPerLumi'])
runAddedTime = eventsAdded * timePerEvent
runAddedSize = eventsAdded * sizePerEvent
self.currentJob.addResourceEstimates(jobTime = runAddedTime, disk = runAddedSize)
firstLumi = None
lastLumi = None
if firstLumi == None:
# Set the first lumi in the run
firstLumi = lumi
# If we're full, end the job
if lumisInJob == lumisPerJob:
stopJob = True
# Actually do the new job creation
if stopJob:
if firstLumi != None and lastLumi != None and lastRun != None:
self.currentJob['mask'].addRunAndLumis(run = lastRun,
lumis = [firstLumi, lastLumi])
eventsAdded = ((lastLumi - firstLumi + 1) * f['avgEvtsPerLumi'])
runAddedTime = eventsAdded * timePerEvent
runAddedSize = eventsAdded * sizePerEvent
self.currentJob.addResourceEstimates(jobTime = runAddedTime, disk = runAddedSize)
msg = None
if failNextJob:
msg = "File %s has too many events (%d) in %d lumi(s)" % (f['lfn'],
f['events'],
f['lumiCount'])
self.newJob(name = self.getJobName(), failedJob = failNextJob,
failedReason = msg)
if deterministicPileup:
self.currentJob.addBaggageParameter("skipPileupEvents", (self.nJobs - 1) * lumisPerJob * eventsPerLumiInDataset)
self.currentJob.addResourceEstimates(memory = memoryRequirement)
failNextJob = False
firstLumi = lumi
lumisInJob = 0
lumisInJobInFile = 0
currentJobAvgEventCount = 0
totalJobs += 1
# Add the file to new jobs
self.currentJob.addFile(f)
if updateSplitOnJobStop:
#Then we were carrying from a previous file
#Reset calculations for this file
updateSplitOnJobStop = False
if f['avgEvtsPerLumi']:
ratio = float(avgEventsPerJob) / f['avgEvtsPerLumi']
lumisPerJob = max(int(math.floor(ratio)), 1)
else:
lumisPerJob = f['lumiCount']
lumisInJob += 1
lumisInJobInFile += 1
lastLumi = lumi
stopJob = False
lastRun = run.run
totalAvgEventCount += f['avgEvtsPerLumi']
if self.currentJob and not f in self.currentJob['input_files']:
self.currentJob.addFile(f)
# We stop here if there are more total events than requested.
if totalEvents > 0 and totalAvgEventCount >= totalEvents:
stopTask = True
break
if firstLumi != None and lastLumi != None:
# Add this run to the mask
self.currentJob['mask'].addRunAndLumis(run = run.run,
lumis = [firstLumi, lastLumi])
eventsAdded = ((lastLumi - firstLumi + 1) * f['avgEvtsPerLumi'])
runAddedTime = eventsAdded * timePerEvent
runAddedSize = eventsAdded * sizePerEvent
self.currentJob.addResourceEstimates(jobTime = runAddedTime, disk = runAddedSize)
firstLumi = None
lastLumi = None
if stopTask:
break
if not splitOnFile:
currentJobAvgEventCount += f['avgEvtsPerLumi'] * lumisInJobInFile
if stopTask:
break
if stopTask:
break
return
class ErrorHandlerPoller(BaseWorkerThread):
"""
Polls for Error Conditions, handles them
"""
def __init__(self, config):
"""
Initialise class members
"""
BaseWorkerThread.__init__(self)
self.config = config
myThread = threading.currentThread()
self.daoFactory = DAOFactory(package = "WMCore.WMBS",
logger = myThread.logger,
dbinterface = myThread.dbi)
self.changeState = ChangeState(self.config)
self.maxRetries = self.config.ErrorHandler.maxRetries
self.maxProcessSize = getattr(self.config.ErrorHandler, 'maxProcessSize', 250)
self.getJobs = self.daoFactory(classname = "Jobs.GetAllJobs")
self.dataCollection = DataCollectionService(url = config.ACDC.couchurl,
database = config.ACDC.database)
# initialize the alert framework (if available - config.Alert present)
# self.sendAlert will be then be available
self.initAlerts(compName = "ErrorHandler")
return
def setup(self, parameters):
"""
Load DB objects required for queries
"""
# For now, does nothing
return
def terminate(self, params):
"""
_terminate_
Do one pass, then commit suicide
"""
logging.debug("terminating. doing one more pass before we die")
self.algorithm(params)
def processRetries(self, jobs, jobType):
"""
Actually do the retries
"""
logging.info("Processing retries for %i failed jobs of type %s." % (len(jobs), jobType))
exhaustJobs = []
cooloffJobs = []
# Retries < max retry count
for ajob in jobs:
# Retries < max retry count
if ajob['retry_count'] < self.maxRetries:
cooloffJobs.append(ajob)
# Check if Retries >= max retry count
elif ajob['retry_count'] >= self.maxRetries:
exhaustJobs.append(ajob)
msg = "Exhausting job %i" % ajob['id']
logging.error(msg)
self.sendAlert(6, msg = msg)
logging.debug("JobInfo: %s" % ajob)
else:
logging.debug("Job %i had %s retries remaining" \
% (ajob['id'], str(ajob['retry_count'])))
#Now to actually do something.
logging.debug("About to propagate jobs")
self.changeState.propagate(exhaustJobs, 'exhausted', \
'%sfailed' %(jobType))
self.changeState.propagate(cooloffJobs, '%scooloff' %(jobType), \
'%sfailed' %(jobType))
# Remove all the files in the exhausted jobs.
logging.debug("About to fail input files for exhausted jobs")
for job in exhaustJobs:
job.failInputFiles()
return exhaustJobs
def handleACDC(self, jobList):
"""
_handleACDC_
Do the ACDC creation and hope it works
"""
logging.debug("Entering ACDC with %i jobs" % len(jobList))
for job in jobList:
job.getMask()
self.dataCollection.failedJobs(jobList)
return
def splitJobList(self, jobList, jobType):
"""
_splitJobList_
Split up list of jobs into more manageable chunks if necessary
"""
if len(jobList) < 1:
# Nothing to do
return
myThread = threading.currentThread()
while len(jobList) > 0:
# Loop over the list and handle it one chunk at a time
tmpList = jobList[:self.maxProcessSize]
jobList = jobList[self.maxProcessSize:]
logging.debug("About to process %i errors" % len(tmpList))
myThread.transaction.begin()
exhaustList = self.processRetries(tmpList, jobType)
self.handleACDC(jobList = exhaustList)
myThread.transaction.commit()
return
def handleErrors(self):
"""
Queries DB for all watched filesets, if matching filesets become
available, create the subscriptions
"""
createList = []
submitList = []
jobList = []
# Run over created jobs
idList = self.getJobs.execute(state = 'CreateFailed')
logging.info("Found %s failed jobs failed during creation" \
% len(idList))
if len(idList) > 0:
createList = self.loadJobsFromList(idList = idList)
# Run over submitted jobs
idList = self.getJobs.execute(state = 'SubmitFailed')
logging.info("Found %s failed jobs failed during submit" \
% len(idList))
if len(idList) > 0:
submitList = self.loadJobsFromList(idList = idList)
# Run over executed jobs
idList = self.getJobs.execute(state = 'JobFailed')
logging.info("Found %s failed jobs failed during execution" \
% len(idList))
if len(idList) > 0:
jobList = self.loadJobsFromList(idList = idList)
self.splitJobList(jobList = createList, jobType = 'create')
self.splitJobList(jobList = submitList, jobType = 'submit')
self.splitJobList(jobList = jobList, jobType = 'job')
return
def loadJobsFromList(self, idList):
"""
_loadJobsFromList_
Load jobs in bulk
"""
loadAction = self.daoFactory(classname = "Jobs.LoadForErrorHandler")
binds = []
for jobID in idList:
binds.append({"jobid": jobID})
results = loadAction.execute(jobID = binds)
# You have to have a list
if type(results) == dict:
results = [results]
listOfJobs = []
for entry in results:
# One job per entry
tmpJob = Job(id = entry['id'])
tmpJob.update(entry)
listOfJobs.append(tmpJob)
return listOfJobs
def algorithm(self, parameters = None):
"""
Performs the handleErrors method, looking for each type of failure
And deal with it as desired.
"""
logging.debug("Running error handling algorithm")
myThread = threading.currentThread()
try:
self.handleErrors()
except WMException, ex:
try:
myThread.transaction.rollback()
except:
pass
raise
except Exception, ex:
msg = "Caught exception in ErrorHandler\n"
msg += str(ex)
msg += str(traceback.format_exc())
msg += "\n\n"
logging.error(msg)
self.sendAlert(6, msg = msg)
if getattr(myThread, 'transaction', None) != None \
and getattr(myThread.transaction, 'transaction', None) != None:
myThread.transaction.rollback()
raise ErrorHandlerException(msg)
def __init__(self, config):
"""
__init__
Create all DAO objects that are used by this class.
"""
WMConnectionBase.__init__(self, "WMCore.WMBS")
myThread = threading.currentThread()
self.dbsDaoFactory = DAOFactory(package = "WMComponent.DBS3Buffer",
logger = myThread.logger,
dbinterface = myThread.dbi)
self.getOutputMapAction = self.daofactory(classname = "Jobs.GetOutputMap")
self.bulkAddToFilesetAction = self.daofactory(classname = "Fileset.BulkAddByLFN")
self.bulkParentageAction = self.daofactory(classname = "Files.AddBulkParentage")
self.getJobTypeAction = self.daofactory(classname = "Jobs.GetType")
self.getParentInfoAction = self.daofactory(classname = "Files.GetParentInfo")
self.setParentageByJob = self.daofactory(classname = "Files.SetParentageByJob")
self.setParentageByMergeJob = self.daofactory(classname = "Files.SetParentageByMergeJob")
self.setFileRunLumi = self.daofactory(classname = "Files.AddRunLumi")
self.setFileLocation = self.daofactory(classname = "Files.SetLocationByLFN")
self.setFileAddChecksum = self.daofactory(classname = "Files.AddChecksumByLFN")
self.addFileAction = self.daofactory(classname = "Files.Add")
self.jobCompleteInput = self.daofactory(classname = "Jobs.CompleteInput")
self.setBulkOutcome = self.daofactory(classname = "Jobs.SetOutcomeBulk")
self.getWorkflowSpec = self.daofactory(classname = "Workflow.GetSpecAndNameFromTask")
self.getJobInfoByID = self.daofactory(classname = "Jobs.LoadFromID")
self.getFullJobInfo = self.daofactory(classname = "Jobs.LoadForErrorHandler")
self.getJobTaskNameAction = self.daofactory(classname = "Jobs.GetFWJRTaskName")
self.dbsStatusAction = self.dbsDaoFactory(classname = "DBSBufferFiles.SetStatus")
self.dbsParentStatusAction = self.dbsDaoFactory(classname = "DBSBufferFiles.GetParentStatus")
self.dbsChildrenAction = self.dbsDaoFactory(classname = "DBSBufferFiles.GetChildren")
self.dbsCreateFiles = self.dbsDaoFactory(classname = "DBSBufferFiles.Add")
self.dbsSetLocation = self.dbsDaoFactory(classname = "DBSBufferFiles.SetLocationByLFN")
self.dbsInsertLocation = self.dbsDaoFactory(classname = "DBSBufferFiles.AddLocation")
self.dbsSetChecksum = self.dbsDaoFactory(classname = "DBSBufferFiles.AddChecksumByLFN")
self.dbsSetRunLumi = self.dbsDaoFactory(classname = "DBSBufferFiles.AddRunLumi")
self.dbsGetWorkflow = self.dbsDaoFactory(classname = "ListWorkflow")
self.dbsLFNHeritage = self.dbsDaoFactory(classname = "DBSBufferFiles.BulkHeritageParent")
self.stateChanger = ChangeState(config)
# Decide whether or not to attach jobReport to returned value
self.returnJobReport = getattr(config.JobAccountant, 'returnReportFromWorker', False)
# Store location for the specs for DBS
self.specDir = getattr(config.JobAccountant, 'specDir', None)
# ACDC service
self.dataCollection = DataCollectionService(url = config.ACDC.couchurl,
database = config.ACDC.database)
jobDBurl = sanitizeURL(config.JobStateMachine.couchurl)['url']
jobDBName = config.JobStateMachine.couchDBName
jobCouchdb = CouchServer(jobDBurl)
self.fwjrCouchDB = jobCouchdb.connectDatabase("%s/fwjrs" % jobDBName)
self.localWMStats = WMStatsWriter(config.TaskArchiver.localWMStatsURL, appName="WMStatsAgent")
# Hold data for later commital
self.dbsFilesToCreate = []
self.wmbsFilesToBuild = []
self.wmbsMergeFilesToBuild = []
self.fileLocation = None
self.mergedOutputFiles = []
self.listOfJobsToSave = []
self.listOfJobsToFail = []
self.filesetAssoc = []
self.parentageBinds = []
self.parentageBindsForMerge = []
self.jobsWithSkippedFiles = {}
self.count = 0
self.datasetAlgoID = collections.deque(maxlen = 1000)
self.datasetAlgoPaths = collections.deque(maxlen = 1000)
self.dbsLocations = set()
self.workflowIDs = collections.deque(maxlen = 1000)
self.workflowPaths = collections.deque(maxlen = 1000)
self.phedex = PhEDEx()
self.locLists = self.phedex.getNodeMap()
return
def testChunking(self):
"""
_testChunking_
Insert a workload and files that have several distinct sets of
locations. Verify that the chunks are created correctly and that they
only groups files that have the same set of locations. Also verify that
the chunks are pulled out of ACDC correctly.
"""
dcs = DataCollectionService(url=self.testInit.couchUrl,
database="wmcore-acdc-datacollectionsvc")
testFileA = File(lfn=makeUUID(), size=1024, events=1024)
testFileA.setLocation(["cmssrm.fnal.gov", "castor.cern.ch"])
testFileA.addRun(Run(1, 1, 2))
testFileB = File(lfn=makeUUID(), size=1024, events=1024)
testFileB.setLocation(["cmssrm.fnal.gov", "castor.cern.ch"])
testFileB.addRun(Run(1, 3, 4))
testFileC = File(lfn=makeUUID(), size=1024, events=1024)
testFileC.setLocation(["cmssrm.fnal.gov", "castor.cern.ch"])
testFileC.addRun(Run(1, 5, 6))
testJobA = self.getMinimalJob()
testJobA.addFile(testFileA)
testJobA.addFile(testFileB)
testJobA.addFile(testFileC)
testFileD = File(lfn=makeUUID(), size=1024, events=1024)
testFileD.setLocation(["cmssrm.fnal.gov"])
testFileD.addRun(Run(2, 1, 2))
testFileE = File(lfn=makeUUID(), size=1024, events=1024)
testFileE.setLocation(["cmssrm.fnal.gov"])
testFileE.addRun(Run(2, 3, 4))
testJobB = self.getMinimalJob()
testJobB.addFile(testFileD)
testJobB.addFile(testFileE)
testFileF = File(lfn=makeUUID(),
size=1024,
events=1024,
parents={"/some/parent/F"})
testFileF.setLocation(
["cmssrm.fnal.gov", "castor.cern.ch", "srm.ral.uk"])
testFileF.addRun(Run(3, 1, 2))
testFileG = File(lfn=makeUUID(),
size=1024,
events=1024,
parents={"/some/parent/G"})
testFileG.setLocation(
["cmssrm.fnal.gov", "castor.cern.ch", "srm.ral.uk"])
testFileG.addRun(Run(3, 3, 4))
testFileH = File(lfn=makeUUID(),
size=1024,
events=1024,
parents={"/some/parent/H"})
testFileH.setLocation(
["cmssrm.fnal.gov", "castor.cern.ch", "srm.ral.uk"])
testFileH.addRun(Run(3, 5, 6))
testJobC = self.getMinimalJob()
testJobC.addFile(testFileF)
testJobC.addFile(testFileG)
testJobC.addFile(testFileH)
testFileI = File(lfn=makeUUID(), size=1024, events=1024, merged=True)
testFileI.setLocation(["cmssrm.fnal.gov", "castor.cern.ch"])
testFileI.addRun(Run(4, 1, 2))
testFileJ = File(lfn=makeUUID(), size=1024, events=1024, merged=True)
testFileJ.setLocation(["cmssrm.fnal.gov", "castor.cern.ch"])
testFileJ.addRun(Run(4, 3, 4))
testFileK = File(lfn=makeUUID(), size=1024, events=1024, merged=True)
testFileK.setLocation(["cmssrm.fnal.gov", "castor.cern.ch"])
testFileK.addRun(Run(4, 5, 6))
testJobD = self.getMinimalJob()
testJobD.addFile(testFileI)
testJobD.addFile(testFileJ)
testJobD.addFile(testFileK)
dcs.failedJobs([testJobA, testJobB, testJobC, testJobD])
chunks = dcs.chunkFileset("ACDCTest", "/ACDCTest/reco", chunkSize=5)
self.assertEqual(
len(chunks), 4,
"Error: There should be four chunks: %s" % len(chunks))
goldenMetaData = {
1: {
"lumis": 2,
"locations": ["castor.cern.ch", "cmssrm.fnal.gov"],
"events": 1024
},
2: {
"lumis": 4,
"locations": ["cmssrm.fnal.gov"],
"events": 2048
},
3: {
"lumis": 6,
"locations":
["castor.cern.ch", "cmssrm.fnal.gov", "srm.ral.uk"],
"events": 3072
},
5: {
"lumis": 10,
"locations": ["castor.cern.ch", "cmssrm.fnal.gov"],
"events": 5120
}
}
testFiles = [
testFileA, testFileB, testFileC, testFileI, testFileJ, testFileK
]
lastFile = testFileA
for testFile in testFiles:
if lastFile["lfn"] < testFile["lfn"]:
lastFile = testFile
testFiles.remove(lastFile)
goldenFiles = {
1: [lastFile],
2: [testFileD, testFileE],
3: [testFileF, testFileG, testFileH],
5: testFiles
}
for chunk in chunks:
chunkMetaData = dcs.getChunkInfo("ACDCTest", "/ACDCTest/reco",
chunk["offset"], chunk["files"])
self.assertEqual(chunkMetaData["files"], chunk["files"])
self.assertEqual(chunkMetaData["lumis"], chunk["lumis"])
self.assertEqual(chunkMetaData["events"], chunk["events"])
self.assertEqual(chunkMetaData["locations"], chunk["locations"])
self.assertTrue(chunk["files"] in goldenMetaData.keys(),
"Error: Extra chunk found.")
