def cleanTaskQueues():
tqDB = TaskQueueDB()
jobDB = JobDB()
logDB = JobLoggingDB()
result = tqDB.enableAllTaskQueues()
if not result['OK']:
return result
result = tqDB.findOrphanJobs()
if not result['OK']:
return result
for jid in result['Value']:
result = tqDB.deleteJob(jid)
if not result['OK']:
gLogger.error("Cannot delete from TQ job %s" % jid, result['Message'])
continue
result = jobDB.rescheduleJob(jid)
if not result['OK']:
gLogger.error("Cannot reschedule in JobDB job %s" % jid, result['Message'])
continue
result = logDB.addLoggingRecord(jid, JobStatus.RECEIVED, "", "", source="JobState")
if not result['OK']:
gLogger.error("Cannot add logging record in JobLoggingDB %s" % jid, result['Message'])
continue
return S_OK()
def initializeMatcherHandler(serviceInfo):
""" Matcher Service initialization
"""
global gJobDB
global gTaskQueueDB
global jlDB
global pilotAgentsDB
gJobDB = JobDB()
gTaskQueueDB = TaskQueueDB()
jlDB = JobLoggingDB()
pilotAgentsDB = PilotAgentsDB()
gMonitor.registerActivity('matchTime', "Job matching time", 'Matching',
"secs", gMonitor.OP_MEAN, 300)
gMonitor.registerActivity('matchesDone', "Job Match Request", 'Matching',
"matches", gMonitor.OP_RATE, 300)
gMonitor.registerActivity('matchesOK', "Matched jobs", 'Matching',
"matches", gMonitor.OP_RATE, 300)
gMonitor.registerActivity('numTQs', "Number of Task Queues", 'Matching',
"tqsk queues", gMonitor.OP_MEAN, 300)
gTaskQueueDB.recalculateTQSharesForAll()
gThreadScheduler.addPeriodicTask(120,
gTaskQueueDB.recalculateTQSharesForAll)
gThreadScheduler.addPeriodicTask(60, sendNumTaskQueues)
sendNumTaskQueues()
return S_OK()
def initializeMatcherHandler(serviceInfo):
""" Matcher Service initialization
"""
global jobDB
global jobLoggingDB
global taskQueueDB
jobDB = JobDB()
jobLoggingDB = JobLoggingDB()
taskQueueDB = TaskQueueDB()
gMonitor.registerActivity("matchTime", "Job matching time", "Matching", "secs", gMonitor.OP_MEAN, 300)
gMonitor.registerActivity(
"matchTaskQueues", "Task queues checked per job", "Matching", "task queues", gMonitor.OP_MEAN, 300
)
gMonitor.registerActivity("matchesDone", "Job Matches", "Matching", "matches", gMonitor.OP_MEAN, 300)
gMonitor.registerActivity("numTQs", "Number of Task Queues", "Matching", "tqsk queues", gMonitor.OP_MEAN, 300)
taskQueueDB.recalculateTQSharesForAll()
gThreadScheduler.addPeriodicTask(120, taskQueueDB.recalculateTQSharesForAll)
gThreadScheduler.addPeriodicTask(120, sendNumTaskQueues)
sendNumTaskQueues()
return S_OK()
def initializeMatcherHandler( serviceInfo ):
""" Matcher Service initialization
"""
global gJobDB
global gJobLoggingDB
global gTaskQueueDB
global gPilotAgentsDB
gJobDB = JobDB()
gJobLoggingDB = JobLoggingDB()
gTaskQueueDB = TaskQueueDB()
gPilotAgentsDB = PilotAgentsDB()
gMonitor.registerActivity( 'matchTime', "Job matching time",
'Matching', "secs" , gMonitor.OP_MEAN, 300 )
gMonitor.registerActivity( 'matchesDone', "Job Match Request",
'Matching', "matches" , gMonitor.OP_RATE, 300 )
gMonitor.registerActivity( 'matchesOK', "Matched jobs",
'Matching', "matches" , gMonitor.OP_RATE, 300 )
gMonitor.registerActivity( 'numTQs', "Number of Task Queues",
'Matching', "tqsk queues" , gMonitor.OP_MEAN, 300 )
gTaskQueueDB.recalculateTQSharesForAll()