self.assertEqual(chunk["lumis"],
goldenMetaData[chunk["files"]]["lumis"],
"Error: Lumis in chunk is wrong.")
self.assertEqual(chunk["locations"],
goldenMetaData[chunk["files"]]["locations"],
"Error: Locations in chunk is wrong.")
self.assertEqual(chunk["events"],
goldenMetaData[chunk["files"]]["events"],
"Error: Events in chunk is wrong.")
del goldenMetaData[chunk["files"]]
chunkFiles = dcs.getChunkFiles("ACDCTest", "/ACDCTest/reco",
chunk["offset"], chunk["files"])
self.assertTrue(chunk["files"] in goldenFiles.keys(),
"Error: Extra chunk found.")
goldenChunkFiles = goldenFiles[chunk["files"]]
self.assertEqual(len(chunkFiles), len(goldenChunkFiles))
for chunkFile in chunkFiles:
foundFile = None
for goldenChunkFile in goldenChunkFiles:
if chunkFile["lfn"] == goldenChunkFile["lfn"]:
foundFile = goldenChunkFile
break
self.assertIsNotNone(
foundFile, "Error: Missing chunk file: %s, %s" %
(chunkFiles, goldenChunkFiles))
self.assertEqual(set(foundFile["parents"]),
chunkFile["parents"],
"Error: File parents should match.")
self.assertEqual(foundFile["merged"], chunkFile["merged"],
"Error: File merged status should match.")
self.assertEqual(foundFile["locations"],
chunkFile["locations"],
"Error: File locations should match.")
self.assertEqual(foundFile["events"], chunkFile["events"])
self.assertEqual(foundFile["size"], chunkFile["size"])
self.assertEqual(len(foundFile["runs"]),
len(chunkFile["runs"]),
"Error: Wrong number of runs.")
for run in foundFile["runs"]:
runMatch = False
for chunkRun in chunkFile["runs"]:
if chunkRun.run == run.run and chunkRun.lumis == run.lumis:
runMatch = True
break
self.assertTrue(runMatch,
"Error: Run information is wrong.")
del goldenFiles[chunk["files"]]
singleChunk = dcs.singleChunkFileset("ACDCTest", "/ACDCTest/reco")
self.assertEqual(
singleChunk, {
"offset": 0,
"files": 11,
"events": 11264,
"lumis": 22,
"locations":
{"castor.cern.ch", "cmssrm.fnal.gov", "srm.ral.uk"}
}, "Error: Single chunk metadata is wrong")
return
def testGetLumiWhitelist(self):
"""
_testGetLumiWhitelist_
Verify that the ACDC whitelist generation code works correctly. We'll
add jobs with the following lumi info:
# Run 1, lumis [1, 2, 3], [4, 6], [7], [9], [11, 12]
# Run 2, lumis [5, 6, 7], [10, 11, 12], [15]
# Run 3, lumis [20]
And should get out a whitelist that looks like this:
{"1": [[1, 4], [6, 7], [9, 9], [11, 12]],
"2": [[5, 7], [10, 12], [15, 15]],
"3": [[20, 20]]}
"""
dcs = DataCollectionService(url = self.testInit.couchUrl,
database = "wmcore-acdc-datacollectionsvc")
def getJob():
job = Job()
job["task"] = "/ACDCTest/reco"
job["workflow"] = "ACDCTest"
job["location"] = "cmssrm.fnal.gov"
job["owner"] = "cmsdataops"
job["group"] = "cmsdataops"
return job
testFileA = File(lfn = makeUUID(), size = 1024, events = 1024)
testFileA.addRun(Run(1, 1, 2))
testFileB = File(lfn = makeUUID(), size = 1024, events = 1024)
testFileB.addRun(Run(1, 3))
testJobA = getJob()
testJobA.addFile(testFileA)
testJobA.addFile(testFileB)
testFileC = File(lfn = makeUUID(), size = 1024, events = 1024)
testFileC.addRun(Run(1, 4, 6))
testJobB = getJob()
testJobB.addFile(testFileC)
testFileD = File(lfn = makeUUID(), size = 1024, events = 1024)
testFileD.addRun(Run(1, 7))
testJobC = getJob()
testJobC.addFile(testFileD)
testFileE = File(lfn = makeUUID(), size = 1024, events = 1024)
testFileE.addRun(Run(1, 11, 12))
testJobD = getJob()
testJobD.addFile(testFileE)
testFileF = File(lfn = makeUUID(), size = 1024, events = 1024)
testFileF.addRun(Run(2, 5, 6, 7))
testJobE = getJob()
testJobE.addFile(testFileF)
testFileG = File(lfn = makeUUID(), size = 1024, events = 1024)
testFileG.addRun(Run(2, 10, 11, 12))
testJobF = getJob()
testJobF.addFile(testFileG)
testFileH = File(lfn = makeUUID(), size = 1024, events = 1024)
testFileH.addRun(Run(2, 15))
testJobG = getJob()
testJobG.addFile(testFileH)
testFileI = File(lfn = makeUUID(), size = 1024, events = 1024)
testFileI.addRun(Run(3, 20))
testJobH = getJob()
testJobH.addFile(testFileI)
testFileJ = File(lfn = makeUUID(), size = 1024, events = 1024)
testFileJ.addRun(Run(1, 9))
testJobI = getJob()
testJobI.addFile(testFileJ)
dcs.failedJobs([testJobA, testJobB, testJobC, testJobD, testJobE,
testJobF, testJobG, testJobH, testJobI])
whiteList = dcs.getLumiWhitelist("ACDCTest", "/ACDCTest/reco")
self.assertEqual(len(whiteList.keys()), 3,
"Error: There should be 3 runs.")
self.assertEqual(whiteList["1"], [[1, 4], [6, 7], [9, 9], [11, 12]],
"Error: Whitelist for run 1 is wrong.")
self.assertEqual(whiteList["2"], [[5, 7], [10, 12], [15, 15]],
"Error: Whitelist for run 2 is wrong.")
self.assertEqual(whiteList["3"], [[20, 20]],
"Error: Whitelist for run 3 is wrong.")
return
def testChunking(self):
"""
_testChunking_
Insert a workload and files that have several distinct sets of
locations. Verify that the chunks are created correctly and that they
only groups files that have the same set of locations. Also verify that
the chunks are pulled out of ACDC correctly.
"""
dcs = DataCollectionService(url = self.testInit.couchUrl,
database = "wmcore-acdc-datacollectionsvc")
def getJob():
job = Job()
job["task"] = "/ACDCTest/reco"
job["workflow"] = "ACDCTest"
job["location"] = "cmssrm.fnal.gov"
job["owner"] = "cmsdataops"
job["group"] = "cmsdataops"
return job
testFileA = File(lfn = makeUUID(), size = 1024, events = 1024)
testFileA.setLocation(["cmssrm.fnal.gov", "castor.cern.ch"])
testFileA.addRun(Run(1, 1, 2))
testFileB = File(lfn = makeUUID(), size = 1024, events = 1024)
testFileB.setLocation(["cmssrm.fnal.gov", "castor.cern.ch"])
testFileB.addRun(Run(1, 3, 4))
testFileC = File(lfn = makeUUID(), size = 1024, events = 1024)
testFileC.setLocation(["cmssrm.fnal.gov", "castor.cern.ch"])
testFileC.addRun(Run(1, 5, 6))
testJobA = getJob()
testJobA.addFile(testFileA)
testJobA.addFile(testFileB)
testJobA.addFile(testFileC)
testFileD = File(lfn = makeUUID(), size = 1024, events = 1024)
testFileD.setLocation(["cmssrm.fnal.gov"])
testFileD.addRun(Run(2, 1, 2))
testFileE = File(lfn = makeUUID(), size = 1024, events = 1024)
testFileE.setLocation(["cmssrm.fnal.gov"])
testFileE.addRun(Run(2, 3, 4))
testJobB = getJob()
testJobB.addFile(testFileD)
testJobB.addFile(testFileE)
testFileF = File(lfn = makeUUID(), size = 1024, events = 1024,
parents = set(["/some/parent/F"]))
testFileF.setLocation(["cmssrm.fnal.gov", "castor.cern.ch", "srm.ral.uk"])
testFileF.addRun(Run(3, 1, 2))
testFileG = File(lfn = makeUUID(), size = 1024, events = 1024,
parents = set(["/some/parent/G"]))
testFileG.setLocation(["cmssrm.fnal.gov", "castor.cern.ch", "srm.ral.uk"] )
testFileG.addRun(Run(3, 3, 4))
testFileH = File(lfn = makeUUID(), size = 1024, events = 1024,
parents = set(["/some/parent/H"]))
testFileH.setLocation(["cmssrm.fnal.gov", "castor.cern.ch", "srm.ral.uk"])
testFileH.addRun(Run(3, 5, 6))
testJobC = getJob()
testJobC.addFile(testFileF)
testJobC.addFile(testFileG)
testJobC.addFile(testFileH)
testFileI = File(lfn = makeUUID(), size = 1024, events = 1024, merged = True)
testFileI.setLocation(["cmssrm.fnal.gov", "castor.cern.ch"])
testFileI.addRun(Run(4, 1, 2))
testFileJ = File(lfn = makeUUID(), size = 1024, events = 1024, merged = True)
testFileJ.setLocation(["cmssrm.fnal.gov", "castor.cern.ch"] )
testFileJ.addRun(Run(4, 3, 4))
testFileK = File(lfn = makeUUID(), size = 1024, events = 1024, merged = True)
testFileK.setLocation(["cmssrm.fnal.gov", "castor.cern.ch"])
testFileK.addRun(Run(4, 5, 6))
testJobD = getJob()
testJobD.addFile(testFileI)
testJobD.addFile(testFileJ)
testJobD.addFile(testFileK)
dcs.failedJobs([testJobA, testJobB, testJobC, testJobD])
chunks = dcs.chunkFileset("ACDCTest", "/ACDCTest/reco",
chunkSize = 5)
self.assertEqual(len(chunks), 4, "Error: There should be four chunks: %s" % len(chunks))
goldenMetaData = {1: {"lumis": 2, "locations": ["castor.cern.ch", "cmssrm.fnal.gov"], "events": 1024},
2: {"lumis": 4, "locations": ["cmssrm.fnal.gov"], "events": 2048},
3: {"lumis": 6, "locations": ["castor.cern.ch", "cmssrm.fnal.gov", "srm.ral.uk"], "events": 3072},
5: {"lumis": 10, "locations": ["castor.cern.ch", "cmssrm.fnal.gov"], "events": 5120}}
testFiles =[testFileA, testFileB, testFileC, testFileI, testFileJ, testFileK]
lastFile = testFileA
for testFile in testFiles:
if lastFile["lfn"] < testFile["lfn"]:
lastFile = testFile
testFiles.remove(lastFile)
goldenFiles = {1: [lastFile],
2: [testFileD, testFileE],
3: [testFileF, testFileG, testFileH],
5: testFiles}
for chunk in chunks:
chunkMetaData = dcs.getChunkInfo("ACDCTest", "/ACDCTest/reco",
chunk["offset"], chunk["files"])
self.assertEqual(chunkMetaData["files"], chunk["files"],
"Error: Metadata doesn't match.")
self.assertEqual(chunkMetaData["lumis"], chunk["lumis"],
"Error: Metadata doesn't match.")
self.assertEqual(chunkMetaData["events"], chunk["events"],
"Error: Metadata doesn't match.")
self.assertEqual(chunkMetaData["locations"], chunk["locations"],
"Error: Metadata doesn't match.")
self.assertTrue(chunk["files"] in goldenMetaData.keys(),
"Error: Extra chunk found.")
self.assertEqual(chunk["lumis"], goldenMetaData[chunk["files"]]["lumis"],
"Error: Lumis in chunk is wrong.")
self.assertEqual(chunk["locations"], goldenMetaData[chunk["files"]]["locations"],
"Error: Locations in chunk is wrong.")
self.assertEqual(chunk["events"], goldenMetaData[chunk["files"]]["events"],
"Error: Events in chunk is wrong.")
del goldenMetaData[chunk["files"]]
chunkFiles = dcs.getChunkFiles("ACDCTest", "/ACDCTest/reco",
chunk["offset"], chunk["files"])
self.assertTrue(chunk["files"] in goldenFiles.keys(),
"Error: Extra chunk found.")
goldenChunkFiles = goldenFiles[chunk["files"]]
self.assertEqual(len(chunkFiles), len(goldenChunkFiles))
for chunkFile in chunkFiles:
foundFile = None
for goldenChunkFile in goldenChunkFiles:
if chunkFile["lfn"] == goldenChunkFile["lfn"]:
foundFile = goldenChunkFile
break
self.assertTrue(foundFile != None,
"Error: Missing chunk file: %s, %s" % (chunkFiles, goldenChunkFiles))
self.assertEqual(foundFile["parents"], chunkFile["parents"],
"Error: File parents should match.")
self.assertEqual(foundFile["merged"], chunkFile["merged"],
"Error: File merged status should match.")
self.assertEqual(foundFile["locations"], chunkFile["locations"],
"Error: File locations should match.")
self.assertEqual(foundFile["events"], chunkFile["events"],
"Error: File locations should match: %s" % chunk["files"])
self.assertEqual(foundFile["size"], chunkFile["size"],
"Error: File locations should match.")
self.assertEqual(len(foundFile["runs"]), len(chunkFile["runs"]),
"Error: Wrong number of runs.")
for run in foundFile["runs"]:
runMatch = False
for chunkRun in chunkFile["runs"]:
if chunkRun.run == run.run and chunkRun.lumis == run.lumis:
runMatch = True
break
self.assertTrue(runMatch, "Error: Run information is wrong.")
del goldenFiles[chunk["files"]]
singleChunk = dcs.singleChunkFileset("ACDCTest", "/ACDCTest/reco")
self.assertEqual(singleChunk, {"offset" : 0,
"files" : 11,
"events" : 11264,
"lumis" : 22,
"locations" : set(["castor.cern.ch", "cmssrm.fnal.gov", "srm.ral.uk"])},
"Error: Single chunk metadata is wrong")
return
def testGetLumiWhitelist(self):
"""
_testGetLumiWhitelist_
Verify that the ACDC whitelist generation code works correctly. We'll
add jobs with the following lumi info:
# Run 1, lumis [1, 2, 3], [4, 6], [7], [9], [11, 12]
# Run 2, lumis [5, 6, 7], [10, 11, 12], [15]
# Run 3, lumis [20]
And should get out a whitelist that looks like this:
{"1": [[1, 4], [6, 7], [9, 9], [11, 12]],
"2": [[5, 7], [10, 12], [15, 15]],
"3": [[20, 20]]}
"""
dcs = DataCollectionService(url=self.testInit.couchUrl,
database="wmcore-acdc-datacollectionsvc")
testFileA = File(lfn=makeUUID(), size=1024, events=1024)
testFileA.addRun(Run(1, 1, 2))
testFileB = File(lfn=makeUUID(), size=1024, events=1024)
testFileB.addRun(Run(1, 3))
testJobA = self.getMinimalJob()
testJobA.addFile(testFileA)
testJobA.addFile(testFileB)
testFileC = File(lfn=makeUUID(), size=1024, events=1024)
testFileC.addRun(Run(1, 4, 6))
testJobB = self.getMinimalJob()
testJobB.addFile(testFileC)
testFileD = File(lfn=makeUUID(), size=1024, events=1024)
testFileD.addRun(Run(1, 7))
testJobC = self.getMinimalJob()
testJobC.addFile(testFileD)
testFileE = File(lfn=makeUUID(), size=1024, events=1024)
testFileE.addRun(Run(1, 11, 12))
testJobD = self.getMinimalJob()
testJobD.addFile(testFileE)
testFileF = File(lfn=makeUUID(), size=1024, events=1024)
testFileF.addRun(Run(2, 5, 6, 7))
testJobE = self.getMinimalJob()
testJobE.addFile(testFileF)
testFileG = File(lfn=makeUUID(), size=1024, events=1024)
testFileG.addRun(Run(2, 10, 11, 12))
testJobF = self.getMinimalJob()
testJobF.addFile(testFileG)
testFileH = File(lfn=makeUUID(), size=1024, events=1024)
testFileH.addRun(Run(2, 15))
testJobG = self.getMinimalJob()
testJobG.addFile(testFileH)
testFileI = File(lfn=makeUUID(), size=1024, events=1024)
testFileI.addRun(Run(3, 20))
testJobH = self.getMinimalJob()
testJobH.addFile(testFileI)
testFileJ = File(lfn=makeUUID(), size=1024, events=1024)
testFileJ.addRun(Run(1, 9))
testJobI = self.getMinimalJob()
testJobI.addFile(testFileJ)
dcs.failedJobs([
testJobA, testJobB, testJobC, testJobD, testJobE, testJobF,
testJobG, testJobH, testJobI
])
whiteList = dcs.getLumiWhitelist("ACDCTest", "/ACDCTest/reco")
self.assertEqual(len(whiteList.keys()), 3,
"Error: There should be 3 runs.")
self.assertEqual(whiteList["1"], [[1, 4], [6, 7], [9, 9], [11, 12]],
"Error: Whitelist for run 1 is wrong.")
self.assertEqual(whiteList["2"], [[5, 7], [10, 12], [15, 15]],
"Error: Whitelist for run 2 is wrong.")
self.assertEqual(whiteList["3"], [[20, 20]],
"Error: Whitelist for run 3 is wrong.")
correctLumiList = LumiList(
compactList={
"1": [[1, 4], [6, 7], [9, 9], [11, 12]],
"2": [[5, 7], [10, 12], [15, 15]],
"3": [[20, 20]]
})
testLumiList = dcs.getLumilistWhitelist("ACDCTest", "/ACDCTest/reco")
self.assertEqual(correctLumiList.getCMSSWString(),
testLumiList.getCMSSWString())
return
def __init__(self, config):
"""
__init__
Create all DAO objects that are used by this class.
"""
WMConnectionBase.__init__(self, "WMCore.WMBS")
myThread = threading.currentThread()
self.dbsDaoFactory = DAOFactory(package="WMComponent.DBS3Buffer",
logger=myThread.logger,
dbinterface=myThread.dbi)
self.getOutputMapAction = self.daofactory(
classname="Jobs.GetOutputMap")
self.bulkAddToFilesetAction = self.daofactory(
classname="Fileset.BulkAddByLFN")
self.bulkParentageAction = self.daofactory(
classname="Files.AddBulkParentage")
self.getJobTypeAction = self.daofactory(classname="Jobs.GetType")
self.getParentInfoAction = self.daofactory(
classname="Files.GetParentInfo")
self.setParentageByJob = self.daofactory(
classname="Files.SetParentageByJob")
self.setParentageByMergeJob = self.daofactory(
classname="Files.SetParentageByMergeJob")
self.setFileRunLumi = self.daofactory(classname="Files.AddRunLumi")
self.setFileLocation = self.daofactory(
classname="Files.SetLocationByLFN")
self.setFileAddChecksum = self.daofactory(
classname="Files.AddChecksumByLFN")
self.addFileAction = self.daofactory(classname="Files.Add")
self.jobCompleteInput = self.daofactory(classname="Jobs.CompleteInput")
self.setBulkOutcome = self.daofactory(classname="Jobs.SetOutcomeBulk")
self.getWorkflowSpec = self.daofactory(
classname="Workflow.GetSpecAndNameFromTask")
self.getJobInfoByID = self.daofactory(classname="Jobs.LoadFromID")
self.getFullJobInfo = self.daofactory(
classname="Jobs.LoadForErrorHandler")
self.getJobTaskNameAction = self.daofactory(
classname="Jobs.GetFWJRTaskName")
self.pnn_to_psn = self.daofactory(
classname="Locations.GetPNNtoPSNMapping").execute()
self.dbsStatusAction = self.dbsDaoFactory(
classname="DBSBufferFiles.SetStatus")
self.dbsParentStatusAction = self.dbsDaoFactory(
classname="DBSBufferFiles.GetParentStatus")
self.dbsChildrenAction = self.dbsDaoFactory(
classname="DBSBufferFiles.GetChildren")
self.dbsCreateFiles = self.dbsDaoFactory(
classname="DBSBufferFiles.Add")
self.dbsSetLocation = self.dbsDaoFactory(
classname="DBSBufferFiles.SetLocationByLFN")
self.dbsInsertLocation = self.dbsDaoFactory(
classname="DBSBufferFiles.AddLocation")
self.dbsSetChecksum = self.dbsDaoFactory(
classname="DBSBufferFiles.AddChecksumByLFN")
self.dbsSetRunLumi = self.dbsDaoFactory(
classname="DBSBufferFiles.AddRunLumi")
self.dbsGetWorkflow = self.dbsDaoFactory(classname="ListWorkflow")
self.dbsLFNHeritage = self.dbsDaoFactory(
classname="DBSBufferFiles.BulkHeritageParent")
self.stateChanger = ChangeState(config)
# Decide whether or not to attach jobReport to returned value
self.returnJobReport = getattr(config.JobAccountant,
'returnReportFromWorker', False)
# Store location for the specs for DBS
self.specDir = getattr(config.JobAccountant, 'specDir', None)
# maximum RAW EDM size for Repack output before data is put into Error dataset and skips PromptReco
self.maxAllowedRepackOutputSize = getattr(
config.JobAccountant, 'maxAllowedRepackOutputSize',
12 * 1024 * 1024 * 1024)
# ACDC service
self.dataCollection = DataCollectionService(
url=config.ACDC.couchurl, database=config.ACDC.database)
jobDBurl = sanitizeURL(config.JobStateMachine.couchurl)['url']
jobDBName = config.JobStateMachine.couchDBName
jobCouchdb = CouchServer(jobDBurl)
self.fwjrCouchDB = jobCouchdb.connectDatabase("%s/fwjrs" % jobDBName)
self.localWMStats = WMStatsWriter(config.TaskArchiver.localWMStatsURL,
appName="WMStatsAgent")
# Hold data for later commital
self.dbsFilesToCreate = []
self.wmbsFilesToBuild = []
self.wmbsMergeFilesToBuild = []
self.fileLocation = None
self.mergedOutputFiles = []
self.listOfJobsToSave = []
self.listOfJobsToFail = []
self.filesetAssoc = []
self.parentageBinds = []
self.parentageBindsForMerge = []
self.jobsWithSkippedFiles = {}
self.count = 0
self.datasetAlgoID = collections.deque(maxlen=1000)
self.datasetAlgoPaths = collections.deque(maxlen=1000)
self.dbsLocations = set()
self.workflowIDs = collections.deque(maxlen=1000)
self.workflowPaths = collections.deque(maxlen=1000)
self.phedex = PhEDEx()
self.locLists = self.phedex.getNodeMap()
return
def testC_ACDCTest(self):
"""
_ACDCTest_
Test whether we can get a goodRunList out of ACDC
and process it correctly.
"""
workload = self.createTestWorkload()
dcs = DataCollectionService(url=self.testInit.couchUrl, database=self.testInit.couchDbName)
testFileA = File(lfn=makeUUID(), size=1024, events=1024, locations="T1_US_FNAL_Disk")
testFileA.addRun(Run(1, 1, 2))
testFileA.create()
testFileB = File(lfn=makeUUID(), size=1024, events=1024, locations="T1_US_FNAL_Disk")
testFileB.addRun(Run(1, 3))
testFileB.create()
testJobA = getJob(workload)
testJobA.addFile(testFileA)
testJobA.addFile(testFileB)
testFileC = File(lfn=makeUUID(), size=1024, events=1024, locations="T1_US_FNAL_Disk")
testFileC.addRun(Run(1, 4, 6))
testFileC.create()
testJobB = getJob(workload)
testJobB.addFile(testFileC)
testFileD = File(lfn=makeUUID(), size=1024, events=1024, locations="T1_US_FNAL_Disk")
testFileD.addRun(Run(1, 7))
testFileD.create()
testJobC = getJob(workload)
testJobC.addFile(testFileD)
testFileE = File(lfn=makeUUID(), size=1024, events=1024, locations="T1_US_FNAL_Disk")
testFileE.addRun(Run(1, 11, 12))
testFileE.create()
testJobD = getJob(workload)
testJobD.addFile(testFileE)
testFileF = File(lfn=makeUUID(), size=1024, events=1024, locations="T1_US_FNAL_Disk")
testFileF.addRun(Run(2, 5, 6, 7))
testFileF.create()
testJobE = getJob(workload)
testJobE.addFile(testFileF)
testFileG = File(lfn=makeUUID(), size=1024, events=1024, locations="T1_US_FNAL_Disk")
testFileG.addRun(Run(2, 10, 11, 12))
testFileG.create()
testJobF = getJob(workload)
testJobF.addFile(testFileG)
testFileH = File(lfn=makeUUID(), size=1024, events=1024, locations="T1_US_FNAL_Disk")
testFileH.addRun(Run(2, 15))
testFileH.create()
testJobG = getJob(workload)
testJobG.addFile(testFileH)
testFileI = File(lfn=makeUUID(), size=1024, events=1024, locations="T1_US_FNAL_Disk")
testFileI.addRun(Run(3, 20))
testFileI.create()
testJobH = getJob(workload)
testJobH.addFile(testFileI)
testFileJ = File(lfn=makeUUID(), size=1024, events=1024, locations="T1_US_FNAL_Disk")
testFileJ.addRun(Run(1, 9))
testFileJ.create()
testJobI = getJob(workload)
testJobI.addFile(testFileJ)
# dcs.failedJobs([testJobA, testJobB, testJobC, testJobD, testJobE,
# testJobF, testJobG, testJobH, testJobI])
dcs.failedJobs([testJobA, testJobD, testJobH])
baseName = makeUUID()
testFileset = Fileset(name=baseName)
testFileset.create()
testFileset.addFile(testFileA)
testFileset.addFile(testFileB)
testFileset.addFile(testFileC)
testFileset.addFile(testFileD)
testFileset.addFile(testFileE)
testFileset.addFile(testFileF)
testFileset.addFile(testFileG)
testFileset.addFile(testFileH)
testFileset.addFile(testFileI)
testFileset.addFile(testFileJ)
testFileset.commit()
testSubscription = Subscription(fileset=testFileset,
workflow=self.testWorkflow,
split_algo="LumiBased",
type="Processing")
testSubscription.create()
splitter = SplitterFactory()
jobFactory = splitter(package="WMCore.WMBS",
subscription=testSubscription)
jobGroups = jobFactory(lumis_per_job=100,
halt_job_on_file_boundaries=False,
splitOnRun=True,
collectionName=workload.name(),
filesetName=workload.getTask("reco").getPathName(),
owner="evansde77",
group="DMWM",
couchURL=self.testInit.couchUrl,
couchDB=self.testInit.couchDbName,
performance=self.performanceParams)
self.assertEqual(jobGroups[0].jobs[0]['mask'].getRunAndLumis(), {1: [[1, 2], [3, 3], [11, 12]]})
self.assertEqual(jobGroups[0].jobs[1]['mask'].getRunAndLumis(), {3: [[20, 20]]})
return
def testC_ACDCTest(self):
"""
_ACDCTest_
Test whether we can get a goodRunList out of ACDC
and process it correctly.
"""
workload = self.createTestWorkload()
dcs = DataCollectionService(url=self.testInit.couchUrl,
database=self.testInit.couchDbName)
testFileA = File(lfn=makeUUID(),
size=1024,
events=1024,
locations="T1_US_FNAL_Disk")
testFileA.addRun(Run(1, 1, 2))
testFileA.create()
testFileB = File(lfn=makeUUID(),
size=1024,
events=1024,
locations="T1_US_FNAL_Disk")
testFileB.addRun(Run(1, 3))
testFileB.create()
testJobA = getJob(workload)
testJobA.addFile(testFileA)
testJobA.addFile(testFileB)
testFileC = File(lfn=makeUUID(),
size=1024,
events=1024,
locations="T1_US_FNAL_Disk")
testFileC.addRun(Run(1, 4, 6))
testFileC.create()
testJobB = getJob(workload)
testJobB.addFile(testFileC)
testFileD = File(lfn=makeUUID(),
size=1024,
events=1024,
locations="T1_US_FNAL_Disk")
testFileD.addRun(Run(1, 7))
testFileD.create()
testJobC = getJob(workload)
testJobC.addFile(testFileD)
testFileE = File(lfn=makeUUID(),
size=1024,
events=1024,
locations="T1_US_FNAL_Disk")
testFileE.addRun(Run(1, 11, 12))
testFileE.create()
testJobD = getJob(workload)
testJobD.addFile(testFileE)
testFileF = File(lfn=makeUUID(),
size=1024,
events=1024,
locations="T1_US_FNAL_Disk")
testFileF.addRun(Run(2, 5, 6, 7))
testFileF.create()
testJobE = getJob(workload)
testJobE.addFile(testFileF)
testFileG = File(lfn=makeUUID(),
size=1024,
events=1024,
locations="T1_US_FNAL_Disk")
testFileG.addRun(Run(2, 10, 11, 12))
testFileG.create()
testJobF = getJob(workload)
testJobF.addFile(testFileG)
testFileH = File(lfn=makeUUID(),
size=1024,
events=1024,
locations="T1_US_FNAL_Disk")
testFileH.addRun(Run(2, 15))
testFileH.create()
testJobG = getJob(workload)
testJobG.addFile(testFileH)
testFileI = File(lfn=makeUUID(),
size=1024,
events=1024,
locations="T1_US_FNAL_Disk")
testFileI.addRun(Run(3, 20))
testFileI.create()
testJobH = getJob(workload)
testJobH.addFile(testFileI)
testFileJ = File(lfn=makeUUID(),
size=1024,
events=1024,
locations="T1_US_FNAL_Disk")
testFileJ.addRun(Run(1, 9))
testFileJ.create()
testJobI = getJob(workload)
testJobI.addFile(testFileJ)
# dcs.failedJobs([testJobA, testJobB, testJobC, testJobD, testJobE,
# testJobF, testJobG, testJobH, testJobI])
dcs.failedJobs([testJobA, testJobD, testJobH])
baseName = makeUUID()
testFileset = Fileset(name=baseName)
testFileset.create()
testFileset.addFile(testFileA)
testFileset.addFile(testFileB)
testFileset.addFile(testFileC)
testFileset.addFile(testFileD)
testFileset.addFile(testFileE)
testFileset.addFile(testFileF)
testFileset.addFile(testFileG)
testFileset.addFile(testFileH)
testFileset.addFile(testFileI)
testFileset.addFile(testFileJ)
testFileset.commit()
testSubscription = Subscription(fileset=testFileset,
workflow=self.testWorkflow,
split_algo="LumiBased",
type="Processing")
testSubscription.create()
splitter = SplitterFactory()
jobFactory = splitter(package="WMCore.WMBS",
subscription=testSubscription)
jobGroups = jobFactory(
lumis_per_job=100,
halt_job_on_file_boundaries=False,
splitOnRun=True,
collectionName=workload.name(),
filesetName=workload.getTask("reco").getPathName(),
owner="evansde77",
group="DMWM",
couchURL=self.testInit.couchUrl,
couchDB=self.testInit.couchDbName,
performance=self.performanceParams)
self.assertEqual(jobGroups[0].jobs[0]['mask'].getRunAndLumis(),
{1: [[1, 2], [3, 3], [11, 12]]})
self.assertEqual(jobGroups[0].jobs[1]['mask'].getRunAndLumis(),
{3: [[20, 20]]})
return
class ErrorHandlerTest(EmulatedUnitTestCase):
"""
TestCase for TestErrorHandler module
"""
def setUp(self):
"""
setup for test.
"""
super(ErrorHandlerTest, self).setUp()
myThread = threading.currentThread()
self.testInit = TestInitCouchApp(__file__)
self.testInit.setLogging()
self.testInit.setDatabaseConnection()
self.testInit.setSchema(customModules=["WMCore.WMBS"],
useDefault=False)
self.testInit.setupCouch("errorhandler_t", "GroupUser", "ACDC")
self.testInit.setupCouch("errorhandler_t_jd/jobs", "JobDump")
self.testInit.setupCouch("errorhandler_t_jd/fwjrs", "FWJRDump")
self.daofactory = DAOFactory(package="WMCore.WMBS",
logger=myThread.logger,
dbinterface=myThread.dbi)
self.getJobs = self.daofactory(classname="Jobs.GetAllJobs")
self.setJobTime = self.daofactory(classname="Jobs.SetStateTime")
locationAction = self.daofactory(classname="Locations.New")
locationAction.execute(siteName="malpaquet", pnn="T2_CH_CERN")
self.testDir = self.testInit.generateWorkDir()
self.configFile = EmulatorSetup.setupWMAgentConfig()
self.nJobs = 10
self.dataCS = DataCollectionService(url=self.testInit.couchUrl,
database="errorhandler_t")
return
def tearDown(self):
"""
Database deletion
"""
self.testInit.clearDatabase()
self.testInit.delWorkDir()
self.testInit.tearDownCouch()
EmulatorSetup.deleteConfig(self.configFile)
return
def getConfig(self):
"""
_getConfig_
"""
config = self.testInit.getConfiguration()
self.testInit.generateWorkDir(config)
# First the general stuff
config.section_("General")
config.General.workDir = os.getenv("TESTDIR", self.testDir)
config.section_("CoreDatabase")
config.CoreDatabase.connectUrl = os.getenv("DATABASE")
config.CoreDatabase.socket = os.getenv("DBSOCK")
config.component_("ErrorHandler")
# The log level of the component.
config.ErrorHandler.logLevel = 'INFO'
# The namespace of the component
config.ErrorHandler.namespace = 'WMComponent.ErrorHandler.ErrorHandler'
# maximum number of threads we want to deal
# config.ErrorHandler.maxThreads = 30
# with messages per pool.
config.ErrorHandler.maxProcessSize = 30
config.ErrorHandler.readFWJR = True
# maximum number of retries we want for job
config.ErrorHandler.maxRetries = 5
# The poll interval at which to look for failed jobs
config.ErrorHandler.pollInterval = 60
# this will be overwritten in some unittests
# JobStateMachine
config.component_('JobStateMachine')
config.JobStateMachine.couchurl = os.getenv('COUCHURL', None)
config.JobStateMachine.couchDBName = "errorhandler_t_jd"
config.section_('ACDC')
config.ACDC.couchurl = self.testInit.couchUrl
config.ACDC.database = "errorhandler_t"
return config
def createWorkload(self, workloadName='Test'):
"""
_createTestWorkload_
Creates a test workload for us to run on, hold the basic necessities.
"""
workload = testWorkload(workloadName)
# Add RequestManager stuff
workload.data.request.section_('schema')
workload.data.request.schema.Requestor = 'nobody'
workload.data.request.schema.Group = 'testers'
taskMaker = TaskMaker(workload,
os.path.join(self.testDir, 'workloadTest'))
taskMaker.skipSubscription = True
taskMaker.processWorkload()
return workload
def createTestJobGroup(self,
nJobs=10,
retry_count=1,
workloadPath='test',
fwjrPath=None,
workloadName=makeUUID(),
fileModifier=''):
"""
Creates a group of several jobs
"""
myThread = threading.currentThread()
myThread.transaction.begin()
testWorkflow = Workflow(spec=workloadPath,
owner="cmsdataops",
group="cmsdataops",
name=workloadName,
task="/TestWorkload/ReReco")
testWorkflow.create()
testWMBSFileset = Fileset(name="TestFileset")
testWMBSFileset.create()
testSubscription = Subscription(fileset=testWMBSFileset,
workflow=testWorkflow)
testSubscription.create()
testJobGroup = JobGroup(subscription=testSubscription)
testJobGroup.create()
testFile0 = File(lfn="/this/is/a/parent%s" % fileModifier,
size=1024,
events=10)
testFile0.addRun(Run(10, *[12312]))
testFile0.setLocation('T2_CH_CERN')
testFileA = File(lfn="/this/is/a/lfnA%s" % fileModifier,
size=1024,
events=10,
first_event=88,
merged=False)
testFileA.addRun(Run(10, *[12312, 12313]))
testFileA.setLocation('T2_CH_CERN')
testFileB = File(lfn="/this/is/a/lfnB%s" % fileModifier,
size=1024,
events=10,
first_event=88,
merged=False)
testFileB.addRun(Run(10, *[12314, 12315, 12316]))
testFileB.setLocation('T2_CH_CERN')
testFile0.create()
testFileA.create()
testFileB.create()
testFileA.addParent(lfn="/this/is/a/parent%s" % fileModifier)
testFileB.addParent(lfn="/this/is/a/parent%s" % fileModifier)
for i in range(0, nJobs):
testJob = Job(name=makeUUID())
testJob['retry_count'] = retry_count
testJob['retry_max'] = 10
testJob['mask'].addRunAndLumis(run=10, lumis=[12312])
testJob['mask'].addRunAndLumis(run=10, lumis=[12314, 12316])
testJob['cache_dir'] = os.path.join(self.testDir, testJob['name'])
testJob['fwjr_path'] = fwjrPath
os.mkdir(testJob['cache_dir'])
testJobGroup.add(testJob)
testJob.create(group=testJobGroup)
testJob.addFile(testFileA)
testJob.addFile(testFileB)
testJob.save()
testJobGroup.commit()
testSubscription.acquireFiles(files=[testFileA, testFileB])
testSubscription.save()
myThread.transaction.commit()
return testJobGroup
def testA_Create(self):
"""
WMComponent_t.ErrorHandler_t.ErrorHandler_t:testCreate()
Mimics creation of component and test jobs failed in create stage.
"""
njobs = 4
workloadName = 'TestWorkload'
self.createWorkload(workloadName=workloadName)
workloadPath = os.path.join(self.testDir, 'workloadTest', workloadName,
'WMSandbox', 'WMWorkload.pkl')
# testJobGroup = self.createTestJobGroup(nJobs=self.nJobs,
testJobGroup = self.createTestJobGroup(nJobs=njobs,
workloadPath=workloadPath,
workloadName=workloadName)
config = self.getConfig()
changer = ChangeState(config)
changer.propagate(testJobGroup.jobs, 'created', 'new')
changer.propagate(testJobGroup.jobs, 'createfailed', 'created')
idList = self.getJobs.execute(state='CreateFailed')
self.assertEqual(len(idList), njobs)
testErrorHandler = ErrorHandlerPoller(config)
# set reqAuxDB None for the test,
testErrorHandler.reqAuxDB = None
testErrorHandler.setup(None)
testErrorHandler.algorithm(None)
idList = self.getJobs.execute(state='CreateFailed')
self.assertEqual(len(idList), 0)
# These should go directly to exhausted
idList = self.getJobs.execute(state='Exhausted')
self.assertEqual(len(idList), njobs)
# Check that it showed up in ACDC
collection = self.dataCS.getDataCollection(workloadName)
# Now look at what's inside
self.assertTrue(len(collection['filesets']) > 0)
for fileset in collection["filesets"]:
counter = 0
for f in fileset.listFiles():
counter += 1
self.assertTrue(
f['lfn'] in ["/this/is/a/lfnA", "/this/is/a/lfnB"])
self.assertEqual(f['events'], 10)
self.assertEqual(f['size'], 1024)
self.assertEqual(f['parents'], [u'/this/is/a/parent'])
self.assertTrue(f['runs'][0]['run_number'] == 10)
if f['lfn'] == "/this/is/a/lfnA":
self.assertItemsEqual(f['runs'][0]['lumis'], [12312])
elif f['lfn'] == "/this/is/a/lfnB":
self.assertItemsEqual(f['runs'][0]['lumis'],
[12314, 12315, 12316])
else:
self.assertFail("File name is not known: %s" % f['lfn'])
self.assertEqual(f['merged'], 0)
self.assertEqual(f['first_event'], 88)
self.assertEqual(
counter,
njobs * 2) # each job has 2 files (thus 4 times duplicate)
return
def testB_Submit(self):
"""
WMComponent_t.ErrorHandler_t.ErrorHandler_t:testSubmit()
Mimics creation of component and test jobs failed in submit stage.
"""
workloadName = 'TestWorkload'
self.createWorkload(workloadName=workloadName)
workloadPath = os.path.join(self.testDir, 'workloadTest', workloadName,
'WMSandbox', 'WMWorkload.pkl')
testJobGroup = self.createTestJobGroup(nJobs=self.nJobs,
workloadPath=workloadPath)
config = self.getConfig()
changer = ChangeState(config)
changer.propagate(testJobGroup.jobs, 'created', 'new')
changer.propagate(testJobGroup.jobs, 'submitfailed', 'created')
idList = self.getJobs.execute(state='SubmitFailed')
self.assertEqual(len(idList), self.nJobs)
testErrorHandler = ErrorHandlerPoller(config)
# set reqAuxDB None for the test,
testErrorHandler.reqAuxDB = None
testErrorHandler.setup(None)
testErrorHandler.algorithm(None)
idList = self.getJobs.execute(state='SubmitFailed')
self.assertEqual(len(idList), 0)
idList = self.getJobs.execute(state='SubmitCooloff')
self.assertEqual(len(idList), self.nJobs)
return
def testC_Jobs(self):
"""
WMComponent_t.ErrorHandler_t.ErrorHandler_t.testJobs()
Mimics creation of component and test jobs failed in execute stage.
"""
workloadName = 'TestWorkload'
self.createWorkload(workloadName=workloadName)
workloadPath = os.path.join(self.testDir, 'workloadTest', workloadName,
'WMSandbox', 'WMWorkload.pkl')
testJobGroup = self.createTestJobGroup(nJobs=self.nJobs,
workloadPath=workloadPath)
config = self.getConfig()
changer = ChangeState(config)
changer.propagate(testJobGroup.jobs, 'created', 'new')
changer.propagate(testJobGroup.jobs, 'executing', 'created')
changer.propagate(testJobGroup.jobs, 'complete', 'executing')
changer.propagate(testJobGroup.jobs, 'jobfailed', 'complete')
idList = self.getJobs.execute(state='JobFailed')
self.assertEqual(len(idList), self.nJobs)
testErrorHandler = ErrorHandlerPoller(config)
# set reqAuxDB None for the test,
testErrorHandler.reqAuxDB = None
testErrorHandler.setup(None)
testErrorHandler.algorithm(None)
idList = self.getJobs.execute(state='JobFailed')
self.assertEqual(len(idList), 0)
idList = self.getJobs.execute(state='JobCooloff')
self.assertEqual(len(idList), self.nJobs)
return
def testD_Exhausted(self):
"""
_testExhausted_
Test that the system can exhaust jobs correctly
"""
workloadName = 'TestWorkload'
self.createWorkload(workloadName=workloadName)
workloadPath = os.path.join(self.testDir, 'workloadTest', workloadName,
'WMSandbox', 'WMWorkload.pkl')
testJobGroup = self.createTestJobGroup(nJobs=self.nJobs,
retry_count=5,
workloadPath=workloadPath)
config = self.getConfig()
config.ErrorHandler.maxRetries = 1
changer = ChangeState(config)
changer.propagate(testJobGroup.jobs, 'created', 'new')
changer.propagate(testJobGroup.jobs, 'executing', 'created')
changer.propagate(testJobGroup.jobs, 'complete', 'executing')
changer.propagate(testJobGroup.jobs, 'jobfailed', 'complete')
testSubscription = Subscription(id=1) # You should only have one
testSubscription.load()
testSubscription.loadData()
# Do we have files to start with?
self.assertEqual(len(testSubscription.filesOfStatus("Acquired")), 2)
testErrorHandler = ErrorHandlerPoller(config)
# set reqAuxDB None for the test,
testErrorHandler.reqAuxDB = None
testErrorHandler.setup(None)
testErrorHandler.algorithm(None)
idList = self.getJobs.execute(state='JobFailed')
self.assertEqual(len(idList), 0)
idList = self.getJobs.execute(state='JobCooloff')
self.assertEqual(len(idList), 0)
idList = self.getJobs.execute(state='Exhausted')
self.assertEqual(len(idList), self.nJobs)
# Did we fail the files?
self.assertEqual(len(testSubscription.filesOfStatus("Acquired")), 0)
self.assertEqual(len(testSubscription.filesOfStatus("Failed")), 2)
def testE_FailJobs(self):
"""
_FailJobs_
Test our ability to fail jobs based on the information in the FWJR
"""
workloadName = 'TestWorkload'
self.createWorkload(workloadName=workloadName)
workloadPath = os.path.join(self.testDir, 'workloadTest', workloadName,
'WMSandbox', 'WMWorkload.pkl')
fwjrPath = os.path.join(WMCore.WMBase.getTestBase(),
"WMComponent_t/JobAccountant_t",
"fwjrs/badBackfillJobReport.pkl")
testJobGroup = self.createTestJobGroup(nJobs=self.nJobs,
workloadPath=workloadPath,
fwjrPath=fwjrPath)
badJobGroup = self.createTestJobGroup(nJobs=self.nJobs,
workloadPath=workloadPath,
fwjrPath=None,
fileModifier='bad')
config = self.getConfig()
config.ErrorHandler.readFWJR = True
changer = ChangeState(config)
changer.propagate(testJobGroup.jobs, 'created', 'new')
changer.propagate(testJobGroup.jobs, 'executing', 'created')
changer.propagate(testJobGroup.jobs, 'complete', 'executing')
changer.propagate(testJobGroup.jobs, 'jobfailed', 'complete')
changer.propagate(badJobGroup.jobs, 'created', 'new')
changer.propagate(badJobGroup.jobs, 'executing', 'created')
changer.propagate(badJobGroup.jobs, 'complete', 'executing')
changer.propagate(badJobGroup.jobs, 'jobfailed', 'complete')
testErrorHandler = ErrorHandlerPoller(config)
# set reqAuxDB None for the test,
testErrorHandler.reqAuxDB = None
testErrorHandler.setup(None)
testErrorHandler.exitCodesNoRetry = [8020]
testErrorHandler.algorithm(None)
# This should exhaust all jobs due to exit code
# Except those with no fwjr
idList = self.getJobs.execute(state='JobFailed')
self.assertEqual(len(idList), 0)
idList = self.getJobs.execute(state='JobCooloff')
self.assertEqual(len(idList), self.nJobs)
idList = self.getJobs.execute(state='Exhausted')
self.assertEqual(len(idList), self.nJobs)
config.ErrorHandler.maxFailTime = -10
testErrorHandler2 = ErrorHandlerPoller(config)
# set reqAuxDB None for the test,
testErrorHandler2.reqAuxDB = None
changer.propagate(testJobGroup.jobs, 'created', 'new')
changer.propagate(testJobGroup.jobs, 'executing', 'created')
changer.propagate(testJobGroup.jobs, 'complete', 'executing')
changer.propagate(testJobGroup.jobs, 'jobfailed', 'complete')
testErrorHandler2.algorithm(None)
# This should exhaust all jobs due to timeout
idList = self.getJobs.execute(state='JobFailed')
self.assertEqual(len(idList), 0)
idList = self.getJobs.execute(state='JobCooloff')
self.assertEqual(len(idList), self.nJobs)
idList = self.getJobs.execute(state='Exhausted')
self.assertEqual(len(idList), self.nJobs)
config.ErrorHandler.maxFailTime = 24 * 3600
config.ErrorHandler.passExitCodes = [8020]
testErrorHandler3 = ErrorHandlerPoller(config)
# set reqAuxDB None for the test,
testErrorHandler3.reqAuxDB = None
changer.propagate(testJobGroup.jobs, 'created', 'new')
changer.propagate(testJobGroup.jobs, 'executing', 'created')
changer.propagate(testJobGroup.jobs, 'complete', 'executing')
changer.propagate(testJobGroup.jobs, 'jobfailed', 'complete')
testErrorHandler3.algorithm(None)
# This should pass all jobs due to exit code
idList = self.getJobs.execute(state='Created')
self.assertEqual(len(idList), self.nJobs)
return
@attr('integration')
def testZ_Profile(self):
"""
_testProfile_
Do a full profile of the poller
"""
nJobs = 100
workloadName = 'TestWorkload'
self.createWorkload(workloadName=workloadName)
workloadPath = os.path.join(self.testDir, 'workloadTest', workloadName,
'WMSandbox', 'WMWorkload.pkl')
testJobGroup = self.createTestJobGroup(nJobs=nJobs,
workloadPath=workloadPath)
config = self.getConfig()
changer = ChangeState(config)
changer.propagate(testJobGroup.jobs, 'created', 'new')
changer.propagate(testJobGroup.jobs, 'executing', 'created')
changer.propagate(testJobGroup.jobs, 'complete', 'executing')
changer.propagate(testJobGroup.jobs, 'jobfailed', 'complete')
idList = self.getJobs.execute(state='JobFailed')
self.assertEqual(len(idList), nJobs)
testErrorHandler = ErrorHandlerPoller(config)
# set reqAuxDB None for the test,
testErrorHandler.reqAuxDB = None
testErrorHandler.setup(None)
startTime = time.time()
cProfile.runctx("testErrorHandler.algorithm()",
globals(),
locals(),
filename="profStats.stat")
stopTime = time.time()
idList = self.getJobs.execute(state='CreateFailed')
self.assertEqual(len(idList), 0)
idList = self.getJobs.execute(state='JobFailed')
self.assertEqual(len(idList), 0)
idList = self.getJobs.execute(state='JobCooloff')
self.assertEqual(len(idList), nJobs)
print("Took %f seconds to run polling algo" % (stopTime - startTime))
p = pstats.Stats('profStats.stat')
p.sort_stats('cumulative')
p.print_stats(0.2)
return
def algorithm(self, *args, **kwargs):
"""
_algorithm_
Split files into a number of lumis per job
Allow a flag to determine if we split files between jobs
"""
myThread = threading.currentThread()
lumisPerJob = int(kwargs.get('lumis_per_job', 1))
splitOnFile = bool(kwargs.get('halt_job_on_file_boundaries', True))
ignoreACDC = bool(kwargs.get('ignore_acdc_except', False))
collectionName = kwargs.get('collectionName', None)
splitOnRun = kwargs.get('splitOnRun', True)
getParents = kwargs.get('include_parents', False)
runWhitelist = kwargs.get('runWhitelist', [])
runs = kwargs.get('runs', None)
lumis = kwargs.get('lumis', None)
deterministicPileup = kwargs.get('deterministicPileup', False)
eventsPerLumiInDataset = 0
if deterministicPileup and self.package == 'WMCore.WMBS':
getJobNumber = self.daoFactory(
classname="Jobs.GetNumberOfJobsPerWorkflow")
jobNumber = getJobNumber.execute(
workflow=self.subscription.getWorkflow().id)
self.nJobs = jobNumber
timePerEvent, sizePerEvent, memoryRequirement = \
self.getPerformanceParameters(kwargs.get('performance', {}))
goodRunList = {}
if runs and lumis:
goodRunList = buildLumiMask(runs, lumis)
# If we have runLumi info, we need to load it from couch
if collectionName:
try:
from WMCore.ACDC.DataCollectionService import DataCollectionService
couchURL = kwargs.get('couchURL')
couchDB = kwargs.get('couchDB')
filesetName = kwargs.get('filesetName')
collectionName = kwargs.get('collectionName')
owner = kwargs.get('owner')
group = kwargs.get('group')
logging.info('Creating jobs for ACDC fileset %s' % filesetName)
dcs = DataCollectionService(couchURL, couchDB)
goodRunList = dcs.getLumiWhitelist(collectionName, filesetName,
owner, group)
except Exception, ex:
msg = "Exception while trying to load goodRunList\n"
if ignoreACDC:
msg += "Ditching goodRunList\n"
msg += str(ex)
msg += str(traceback.format_exc())
logging.error(msg)
goodRunList = {}
else:
msg += "Refusing to create any jobs.\n"
msg += str(ex)
msg += str(traceback.format_exc())
logging.error(msg)
return
class ErrorHandlerPoller(BaseWorkerThread):
"""
Polls for Error Conditions, handles them
"""
def __init__(self, config):
"""
Initialise class members
"""
BaseWorkerThread.__init__(self)
self.config = config
myThread = threading.currentThread()
self.daoFactory = DAOFactory(package = "WMCore.WMBS",
logger = myThread.logger,
dbinterface = myThread.dbi)
self.changeState = ChangeState(self.config)
self.maxRetries = self.config.ErrorHandler.maxRetries
if type(self.maxRetries) != dict:
self.maxRetries = {'default' : self.maxRetries}
if 'default' not in self.maxRetries:
raise ErrorHandlerException('Max retries for the default job type must be specified')
self.maxProcessSize = getattr(self.config.ErrorHandler, 'maxProcessSize', 250)
self.exitCodes = getattr(self.config.ErrorHandler, 'failureExitCodes', [])
self.maxFailTime = getattr(self.config.ErrorHandler, 'maxFailTime', 32 * 3600)
self.readFWJR = getattr(self.config.ErrorHandler, 'readFWJR', False)
self.passCodes = getattr(self.config.ErrorHandler, 'passExitCodes', [])
self.getJobs = self.daoFactory(classname = "Jobs.GetAllJobs")
self.idLoad = self.daoFactory(classname = "Jobs.LoadFromIDWithType")
self.loadAction = self.daoFactory(classname = "Jobs.LoadForErrorHandler")
self.dataCollection = DataCollectionService(url = config.ACDC.couchurl,
database = config.ACDC.database)
# initialize the alert framework (if available - config.Alert present)
# self.sendAlert will be then be available
self.initAlerts(compName = "ErrorHandler")
# Some exit codes imply an immediate failure, non-configurable
self.exitCodes.extend(WMJobPermanentSystemErrors)
return
def setup(self, parameters = None):
"""
Load DB objects required for queries
"""
# For now, does nothing
return
def terminate(self, params):
"""
_terminate_
Do one pass, then commit suicide
"""
logging.debug("terminating. doing one more pass before we die")
self.algorithm(params)
def exhaustJobs(self, jobList):
"""
_exhaustJobs_
Actually do the jobs exhaustion
"""
self.changeState.propagate(jobList, 'exhausted', 'retrydone')
# Remove all the files in the exhausted jobs.
logging.debug("About to fail input files for exhausted jobs")
for job in jobList:
job.failInputFiles()
self.handleACDC(jobList)
return
def processRetries(self, jobList, state):
"""
_processRetries_
Actually do the retries
"""
logging.info("Processing retries for %d failed jobs of type %sfailed" % (len(jobList), state))
retrydoneJobs = []
cooloffJobs = []
passJobs = []
# Retries < max retry count
for job in jobList:
allowedRetries = self.maxRetries.get(job['type'], self.maxRetries['default'])
# Retries < allowed max retry count
if job['retry_count'] < allowedRetries and state != 'create':
cooloffJobs.append(job)
# Check if Retries >= allowed max retry count
elif job['retry_count'] >= allowedRetries or state == 'create':
retrydoneJobs.append(job)
msg = "Stopping retries for job %d" % job['id']
logging.debug(msg)
self.sendAlert(4, msg = msg)
logging.debug("JobInfo: %s" % job)
if self.readFWJR:
# Then we have to check each FWJR for exit status
cooloffJobs, passJobs, retrydoneFWJRJobs = self.readFWJRForErrors(cooloffJobs)
retrydoneJobs.extend(retrydoneFWJRJobs)
# Now to actually do something.
logging.debug("About to propagate jobs")
if len(retrydoneJobs) > 0:
self.changeState.propagate(retrydoneJobs, 'retrydone',
'%sfailed' % state, updatesummary = True)
if len(cooloffJobs) > 0:
self.changeState.propagate(cooloffJobs, '%scooloff' % state,
'%sfailed' % state, updatesummary = True)
if len(passJobs) > 0:
# Overwrite the transition states and move directly to created
self.changeState.propagate(passJobs, 'created', 'new')
return
def handleACDC(self, jobList):
"""
_handleACDC_
Do the ACDC creation and hope it works
"""
idList = [x['id'] for x in jobList]
logging.info("Starting to build ACDC with %i jobs" % len(idList))
logging.info("This operation will take some time...")
loadList = self.loadJobsFromListFull(idList)
for job in loadList:
job.getMask()
self.dataCollection.failedJobs(loadList)
return
def readFWJRForErrors(self, jobList):
"""
_readFWJRForErrors_
Check the FWJRs of the failed jobs
and determine those that can be retried
and which must be retried without going through cooloff.
Returns a triplet with cooloff, passed and exhausted jobs.
"""
cooloffJobs = []
passJobs = []
exhaustJobs = []
for job in jobList:
report = Report()
reportPath = job['fwjr_path']
if reportPath is None:
logging.error("No FWJR in job %i, ErrorHandler can't process it.\n Passing it to cooloff." % job['id'])
cooloffJobs.append(job)
continue
if not os.path.isfile(reportPath):
logging.error("Failed to find FWJR for job %i in location %s.\n Passing it to cooloff." % (job['id'], reportPath))
cooloffJobs.append(job)
continue
try:
report.load(reportPath)
# First let's check the time conditions
times = report.getFirstStartLastStop()
startTime = None
stopTime = None
if times is not None:
startTime = times['startTime']
stopTime = times['stopTime']
if startTime == None or stopTime == None:
# We have no information to make a decision, keep going.
logging.debug("No start, stop times for steps for job %i" % job['id'])
elif stopTime - startTime > self.maxFailTime:
msg = "Job %i exhausted after running on node for %i seconds" % (job['id'], stopTime - startTime)
logging.debug(msg)
exhaustJobs.append(job)
continue
if len([x for x in report.getExitCodes() if x in self.exitCodes]):
msg = "Job %i exhausted due to a bad exit code (%s)" % (job['id'], str(report.getExitCodes()))
logging.error(msg)
self.sendAlert(4, msg = msg)
exhaustJobs.append(job)
continue
if len([x for x in report.getExitCodes() if x in self.passCodes]):
msg = "Job %i restarted immediately due to an exit code (%s)" % (job['id'], str(report.getExitCodes()))
passJobs.append(job)
continue
cooloffJobs.append(job)
except Exception, ex:
logging.warning("Exception while trying to check jobs for failures!")
logging.warning(str(ex))
logging.warning("Ignoring and sending job to cooloff")
cooloffJobs.append(job)
return cooloffJobs, passJobs, exhaustJobs
def algorithm(self, *args, **kwargs):
"""
_algorithm_
An event base splitting algorithm. All available files are split into a
set number of events per job.
"""
eventsPerJob = int(kwargs.get("events_per_job", 100))
eventsPerLumi = int(kwargs.get("events_per_lumi", eventsPerJob))
getParents = kwargs.get("include_parents", False)
lheInput = kwargs.get("lheInputFiles", False)
collectionName = kwargs.get('collectionName', None)
timePerEvent, sizePerEvent, memoryRequirement = \
self.getPerformanceParameters(kwargs.get('performance', {}))
acdcFileList = []
deterministicPileup = kwargs.get('deterministicPileup', False)
if deterministicPileup and self.package == 'WMCore.WMBS':
getJobNumber = self.daoFactory(classname="Jobs.GetNumberOfJobsPerWorkflow")
self.nJobs = getJobNumber.execute(workflow=self.subscription.getWorkflow().id)
logging.info('Creating %d jobs in DeterministicPileup mode', self.nJobs)
# If we have runLumi info, we need to load it from couch
if collectionName:
try:
from WMCore.ACDC.DataCollectionService import DataCollectionService
couchURL = kwargs.get('couchURL')
couchDB = kwargs.get('couchDB')
filesetName = kwargs.get('filesetName')
collectionName = kwargs.get('collectionName')
owner = kwargs.get('owner')
group = kwargs.get('group')
logging.info('Creating jobs for ACDC fileset %s', filesetName)
dcs = DataCollectionService(couchURL, couchDB)
acdcFileList = dcs.getProductionACDCInfo(collectionName, filesetName, owner, group)
except Exception as ex:
msg = "Exception while trying to load goodRunList\n"
msg += "Refusing to create any jobs.\n"
msg += str(ex)
msg += str(traceback.format_exc())
logging.error(msg)
return
totalJobs = 0
locationDict = self.sortByLocation()
for location in locationDict:
self.newGroup()
fileList = locationDict[location]
getRunLumiInformation = False
for f in fileList:
if f['lfn'].startswith("MCFakeFile"):
# We have one MCFakeFile, then it needs run information
getRunLumiInformation = True
break
if getRunLumiInformation:
if self.package == 'WMCore.WMBS':
loadRunLumi = self.daoFactory(classname="Files.GetBulkRunLumi")
fileLumis = loadRunLumi.execute(files=fileList)
for f in fileList:
lumiDict = fileLumis.get(f['id'], {})
for run in lumiDict.keys():
f.addRun(run=Run(run, *lumiDict[run]))
for f in fileList:
currentEvent = f['first_event']
eventsInFile = f['events']
runs = list(f['runs'])
# We got the runs, clean the file.
f['runs'] = set()
if getParents:
parentLFNs = self.findParent(lfn=f['lfn'])
for lfn in parentLFNs:
parent = File(lfn=lfn)
f['parents'].add(parent)
if acdcFileList:
if f['lfn'] in [x['lfn'] for x in acdcFileList]:
totalJobs = self.createACDCJobs(f, acdcFileList, timePerEvent,
sizePerEvent, memoryRequirement,
lheInput, eventsPerJob, eventsPerLumi,
deterministicPileup, totalJobs)
continue
if not f['lfn'].startswith("MCFakeFile"):
# Very very uncommon, but it has real input dataset
if eventsInFile >= eventsPerJob:
while currentEvent < eventsInFile:
self.newJob(name=self.getJobName(length=totalJobs))
self.currentJob.addFile(f)
if eventsPerJob + currentEvent < eventsInFile:
jobTime = eventsPerJob * timePerEvent
diskRequired = eventsPerJob * sizePerEvent
self.currentJob["mask"].setMaxAndSkipEvents(eventsPerJob, currentEvent)
else:
jobTime = (eventsInFile - currentEvent) * timePerEvent
diskRequired = (eventsInFile - currentEvent) * sizePerEvent
self.currentJob["mask"].setMaxAndSkipEvents(None,
currentEvent)
self.currentJob.addResourceEstimates(jobTime=jobTime,
memory=memoryRequirement,
disk=diskRequired)
if deterministicPileup:
self.currentJob.addBaggageParameter("skipPileupEvents", (self.nJobs - 1) * eventsPerJob)
logging.debug("Job created for real input with %s", self.currentJob)
currentEvent += eventsPerJob
totalJobs += 1
else:
self.newJob(name=self.getJobName(length=totalJobs))
self.currentJob.addFile(f)
jobTime = eventsInFile * timePerEvent
diskRequired = eventsInFile * sizePerEvent
self.currentJob.addResourceEstimates(jobTime=jobTime,
memory=memoryRequirement,
disk=diskRequired)
if deterministicPileup:
self.currentJob.addBaggageParameter("skipPileupEvents", (self.nJobs - 1) * eventsPerJob)
logging.debug("Last job created for real input with %s", self.currentJob)
totalJobs += 1
else:
# This assumes there's only one run which is the case for MC
lumis = runs[0].lumis
(firstLumi, lastLumi) = (min(lumis), max(lumis))
currentLumi = firstLumi
totalEvents = 0
if eventsInFile >= eventsPerJob:
while totalEvents < eventsInFile:
self.newJob(name=self.getJobName(length=totalJobs))
self.currentJob.addFile(f)
self.currentJob.addBaggageParameter("lheInputFiles", lheInput)
lumisPerJob = int(ceil(float(eventsPerJob)
/ eventsPerLumi))
# Limit the number of events to a unsigned 32bit int
eventsRemaining = eventsInFile - totalEvents
if (currentEvent + eventsPerJob - 1) > (2 ** 32 - 1) and (
currentEvent + eventsRemaining - 1) > (2 ** 32 - 1):
currentEvent = 1
if eventsRemaining > eventsPerJob:
self.currentJob["mask"].setMaxAndSkipEvents(eventsPerJob,
currentEvent)
self.currentJob["mask"].setMaxAndSkipLumis(lumisPerJob,
currentLumi)
jobTime = eventsPerJob * timePerEvent
diskRequired = eventsPerJob * sizePerEvent
else:
jobTime = eventsRemaining * timePerEvent
diskRequired = eventsRemaining * sizePerEvent
lumisPerJob = int(ceil(float(eventsRemaining) / eventsPerLumi))
self.currentJob["mask"].setMaxAndSkipEvents(eventsRemaining,
currentEvent)
self.currentJob["mask"].setMaxAndSkipLumis(lumisPerJob,
currentLumi)
if deterministicPileup:
self.currentJob.addBaggageParameter("skipPileupEvents", (self.nJobs - 1) * eventsPerJob)
currentLumi += lumisPerJob
currentEvent += eventsPerJob
totalEvents += eventsPerJob
totalJobs += 1
self.currentJob.addResourceEstimates(jobTime=jobTime,
memory=memoryRequirement,
disk=diskRequired)
else:
self.newJob(name=self.getJobName(length=totalJobs))
self.currentJob.addFile(f)
# For MC we use firstEvent instead of skipEvents so set it to 1
# We must check for events going over 2**32 - 1 here too
if (eventsInFile + currentEvent - 1) > (2 ** 32 - 1):
currentEvent = 1
self.currentJob["mask"].setMaxAndSkipEvents(eventsInFile,
currentEvent)
self.currentJob["mask"].setMaxAndSkipLumis(lastLumi -
currentLumi + 1, currentLumi)
jobTime = eventsInFile * timePerEvent
diskRequired = eventsInFile * sizePerEvent
self.currentJob.addResourceEstimates(jobTime=jobTime,
memory=memoryRequirement,
disk=diskRequired)
if deterministicPileup:
self.currentJob.addBaggageParameter("skipPileupEvents", (self.nJobs - 1) * eventsPerJob)
totalJobs += 1
def algorithm(self, *args, **kwargs):
"""
_algorithm_
An event base splitting algorithm. All available files are split into a
set number of events per job.
"""
eventsPerJob = int(kwargs.get("events_per_job", 100))
eventsPerLumi = int(kwargs.get("events_per_lumi", eventsPerJob))
getParents = kwargs.get("include_parents", False)
lheInput = kwargs.get("lheInputFiles", False)
collectionName = kwargs.get('collectionName', None)
timePerEvent, sizePerEvent, memoryRequirement = \
self.getPerformanceParameters(kwargs.get('performance', {}))
acdcFileList = []
deterministicPileup = kwargs.get('deterministicPileup', False)
if eventsPerJob <= 0 or eventsPerLumi <= 0:
msg = "events_per_job and events_per_lumi must be positive. Their values are: "
msg += "events_per_job: %d, events_per_lumi: %d" % (eventsPerJob, eventsPerLumi)
raise RuntimeError(msg)
if deterministicPileup and self.package == 'WMCore.WMBS':
getJobNumber = self.daoFactory(classname="Jobs.GetNumberOfJobsPerWorkflow")
self.nJobs = getJobNumber.execute(workflow=self.subscription.getWorkflow().id)
logging.info('Creating jobs in DeterministicPileup mode for %s',
self.subscription.workflowName())
# If we have runLumi info, we need to load it from couch
if collectionName:
try:
from WMCore.ACDC.DataCollectionService import DataCollectionService
couchURL = kwargs.get('couchURL')
couchDB = kwargs.get('couchDB')
filesetName = kwargs.get('filesetName')
collectionName = kwargs.get('collectionName')
logging.info('Loading ACDC info for collectionName: %s, with filesetName: %s', collectionName,
filesetName)
dcs = DataCollectionService(couchURL, couchDB)
acdcFileList = dcs.getProductionACDCInfo(collectionName, filesetName)
except Exception as ex:
msg = "Exception while trying to load goodRunList\n"
msg += "Refusing to create any jobs.\n"
msg += str(ex)
msg += str(traceback.format_exc())
logging.error(msg)
return
totalJobs = 0
locationDict = self.sortByLocation()
for location in locationDict:
self.newGroup()
fileList = locationDict[location]
getRunLumiInformation = False
for f in fileList:
if f['lfn'].startswith("MCFakeFile"):
# We have one MCFakeFile, then it needs run information
getRunLumiInformation = True
break
if getRunLumiInformation:
if self.package == 'WMCore.WMBS':
loadRunLumi = self.daoFactory(classname="Files.GetBulkRunLumi")
fileLumis = loadRunLumi.execute(files=fileList)
if not fileLumis:
logging.warning("Empty fileLumis dict for workflow %s, subs %s.",
self.subscription.workflowName(), self.subscription['id'])
for f in fileList:
lumiDict = fileLumis.get(f['id'], {})
for run in lumiDict:
f.addRun(run=Run(run, *lumiDict[run]))
for f in fileList:
currentEvent = f['first_event']
eventsInFile = f['events']
runs = list(f['runs'])
# We got the runs, clean the file.
f['runs'] = set()
if getParents:
parentLFNs = self.findParent(lfn=f['lfn'])
for lfn in parentLFNs:
parent = File(lfn=lfn)
f['parents'].add(parent)
if acdcFileList:
totalJobs = self.createACDCJobs(f, acdcFileList, timePerEvent,
sizePerEvent, memoryRequirement,
lheInput, eventsPerJob, eventsPerLumi,
deterministicPileup, totalJobs)
continue
if not f['lfn'].startswith("MCFakeFile"):
# there might be files with 0 event that still have to be processed
if eventsInFile == 0:
self.newJob(name=self.getJobName(length=totalJobs))
self.currentJob.addFile(f)
# Do not set LastEvent
self.currentJob["mask"].setMaxAndSkipEvents(None, currentEvent)
self.currentJob.addResourceEstimates(jobTime=0,
memory=memoryRequirement,
disk=0)
if deterministicPileup:
self.currentJob.addBaggageParameter("skipPileupEvents", (self.nJobs - 1) * eventsPerJob)
totalJobs += 1
logging.info("Job created for 0-event input file with %s", self.currentJob)
# Very very uncommon in production, but it has real input dataset
while eventsInFile:
self.newJob(name=self.getJobName(length=totalJobs))
self.currentJob.addFile(f)
if eventsInFile >= eventsPerJob:
jobTime = eventsPerJob * timePerEvent
diskRequired = eventsPerJob * sizePerEvent
self.currentJob["mask"].setMaxAndSkipEvents(eventsPerJob - 1, currentEvent)
else:
jobTime = eventsInFile * timePerEvent
diskRequired = eventsInFile * sizePerEvent
self.currentJob["mask"].setMaxAndSkipEvents(eventsInFile - 1, currentEvent)
eventsInFile = eventsPerJob
self.currentJob.addResourceEstimates(jobTime=jobTime,
memory=memoryRequirement,
disk=diskRequired)
if deterministicPileup:
self.currentJob.addBaggageParameter("skipPileupEvents", (self.nJobs - 1) * eventsPerJob)
eventsInFile -= eventsPerJob
currentEvent += eventsPerJob
totalJobs += 1
logging.debug("Job created for real input with %s", self.currentJob)
else:
# This assumes there's only one run which is the case for MC
lumis = runs[0].lumis
(firstLumi, lastLumi) = (min(lumis), max(lumis))
currentLumi = firstLumi
lumisPerJob = int(ceil(float(eventsPerJob) / eventsPerLumi))
while eventsInFile:
self.newJob(name=self.getJobName(length=totalJobs))
self.currentJob.addFile(f)
self.currentJob.addBaggageParameter("lheInputFiles", lheInput)
# Limit the number of events to a unsigned 32bit int
if (currentEvent + eventsPerJob - 1) > (2 ** 32 - 1) and \
(currentEvent + eventsInFile) > (2 ** 32 - 1):
currentEvent = 1
if eventsInFile >= eventsPerJob:
jobTime = eventsPerJob * timePerEvent
diskRequired = eventsPerJob * sizePerEvent
# Alan on 16/Apr/2019: inclusiveMask must be a real inclusiveMask, thus
# FirstEvent/FirstLumi and LastEvent/LastLumi are also processed by the job
self.currentJob["mask"].setMaxAndSkipEvents(eventsPerJob - 1, currentEvent)
self.currentJob["mask"].setMaxAndSkipLumis(lumisPerJob - 1, currentLumi)
else:
jobTime = eventsInFile * timePerEvent
diskRequired = eventsInFile * sizePerEvent
lumisPerJob = int(ceil(float(eventsInFile) / eventsPerLumi))
self.currentJob["mask"].setMaxAndSkipEvents(eventsInFile - 1, currentEvent)
self.currentJob["mask"].setMaxAndSkipLumis(lumisPerJob - 1, currentLumi)
eventsInFile = eventsPerJob
self.currentJob.addResourceEstimates(jobTime=jobTime,
memory=memoryRequirement,
disk=diskRequired)
if deterministicPileup:
self.currentJob.addBaggageParameter("skipPileupEvents", (self.nJobs - 1) * eventsPerJob)
eventsInFile -= eventsPerJob
currentEvent += eventsPerJob
currentLumi += lumisPerJob
totalJobs += 1
logging.info("Job created with mask: %s", self.currentJob['mask'])
return
class ErrorHandlerPoller(BaseWorkerThread):
"""
Polls for Error Conditions, handles them
"""
def __init__(self, config):
"""
Initialise class members
"""
BaseWorkerThread.__init__(self)
self.config = config
myThread = threading.currentThread()
self.daoFactory = DAOFactory(package="WMCore.WMBS",
logger=myThread.logger,
dbinterface=myThread.dbi)
self.changeState = ChangeState(self.config)
self.maxRetries = self.config.ErrorHandler.maxRetries
if not isinstance(self.maxRetries, dict):
self.maxRetries = {'default': self.maxRetries}
if 'default' not in self.maxRetries:
raise ErrorHandlerException(
'Max retries for the default job type must be specified')
self.exitCodesNoRetry = []
self.maxProcessSize = getattr(self.config.ErrorHandler,
'maxProcessSize', 250)
self.maxFailTime = getattr(self.config.ErrorHandler, 'maxFailTime',
32 * 3600)
self.readFWJR = getattr(self.config.ErrorHandler, 'readFWJR', False)
self.passCodes = getattr(self.config.ErrorHandler, 'passExitCodes', [])
self.getJobs = self.daoFactory(classname="Jobs.GetAllJobs")
self.idLoad = self.daoFactory(classname="Jobs.LoadFromIDWithType")
self.loadAction = self.daoFactory(classname="Jobs.LoadForErrorHandler")
self.dataCollection = DataCollectionService(
url=config.ACDC.couchurl, database=config.ACDC.database)
if hasattr(self.config, "Tier0Feeder"):
self.reqAuxDB = None
else:
self.reqAuxDB = ReqMgrAux(self.config.General.ReqMgr2ServiceURL)
return
def setup(self, parameters=None):
"""
Load DB objects required for queries
"""
# For now, does nothing
return
def terminate(self, params):
"""
_terminate_
Do one pass, then commit suicide
"""
logging.debug("terminating. doing one more pass before we die")
self.algorithm(params)
def exhaustJobs(self, jobList):
"""
_exhaustJobs_
Actually do the jobs exhaustion
"""
self.changeState.propagate(jobList, 'exhausted', 'retrydone')
# Remove all the files in the exhausted jobs.
logging.debug("About to fail input files for exhausted jobs")
for job in jobList:
job.failInputFiles()
# Do not build ACDC for utilitarian job types
jobList = [
job for job in jobList
if job['type'] not in ['LogCollect', 'Cleanup']
]
self.handleACDC(jobList)
return
def processRetries(self, jobList, state):
"""
_processRetries_
Actually do the retries
"""
logging.info("Processing retries for %d failed jobs of type %sfailed",
len(jobList), state)
retrydoneJobs = []
cooloffJobs = []
passJobs = []
# Retries < max retry count
for job in jobList:
allowedRetries = self.maxRetries.get(job['type'],
self.maxRetries['default'])
# Retries < allowed max retry count
if job['retry_count'] < allowedRetries and state != 'create':
cooloffJobs.append(job)
# Check if Retries >= allowed max retry count
elif job['retry_count'] >= allowedRetries or state == 'create':
retrydoneJobs.append(job)
msg = "Stopping retries for job %d" % job['id']
logging.debug(msg)
logging.debug("JobInfo: %s", job)
if self.readFWJR:
# Then we have to check each FWJR for exit status
cooloffJobs, passJobs, retrydoneFWJRJobs = self.readFWJRForErrors(
cooloffJobs)
retrydoneJobs.extend(retrydoneFWJRJobs)
# Now to actually do something.
logging.debug("About to propagate jobs")
if len(retrydoneJobs) > 0:
self.changeState.propagate(retrydoneJobs,
'retrydone',
'%sfailed' % state,
updatesummary=True)
if len(cooloffJobs) > 0:
self.changeState.propagate(cooloffJobs,
'%scooloff' % state,
'%sfailed' % state,
updatesummary=True)
if len(passJobs) > 0:
# Overwrite the transition states and move directly to created
self.changeState.propagate(passJobs, 'created', 'new')
return
def handleACDC(self, jobList):
"""
_handleACDC_
Do the ACDC creation and hope it works
"""
idList = [x['id'] for x in jobList]
logging.info("Starting to build ACDC with %i jobs", len(idList))
logging.info("This operation will take some time...")
loadList = self.loadJobsFromListFull(idList)
for job in loadList:
job.getMask()
self.dataCollection.failedJobs(loadList)
return
def readFWJRForErrors(self, jobList):
"""
_readFWJRForErrors_
Check the FWJRs of the failed jobs
and determine those that can be retried
and which must be retried without going through cooloff.
Returns a triplet with cooloff, passed and exhausted jobs.
"""
cooloffJobs = []
passJobs = []
exhaustJobs = []
if self.reqAuxDB:
self.exitCodesNoRetry = self.reqAuxDB.getWMAgentConfig(
self.config.Agent.hostName).get("NoRetryExitCodes", [])
for job in jobList:
report = Report()
reportPath = job['fwjr_path']
if reportPath is None:
logging.error(
"No FWJR in job %i, ErrorHandler can't process it.\n Passing it to cooloff.",
job['id'])
cooloffJobs.append(job)
continue
if not os.path.isfile(reportPath):
logging.error(
"Failed to find FWJR for job %i in location %s.\n Passing it to cooloff.",
job['id'], reportPath)
cooloffJobs.append(job)
continue
try:
report.load(reportPath)
# First let's check the time conditions
times = report.getFirstStartLastStop()
startTime = None
stopTime = None
if times is not None:
startTime = times['startTime']
stopTime = times['stopTime']
# correct the location if the original location is different from recorded in wmbs
# WARNING: we are not updating job location in wmbs only updating in couchdb by doing this.
# If location in wmbs needs to be updated, it should happen in JobAccountant.
locationFromFWJR = report.getSiteName()
if locationFromFWJR:
job["location"] = locationFromFWJR
job["site_cms_name"] = locationFromFWJR
if startTime is None or stopTime is None:
# We have no information to make a decision, keep going.
logging.debug("No start, stop times for steps for job %i",
job['id'])
elif stopTime - startTime > self.maxFailTime:
msg = "Job %i exhausted after running on node for %i seconds" % (
job['id'], stopTime - startTime)
logging.debug(msg)
exhaustJobs.append(job)
continue
if len([
x for x in report.getExitCodes()
if x in self.exitCodesNoRetry
]):
msg = "Job %i exhausted due to a bad exit code (%s)" % (
job['id'], str(report.getExitCodes()))
logging.error(msg)
exhaustJobs.append(job)
continue
if len(
[x for x in report.getExitCodes() if x in self.passCodes]):
msg = "Job %i restarted immediately due to an exit code (%s)" % (
job['id'], str(report.getExitCodes()))
logging.debug(msg)
passJobs.append(job)
continue
cooloffJobs.append(job)
except Exception as ex:
logging.warning(
"Exception while trying to check jobs for failures!")
logging.warning(str(ex))
logging.warning("Ignoring and sending job to cooloff")
cooloffJobs.append(job)
return cooloffJobs, passJobs, exhaustJobs
def handleRetryDoneJobs(self, jobList):
"""
_handleRetryDoneJobs_
"""
myThread = threading.currentThread()
logging.info("About to process %d retry done jobs", len(jobList))
myThread.transaction.begin()
self.exhaustJobs(jobList)
myThread.transaction.commit()
return
def handleFailedJobs(self, jobList, state):
"""
_handleFailedJobs_
"""
myThread = threading.currentThread()
logging.info("About to process %d failures", len(jobList))
myThread.transaction.begin()
self.processRetries(jobList, state)
myThread.transaction.commit()
return
def handleErrors(self):
"""
Queries DB for all watched filesets, if matching filesets become
available, create the subscriptions
"""
# Run over created, submitted and executed job failures
failure_states = ['create', 'submit', 'job']
for state in failure_states:
idList = self.getJobs.execute(state="%sfailed" % state)
logging.info("Found %d failed jobs in state %sfailed", len(idList),
state)
while len(idList) > 0:
tmpList = idList[:self.maxProcessSize]
idList = idList[self.maxProcessSize:]
jobList = self.loadJobsFromList(tmpList)
self.handleFailedJobs(jobList, state)
# Run over jobs done with retries
idList = self.getJobs.execute(state='retrydone')
logging.info("Found %d jobs done with all retries", len(idList))
while len(idList) > 0:
tmpList = idList[:self.maxProcessSize]
idList = idList[self.maxProcessSize:]
jobList = self.loadJobsFromList(tmpList)
self.handleRetryDoneJobs(jobList)
return
def loadJobsFromList(self, idList):
"""
_loadJobsFromList_
Load jobs in bulk
"""
binds = []
for jobID in idList:
binds.append({"jobid": jobID})
results = self.idLoad.execute(jobID=binds)
# You have to have a list
if isinstance(results, dict):
results = [results]
listOfJobs = []
for entry in results:
# One job per entry
tmpJob = Job(id=entry['id'])
tmpJob.update(entry)
listOfJobs.append(tmpJob)
return listOfJobs
def loadJobsFromListFull(self, idList):
"""
_loadJobsFromList_
Load jobs in bulk.
Include the full metadata.
"""
binds = []
for jobID in idList:
binds.append({"jobid": jobID})
results = self.loadAction.execute(jobID=binds)
# You have to have a list
if isinstance(results, dict):
results = [results]
listOfJobs = []
for entry in results:
# One job per entry
tmpJob = Job(id=entry['id'])
tmpJob.update(entry)
listOfJobs.append(tmpJob)
return listOfJobs
@timeFunction
def algorithm(self, parameters=None):
"""
Performs the handleErrors method, looking for each type of failure
And deal with it as desired.
"""
logging.debug("Running error handling algorithm")
try:
myThread = threading.currentThread()
self.handleErrors()
except (CouchConnectionError, HTTPException) as ex:
if getattr(myThread, 'transaction', None) is not None:
myThread.transaction.rollback()
msg = "Caught CouchConnectionError/HTTPException exception in ErrorHandler. "
msg += "Transactions postponed until the next polling cycle\n"
msg += str(ex)
logging.error(msg)
except Exception as ex:
if getattr(myThread, 'transaction', None) is not None:
myThread.transaction.rollback()
msg = "Caught unexpected exception in ErrorHandler:\n"
msg += str(ex)
logging.exception(msg)
raise ErrorHandlerException(msg)
class AccountantWorker(WMConnectionBase):
"""
Class that actually does the work of parsing FWJRs for the Accountant
Run through ProcessPool
"""
def __init__(self, config):
"""
__init__
Create all DAO objects that are used by this class.
"""
WMConnectionBase.__init__(self, "WMCore.WMBS")
myThread = threading.currentThread()
self.dbsDaoFactory = DAOFactory(package = "WMComponent.DBS3Buffer",
logger = myThread.logger,
dbinterface = myThread.dbi)
self.getOutputMapAction = self.daofactory(classname = "Jobs.GetOutputMap")
self.bulkAddToFilesetAction = self.daofactory(classname = "Fileset.BulkAddByLFN")
self.bulkParentageAction = self.daofactory(classname = "Files.AddBulkParentage")
self.getJobTypeAction = self.daofactory(classname = "Jobs.GetType")
self.getParentInfoAction = self.daofactory(classname = "Files.GetParentInfo")
self.setParentageByJob = self.daofactory(classname = "Files.SetParentageByJob")
self.setParentageByMergeJob = self.daofactory(classname = "Files.SetParentageByMergeJob")
self.setFileRunLumi = self.daofactory(classname = "Files.AddRunLumi")
self.setFileLocation = self.daofactory(classname = "Files.SetLocationByLFN")
self.setFileAddChecksum = self.daofactory(classname = "Files.AddChecksumByLFN")
self.addFileAction = self.daofactory(classname = "Files.Add")
self.jobCompleteInput = self.daofactory(classname = "Jobs.CompleteInput")
self.setBulkOutcome = self.daofactory(classname = "Jobs.SetOutcomeBulk")
self.getWorkflowSpec = self.daofactory(classname = "Workflow.GetSpecAndNameFromTask")
self.getJobInfoByID = self.daofactory(classname = "Jobs.LoadFromID")
self.getFullJobInfo = self.daofactory(classname = "Jobs.LoadForErrorHandler")
self.getJobTaskNameAction = self.daofactory(classname = "Jobs.GetFWJRTaskName")
self.dbsStatusAction = self.dbsDaoFactory(classname = "DBSBufferFiles.SetStatus")
self.dbsParentStatusAction = self.dbsDaoFactory(classname = "DBSBufferFiles.GetParentStatus")
self.dbsChildrenAction = self.dbsDaoFactory(classname = "DBSBufferFiles.GetChildren")
self.dbsCreateFiles = self.dbsDaoFactory(classname = "DBSBufferFiles.Add")
self.dbsSetLocation = self.dbsDaoFactory(classname = "DBSBufferFiles.SetLocationByLFN")
self.dbsInsertLocation = self.dbsDaoFactory(classname = "DBSBufferFiles.AddLocation")
self.dbsSetChecksum = self.dbsDaoFactory(classname = "DBSBufferFiles.AddChecksumByLFN")
self.dbsSetRunLumi = self.dbsDaoFactory(classname = "DBSBufferFiles.AddRunLumi")
self.dbsGetWorkflow = self.dbsDaoFactory(classname = "ListWorkflow")
self.dbsLFNHeritage = self.dbsDaoFactory(classname = "DBSBufferFiles.BulkHeritageParent")
self.stateChanger = ChangeState(config)
# Decide whether or not to attach jobReport to returned value
self.returnJobReport = getattr(config.JobAccountant, 'returnReportFromWorker', False)
# Store location for the specs for DBS
self.specDir = getattr(config.JobAccountant, 'specDir', None)
# ACDC service
self.dataCollection = DataCollectionService(url = config.ACDC.couchurl,
database = config.ACDC.database)
jobDBurl = sanitizeURL(config.JobStateMachine.couchurl)['url']
jobDBName = config.JobStateMachine.couchDBName
jobCouchdb = CouchServer(jobDBurl)
self.fwjrCouchDB = jobCouchdb.connectDatabase("%s/fwjrs" % jobDBName)
self.localWMStats = WMStatsWriter(config.TaskArchiver.localWMStatsURL, appName="WMStatsAgent")
# Hold data for later commital
self.dbsFilesToCreate = []
self.wmbsFilesToBuild = []
self.wmbsMergeFilesToBuild = []
self.fileLocation = None
self.mergedOutputFiles = []
self.listOfJobsToSave = []
self.listOfJobsToFail = []
self.filesetAssoc = []
self.parentageBinds = []
self.parentageBindsForMerge = []
self.jobsWithSkippedFiles = {}
self.count = 0
self.datasetAlgoID = collections.deque(maxlen = 1000)
self.datasetAlgoPaths = collections.deque(maxlen = 1000)
self.dbsLocations = set()
self.workflowIDs = collections.deque(maxlen = 1000)
self.workflowPaths = collections.deque(maxlen = 1000)
self.phedex = PhEDEx()
self.locLists = self.phedex.getNodeMap()
return
def reset(self):
"""
_reset_
Reset all global vars between runs.
"""
self.dbsFilesToCreate = []
self.wmbsFilesToBuild = []
self.wmbsMergeFilesToBuild = []
self.fileLocation = None
self.mergedOutputFiles = []
self.listOfJobsToSave = []
self.listOfJobsToFail = []
self.filesetAssoc = []
self.parentageBinds = []
self.parentageBindsForMerge = []
self.jobsWithSkippedFiles = {}
gc.collect()
return
def loadJobReport(self, parameters):
"""
_loadJobReport_
Given a framework job report on disk, load it and return a
FwkJobReport instance. If there is any problem loading or parsing the
framework job report return None.
"""
# The jobReportPath may be prefixed with "file://" which needs to be
# removed so it doesn't confuse the FwkJobReport() parser.
jobReportPath = parameters.get("fwjr_path", None)
if not jobReportPath:
logging.error("Bad FwkJobReport Path: %s" % jobReportPath)
return self.createMissingFWKJR(parameters, 99999, "FWJR path is empty")
jobReportPath = jobReportPath.replace("file://","")
if not os.path.exists(jobReportPath):
logging.error("Bad FwkJobReport Path: %s" % jobReportPath)
return self.createMissingFWKJR(parameters, 99999, 'Cannot find file in jobReport path: %s' % jobReportPath)
if os.path.getsize(jobReportPath) == 0:
logging.error("Empty FwkJobReport: %s" % jobReportPath)
return self.createMissingFWKJR(parameters, 99998, 'jobReport of size 0: %s ' % jobReportPath)
jobReport = Report()
try:
jobReport.load(jobReportPath)
except Exception as ex:
msg = "Error loading jobReport %s\n" % jobReportPath
msg += str(ex)
logging.error(msg)
logging.debug("Failing job: %s\n" % parameters)
return self.createMissingFWKJR(parameters, 99997, 'Cannot load jobReport')
if len(jobReport.listSteps()) == 0:
logging.error("FwkJobReport with no steps: %s" % jobReportPath)
return self.createMissingFWKJR(parameters, 99997, 'jobReport with no steps: %s ' % jobReportPath)
return jobReport
def isTaskExistInFWJR(self, jobReport, jobStatus):
"""
If taskName is not available in the FWJR, then tries to
recover it getting data from the SQL database.
"""
if not jobReport.getTaskName():
logging.warning("Trying to recover a corrupted FWJR for a %s job with job id %s" % (jobStatus,
jobReport.getJobID()))
jobInfo = self.getJobTaskNameAction.execute(jobId = jobReport.getJobID(),
conn = self.getDBConn(),
transaction = self.existingTransaction())
jobReport.setTaskName(jobInfo['taskName'])
jobReport.save(jobInfo['fwjr_path'])
if not jobReport.getTaskName():
msg = "Report to developers. Failed to recover corrupted fwjr for %s job id %s" % (jobStatus,
jobReport.getJobID())
raise AccountantWorkerException(msg)
else:
logging.info("TaskName '%s' successfully recovered and added to fwjr id %s." % (jobReport.getTaskName(),
jobReport.getJobID()))
return
def __call__(self, parameters):
"""
__call__
Handle a completed job. The parameters dictionary will contain the job
ID and the path to the framework job report.
"""
returnList = []
self.reset()
for job in parameters:
logging.info("Handling %s" % job["fwjr_path"])
# Load the job and set the ID
fwkJobReport = self.loadJobReport(job)
fwkJobReport.setJobID(job['id'])
jobSuccess = self.handleJob(jobID = job["id"],
fwkJobReport = fwkJobReport)
if self.returnJobReport:
returnList.append({'id': job["id"], 'jobSuccess': jobSuccess,
'jobReport': fwkJobReport})
else:
returnList.append({'id': job["id"], 'jobSuccess': jobSuccess})
self.count += 1
self.beginTransaction()
# Now things done at the end of the job
# Do what we can with WMBS files
self.handleWMBSFiles(self.wmbsFilesToBuild, self.parentageBinds)
# handle merge files separately since parentage need to set
# separately to support robust merge
self.handleWMBSFiles(self.wmbsMergeFilesToBuild, self.parentageBindsForMerge)
# Create DBSBufferFiles
self.createFilesInDBSBuffer()
# Handle filesetAssoc
if len(self.filesetAssoc) > 0:
self.bulkAddToFilesetAction.execute(binds = self.filesetAssoc,
conn = self.getDBConn(),
transaction = self.existingTransaction())
# Move successful jobs to successful
if len(self.listOfJobsToSave) > 0:
idList = [x['id'] for x in self.listOfJobsToSave]
outcomeBinds = [{'jobid': x['id'], 'outcome': x['outcome']} for x in self.listOfJobsToSave]
self.setBulkOutcome.execute(binds = outcomeBinds,
conn = self.getDBConn(),
transaction = self.existingTransaction())
self.jobCompleteInput.execute(id = idList,
lfnsToSkip = self.jobsWithSkippedFiles,
conn = self.getDBConn(),
transaction = self.existingTransaction())
self.stateChanger.propagate(self.listOfJobsToSave, "success", "complete")
# If we have failed jobs, fail them
if len(self.listOfJobsToFail) > 0:
outcomeBinds = [{'jobid': x['id'], 'outcome': x['outcome']} for x in self.listOfJobsToFail]
self.setBulkOutcome.execute(binds = outcomeBinds,
conn = self.getDBConn(),
transaction = self.existingTransaction())
self.stateChanger.propagate(self.listOfJobsToFail, "jobfailed", "complete")
# Arrange WMBS parentage
if len(self.parentageBinds) > 0:
self.setParentageByJob.execute(binds = self.parentageBinds,
conn = self.getDBConn(),
transaction = self.existingTransaction())
if len(self.parentageBindsForMerge) > 0:
self.setParentageByMergeJob.execute(binds = self.parentageBindsForMerge,
conn = self.getDBConn(),
transaction = self.existingTransaction())
# Straighten out DBS Parentage
if len(self.mergedOutputFiles) > 0:
self.handleDBSBufferParentage()
if len(self.jobsWithSkippedFiles) > 0:
self.handleSkippedFiles()
self.commitTransaction(existingTransaction = False)
return returnList
def outputFilesetsForJob(self, outputMap, merged, moduleLabel):
"""
_outputFilesetsForJob_
Determine if the file should be placed in any other fileset. Note that
this will not return the JobGroup output fileset as all jobs will have
their output placed there.
"""
if moduleLabel not in outputMap:
logging.info("Output module label missing from output map.")
return []
outputFilesets = []
for outputFileset in outputMap[moduleLabel]:
if merged == False and outputFileset["output_fileset"] != None:
outputFilesets.append(outputFileset["output_fileset"])
else:
if outputFileset["merged_output_fileset"] != None:
outputFilesets.append(outputFileset["merged_output_fileset"])
return outputFilesets
def addFileToDBS(self, jobReportFile, task):
"""
_addFileToDBS_
Add a file that was output from a job to the DBS buffer.
"""
datasetInfo = jobReportFile["dataset"]
dbsFile = DBSBufferFile(lfn = jobReportFile["lfn"],
size = jobReportFile["size"],
events = jobReportFile["events"],
checksums = jobReportFile["checksums"],
status = "NOTUPLOADED")
dbsFile.setAlgorithm(appName = datasetInfo["applicationName"],
appVer = datasetInfo["applicationVersion"],
appFam = jobReportFile["module_label"],
psetHash = "GIBBERISH",
configContent = jobReportFile.get('configURL'))
dbsFile.setDatasetPath("/%s/%s/%s" % (datasetInfo["primaryDataset"],
datasetInfo["processedDataset"],
datasetInfo["dataTier"]))
dbsFile.setValidStatus(validStatus = jobReportFile.get("validStatus", None))
dbsFile.setProcessingVer(ver = jobReportFile.get('processingVer', None))
dbsFile.setAcquisitionEra(era = jobReportFile.get('acquisitionEra', None))
dbsFile.setGlobalTag(globalTag = jobReportFile.get('globalTag', None))
#TODO need to find where to get the prep id
dbsFile.setPrepID(prep_id = jobReportFile.get('prep_id', None))
dbsFile['task'] = task
for run in jobReportFile["runs"]:
newRun = Run(runNumber = run.run)
newRun.extend(run.lumis)
dbsFile.addRun(newRun)
dbsFile.setLocation(pnn = list(jobReportFile["locations"])[0], immediateSave = False)
self.dbsFilesToCreate.append(dbsFile)
return
def findDBSParents(self, lfn):
"""
_findDBSParents_
Find the parent of the file in DBS
This is meant to be called recursively
"""
parentsInfo = self.getParentInfoAction.execute([lfn],
conn = self.getDBConn(),
transaction = self.existingTransaction())
newParents = set()
for parentInfo in parentsInfo:
# This will catch straight to merge files that do not have redneck
# parents. We will mark the straight to merge file from the job
# as a child of the merged parent.
if int(parentInfo["merged"]) == 1:
newParents.add(parentInfo["lfn"])
elif parentInfo['gpmerged'] == None:
continue
# Handle the files that result from merge jobs that aren't redneck
# children. We have to setup parentage and then check on whether or
# not this file has any redneck children and update their parentage
# information.
elif int(parentInfo["gpmerged"]) == 1:
newParents.add(parentInfo["gplfn"])
# If that didn't work, we've reached the great-grandparents
# And we have to work via recursion
else:
parentSet = self.findDBSParents(lfn = parentInfo['gplfn'])
for parent in parentSet:
newParents.add(parent)
return newParents
def addFileToWMBS(self, jobType, fwjrFile, jobMask, task, jobID = None):
"""
_addFileToWMBS_
Add a file that was produced in a job to WMBS.
"""
fwjrFile["first_event"] = jobMask["FirstEvent"]
if fwjrFile["first_event"] == None:
fwjrFile["first_event"] = 0
if jobType == "Merge" and fwjrFile["module_label"] != "logArchive":
setattr(fwjrFile["fileRef"], 'merged', True)
fwjrFile["merged"] = True
wmbsFile = self.createFileFromDataStructsFile(file = fwjrFile, jobID = jobID)
if jobType == "Merge":
self.wmbsMergeFilesToBuild.append(wmbsFile)
else:
self.wmbsFilesToBuild.append(wmbsFile)
if fwjrFile["merged"]:
self.addFileToDBS(fwjrFile, task)
return wmbsFile
def _mapLocation(self, fwkJobReport):
for file in fwkJobReport.getAllFileRefs():
if file and hasattr(file, 'location'):
file.location = self.phedex.getBestNodeName(file.location, self.locLists)
def handleJob(self, jobID, fwkJobReport):
"""
_handleJob_
Figure out if a job was successful or not, handle it appropriately
(parse FWJR, update WMBS) and return the success status as a boolean
"""
jobSuccess = fwkJobReport.taskSuccessful()
outputMap = self.getOutputMapAction.execute(jobID = jobID,
conn = self.getDBConn(),
transaction = self.existingTransaction())
jobType = self.getJobTypeAction.execute(jobID = jobID,
conn = self.getDBConn(),
transaction = self.existingTransaction())
if jobSuccess:
fileList = fwkJobReport.getAllFiles()
# consistency check comparing outputMap to fileList
# they should match except for some limited special cases
outputModules = set([])
for fwjrFile in fileList:
outputModules.add(fwjrFile['outputModule'])
if set(outputMap.keys()) == outputModules:
pass
elif jobType == "LogCollect" and len(outputMap.keys()) == 0 and outputModules == set(['LogCollect']):
pass
elif jobType == "Merge" and set(outputMap.keys()) == set(['Merged', 'MergedError', 'logArchive']) and outputModules == set(['Merged', 'logArchive']):
pass
elif jobType == "Merge" and set(outputMap.keys()) == set(['Merged', 'MergedError', 'logArchive']) and outputModules == set(['MergedError', 'logArchive']):
pass
elif jobType == "Express" and set(outputMap.keys()).difference(outputModules) == set(['write_RAW']):
pass
else:
failJob = True
if jobType in [ "Processing", "Production" ]:
cmsRunSteps = 0
for step in fwkJobReport.listSteps():
if step.startswith("cmsRun"):
cmsRunSteps += 1
if cmsRunSteps > 1:
failJob = False
if failJob:
jobSuccess = False
logging.error("Job %d , list of expected outputModules does not match job report, failing job", jobID)
logging.debug("Job %d , expected outputModules %s", jobID, sorted(outputMap.keys()))
logging.debug("Job %d , fwjr outputModules %s", jobID, sorted(outputModules))
fileList = fwkJobReport.getAllFilesFromStep(step = 'logArch1')
else:
logging.debug("Job %d , list of expected outputModules does not match job report, accepted for multi-step CMSSW job", jobID)
else:
fileList = fwkJobReport.getAllFilesFromStep(step = 'logArch1')
if jobSuccess:
logging.info("Job %d , handle successful job", jobID)
else:
logging.error("Job %d , bad jobReport, failing job", jobID)
# make sure the task name is present in FWJR (recover from WMBS if needed)
if len(fileList) > 0:
if jobSuccess:
self.isTaskExistInFWJR(fwkJobReport, "success")
else:
self.isTaskExistInFWJR(fwkJobReport, "failed")
# special check for LogCollect jobs
skipLogCollect = False
if jobSuccess and jobType == "LogCollect":
for fwjrFile in fileList:
try:
# this assumes there is only one file for LogCollect jobs, not sure what happend if that changes
self.associateLogCollectToParentJobsInWMStats(fwkJobReport, fwjrFile["lfn"], fwkJobReport.getTaskName())
except Exception as ex:
skipLogCollect = True
logging.error("Error occurred: associating log collect location, will try again\n %s" % str(ex))
break
# now handle the job (unless the special LogCollect check failed)
if not skipLogCollect:
wmbsJob = Job(id = jobID)
wmbsJob.load()
outputID = wmbsJob.loadOutputID()
wmbsJob.getMask()
wmbsJob["fwjr"] = fwkJobReport
if jobSuccess:
wmbsJob["outcome"] = "success"
else:
wmbsJob["outcome"] = "failure"
for fwjrFile in fileList:
logging.debug("Job %d , register output %s", jobID, fwjrFile["lfn"])
wmbsFile = self.addFileToWMBS(jobType, fwjrFile, wmbsJob["mask"],
jobID = jobID, task = fwkJobReport.getTaskName())
merged = fwjrFile['merged']
moduleLabel = fwjrFile["module_label"]
if merged:
self.mergedOutputFiles.append(wmbsFile)
self.filesetAssoc.append({"lfn": wmbsFile["lfn"], "fileset": outputID})
# LogCollect jobs have no output fileset
if jobType != "LogCollect":
outputFilesets = self.outputFilesetsForJob(outputMap, merged, moduleLabel)
for outputFileset in outputFilesets:
self.filesetAssoc.append({"lfn": wmbsFile["lfn"], "fileset": outputFileset})
# Check if the job had any skipped files, put them in ACDC containers
# We assume full file processing (no job masks)
if jobSuccess:
skippedFiles = fwkJobReport.getAllSkippedFiles()
if skippedFiles:
self.jobsWithSkippedFiles[jobID] = skippedFiles
# Only save once job is done, and we're sure we made it through okay
self._mapLocation(wmbsJob['fwjr'])
if jobSuccess:
self.listOfJobsToSave.append(wmbsJob)
else:
self.listOfJobsToFail.append(wmbsJob)
return jobSuccess
def associateLogCollectToParentJobsInWMStats(self, fwkJobReport, logAchiveLFN, task):
"""
_associateLogCollectToParentJobsInWMStats_
Associate a logArchive output to its parent job
"""
inputFileList = fwkJobReport.getAllInputFiles()
requestName = task.split('/')[1]
keys = []
for inputFile in inputFileList:
keys.append([requestName, inputFile["lfn"]])
resultRows = self.fwjrCouchDB.loadView("FWJRDump", 'jobsByOutputLFN',
options = {"stale": "update_after"},
keys = keys)['rows']
if len(resultRows) > 0:
#get data from wmbs
parentWMBSJobIDs = []
for row in resultRows:
parentWMBSJobIDs.append({"jobid": row["value"]})
#update Job doc in wmstats
results = self.getJobInfoByID.execute(parentWMBSJobIDs)
parentJobNames = []
if isinstance(results, list):
for jobInfo in results:
parentJobNames.append(jobInfo['name'])
else:
parentJobNames.append(results['name'])
self.localWMStats.updateLogArchiveLFN(parentJobNames, logAchiveLFN)
else:
#TODO: if the couch db is consistent with DB this should be removed (checking resultRow > 0)
#It need to be failed and retried.
logging.error("job report is missing for updating log archive mapping\n Input file list\n %s" % inputFileList)
return
def createMissingFWKJR(self, parameters, errorCode = 999,
errorDescription = 'Failure of unknown type'):
"""
_createMissingFWJR_
Create a missing FWJR if the report can't be found by the code in the
path location.
"""
report = Report()
report.addError("cmsRun1", 84, errorCode, errorDescription)
report.data.cmsRun1.status = "Failed"
return report
def createFilesInDBSBuffer(self):
"""
_createFilesInDBSBuffer_
It does the actual job of creating things in DBSBuffer
WARNING: This assumes all files in a job have the same final location
"""
if len(self.dbsFilesToCreate) == 0:
# Whoops, nothing to do!
return
dbsFileTuples = []
dbsFileLoc = []
dbsCksumBinds = []
runLumiBinds = []
selfChecksums = None
jobLocations = set()
for dbsFile in self.dbsFilesToCreate:
# Append a tuple in the format specified by DBSBufferFiles.Add
# Also run insertDatasetAlgo
assocID = None
datasetAlgoPath = '%s:%s:%s:%s:%s:%s:%s:%s' % (dbsFile['datasetPath'],
dbsFile["appName"],
dbsFile["appVer"],
dbsFile["appFam"],
dbsFile["psetHash"],
dbsFile['processingVer'],
dbsFile['acquisitionEra'],
dbsFile['globalTag'])
# First, check if this is in the cache
if datasetAlgoPath in self.datasetAlgoPaths:
for da in self.datasetAlgoID:
if da['datasetAlgoPath'] == datasetAlgoPath:
assocID = da['assocID']
break
if not assocID:
# Then we have to get it ourselves
try:
assocID = dbsFile.insertDatasetAlgo()
self.datasetAlgoPaths.append(datasetAlgoPath)
self.datasetAlgoID.append({'datasetAlgoPath': datasetAlgoPath,
'assocID': assocID})
except WMException:
raise
except Exception as ex:
msg = "Unhandled exception while inserting datasetAlgo: %s\n" % datasetAlgoPath
msg += str(ex)
logging.error(msg)
raise AccountantWorkerException(msg)
# Associate the workflow to the file using the taskPath and the requestName
# TODO: debug why it happens and then drop/recover these cases automatically
taskPath = dbsFile.get('task')
if not taskPath:
msg = "Can't do workflow association, report this error to a developer.\n"
msg += "DbsFile : %s" % str(dbsFile)
raise AccountantWorkerException(msg)
workflowName = taskPath.split('/')[1]
workflowPath = '%s:%s' % (workflowName, taskPath)
if workflowPath in self.workflowPaths:
for wf in self.workflowIDs:
if wf['workflowPath'] == workflowPath:
workflowID = wf['workflowID']
break
else:
result = self.dbsGetWorkflow.execute(workflowName, taskPath, conn = self.getDBConn(),
transaction = self.existingTransaction())
workflowID = result['id']
self.workflowPaths.append(workflowPath)
self.workflowIDs.append({'workflowPath': workflowPath, 'workflowID': workflowID})
lfn = dbsFile['lfn']
selfChecksums = dbsFile['checksums']
jobLocation = dbsFile.getLocations()[0]
jobLocations.add(jobLocation)
dbsFileTuples.append((lfn, dbsFile['size'],
dbsFile['events'], assocID,
dbsFile['status'], workflowID))
dbsFileLoc.append({'lfn': lfn, 'sename' : jobLocation})
if dbsFile['runs']:
runLumiBinds.append({'lfn': lfn, 'runs': dbsFile['runs']})
if selfChecksums:
# If we have checksums we have to create a bind
# For each different checksum
for entry in selfChecksums.keys():
dbsCksumBinds.append({'lfn': lfn, 'cksum' : selfChecksums[entry],
'cktype' : entry})
try:
diffLocation = jobLocations.difference(self.dbsLocations)
for jobLocation in diffLocation:
self.dbsInsertLocation.execute(siteName = jobLocation,
conn = self.getDBConn(),
transaction = self.existingTransaction())
self.dbsLocations.add(jobLocation)
self.dbsCreateFiles.execute(files = dbsFileTuples,
conn = self.getDBConn(),
transaction = self.existingTransaction())
self.dbsSetLocation.execute(binds = dbsFileLoc,
conn = self.getDBConn(),
transaction = self.existingTransaction())
self.dbsSetChecksum.execute(bulkList = dbsCksumBinds,
conn = self.getDBConn(),
transaction = self.existingTransaction())
if len(runLumiBinds) > 0:
self.dbsSetRunLumi.execute(file = runLumiBinds,
conn = self.getDBConn(),
transaction = self.existingTransaction())
except WMException:
raise
except Exception as ex:
msg = "Got exception while inserting files into DBSBuffer!\n"
msg += str(ex)
logging.error(msg)
logging.debug("Listing binds:")
logging.debug("jobLocation: %s\n" % jobLocation)
logging.debug("dbsFiles: %s\n" % dbsFileTuples)
logging.debug("dbsFileLoc: %s\n" %dbsFileLoc)
logging.debug("Checksum binds: %s\n" % dbsCksumBinds)
logging.debug("RunLumi binds: %s\n" % runLumiBinds)
raise AccountantWorkerException(msg)
# Now that we've created those files, clear the list
self.dbsFilesToCreate = []
return
def handleWMBSFiles(self, wmbsFilesToBuild, parentageBinds):
"""
_handleWMBSFiles_
Do what can be done in bulk in bulk
"""
if len(wmbsFilesToBuild) == 0:
# Nothing to do
return
runLumiBinds = []
fileCksumBinds = []
fileLocations = []
fileCreate = []
for wmbsFile in wmbsFilesToBuild:
lfn = wmbsFile['lfn']
if lfn == None:
continue
selfChecksums = wmbsFile['checksums']
# by jobType add to different parentage relation
# if it is the merge job, don't include the parentage on failed input files.
# otherwise parentage is set for all input files.
parentageBinds.append({'child': lfn, 'jobid': wmbsFile['jid']})
if wmbsFile['runs']:
runLumiBinds.append({'lfn': lfn, 'runs': wmbsFile['runs']})
if len(wmbsFile.getLocations()) > 0:
fileLocations.append({'lfn': lfn, 'location': wmbsFile.getLocations()[0]})
if selfChecksums:
# If we have checksums we have to create a bind
# For each different checksum
for entry in selfChecksums.keys():
fileCksumBinds.append({'lfn': lfn, 'cksum' : selfChecksums[entry],
'cktype' : entry})
fileCreate.append([lfn,
wmbsFile['size'],
wmbsFile['events'],
None,
wmbsFile["first_event"],
wmbsFile['merged']])
if len(fileCreate) == 0:
return
try:
self.addFileAction.execute(files = fileCreate,
conn = self.getDBConn(),
transaction = self.existingTransaction())
if runLumiBinds:
self.setFileRunLumi.execute(file = runLumiBinds,
conn = self.getDBConn(),
transaction = self.existingTransaction())
self.setFileAddChecksum.execute(bulkList = fileCksumBinds,
conn = self.getDBConn(),
transaction = self.existingTransaction())
self.setFileLocation.execute(lfn = fileLocations,
location = self.fileLocation,
conn = self.getDBConn(),
transaction = self.existingTransaction())
except WMException:
raise
except Exception as ex:
msg = "Error while adding files to WMBS!\n"
msg += str(ex)
logging.error(msg)
logging.debug("Printing binds: \n")
logging.debug("FileCreate binds: %s\n" % fileCreate)
logging.debug("Runlumi binds: %s\n" % runLumiBinds)
logging.debug("Checksum binds: %s\n" % fileCksumBinds)
logging.debug("FileLocation binds: %s\n" % fileLocations)
raise AccountantWorkerException(msg)
# Clear out finished files
wmbsFilesToBuild = []
return
def createFileFromDataStructsFile(self, file, jobID):
"""
_createFileFromDataStructsFile_
This function will create a WMBS File given a DataStructs file
"""
wmbsFile = File()
wmbsFile.update(file)
if isinstance(file["locations"], set):
pnn = list(file["locations"])[0]
elif isinstance(file["locations"], list):
if len(file['locations']) > 1:
logging.error("Have more then one location for a file in job %i" % (jobID))
logging.error("Choosing location %s" % (file['locations'][0]))
pnn = file["locations"][0]
else:
pnn = file["locations"]
wmbsFile["locations"] = set()
if pnn != None:
wmbsFile.setLocation(pnn = pnn, immediateSave = False)
wmbsFile['jid'] = jobID
return wmbsFile
def handleDBSBufferParentage(self):
"""
_handleDBSBufferParentage_
Handle all the DBSBuffer Parentage in bulk if you can
"""
outputLFNs = [f['lfn'] for f in self.mergedOutputFiles]
bindList = []
for lfn in outputLFNs:
newParents = self.findDBSParents(lfn = lfn)
for parentLFN in newParents:
bindList.append({'child': lfn, 'parent': parentLFN})
# Now all the parents should exist
# Commit them to DBSBuffer
logging.info("About to commit all DBSBuffer Heritage information")
logging.info(len(bindList))
if len(bindList) > 0:
try:
self.dbsLFNHeritage.execute(binds = bindList,
conn = self.getDBConn(),
transaction = self.existingTransaction())
except WMException:
raise
except Exception as ex:
msg = "Error while trying to handle the DBS LFN heritage\n"
msg += str(ex)
msg += "BindList: %s" % bindList
logging.error(msg)
raise AccountantWorkerException(msg)
return
def handleSkippedFiles(self):
"""
_handleSkippedFiles_
Handle all the skipped files in bulk,
the way it handles the skipped files
imposes an important restriction:
Skipped files should have been processed by a single job
in the task and no job mask exists in it.
This is suitable for jobs using ParentlessMergeBySize/FileBased/MinFileBased
splitting algorithms.
Here ACDC records and created and the file are moved
to wmbs_sub_files_failed from completed.
"""
jobList = self.getFullJobInfo.execute([{'jobid' : x} for x in self.jobsWithSkippedFiles.keys()],
fileSelection = self.jobsWithSkippedFiles,
conn = self.getDBConn(),
transaction = self.existingTransaction())
self.dataCollection.failedJobs(jobList, useMask = False)
return