gThreadScheduler.addPeriodicTask( 120, gTaskQueueDB.recalculateTQSharesForAll )
gThreadScheduler.addPeriodicTask( 60, sendNumTaskQueues )
sendNumTaskQueues()
return S_OK()
def test_matcher( self ):
# insert a proper DN to run the test
resourceDescription = {'OwnerGroup': 'prod', 'OwnerDN':'/C=ch/O=DIRAC/OU=DIRAC CI/CN=ciuser/[email protected]',
'DIRACVersion': 'pippo', 'ReleaseVersion':'blabla', 'VirtualOrganization':'LHCB',
'PilotInfoReportedFlag':'True', 'PilotBenchmark':'anotherPilot', 'LHCbPlatform':'CERTO',
'Site':'DIRAC.Jenkins.org', 'CPUTime' : 86400 }
matcher = RPCClient( 'WorkloadManagement/Matcher' )
JobStateUpdate = RPCClient( 'WorkloadManagement/JobStateUpdate' )
wmsClient = WMSClient()
job = helloWorldJob()
job.setDestination( 'DIRAC.Jenkins.org' )
job.setInputData( '/a/bbb' )
job.setType( 'User' )
jobDescription = createFile( job )
res = wmsClient.submitJob( job._toJDL( xmlFile = jobDescription ) )
self.assert_( res['OK'] )
jobID = res['Value']
res = JobStateUpdate.setJobStatus( jobID, 'Waiting', 'matching', 'source' )
self.assert_( res['OK'] )
tqDB = TaskQueueDB()
tqDefDict = {'OwnerDN': '/C=ch/O=DIRAC/OU=DIRAC CI/CN=ciuser/[email protected]',
'OwnerGroup':'prod', 'Setup':'JenkinsSetup', 'CPUTime':86400}
res = tqDB.insertJob( jobID, tqDefDict, 10 )
self.assert_( res['OK'] )
res = matcher.requestJob( resourceDescription )
print res
self.assert_( res['OK'] )
wmsClient.deleteJob( jobID )
def initialize(self):
""" Sets defaults
"""
self.am_setOption("PollingTime", 120)
self.jobDB = JobDB()
self.taskQueueDB = TaskQueueDB()
self.jobLoggingDB = JobLoggingDB()
# self.sandboxDB = SandboxDB( 'SandboxDB' )
agentTSTypes = self.am_getOption('ProductionTypes', [])
if agentTSTypes:
self.prod_types = agentTSTypes
else:
self.prod_types = Operations().getValue(
'Transformations/DataProcessing', ['MCSimulation', 'Merge'])
gLogger.info(
"Will exclude the following Production types from cleaning %s" %
(', '.join(self.prod_types)))
self.maxJobsAtOnce = self.am_getOption('MaxJobsAtOnce', 500)
self.jobByJob = self.am_getOption('JobByJob', False)
self.throttlingPeriod = self.am_getOption('ThrottlingPeriod', 0.)
self.removeStatusDelay['Done'] = self.am_getOption(
'RemoveStatusDelay/Done', 7)
self.removeStatusDelay['Killed'] = self.am_getOption(
'RemoveStatusDelay/Killed', 7)
self.removeStatusDelay['Failed'] = self.am_getOption(
'RemoveStatusDelay/Failed', 7)
self.removeStatusDelay['Any'] = self.am_getOption(
'RemoveStatusDelay/Any', -1)
return S_OK()
def test_matcher( self ):
# insert a proper DN to run the test
resourceDescription = {'OwnerGroup': 'prod', 'OwnerDN':'/C=ch/O=DIRAC/OU=DIRAC CI/CN=ciuser/[email protected]',
'DIRACVersion': 'pippo', 'ReleaseVersion':'blabla', 'VirtualOrganization':'LHCB',
'PilotInfoReportedFlag':'True', 'PilotBenchmark':'anotherPilot', 'LHCbPlatform':'CERTO',
'Site':'DIRAC.Jenkins.org', 'CPUTime' : 86400 }
matcher = RPCClient( 'WorkloadManagement/Matcher' )
JobStateUpdate = RPCClient( 'WorkloadManagement/JobStateUpdate' )
wmsClient = WMSClient()
job = helloWorldJob()
job.setDestination( 'DIRAC.Jenkins.org' )
job.setInputData( '/a/bbb' )
job.setType( 'User' )
jobDescription = createFile( job )
res = wmsClient.submitJob( job._toJDL( xmlFile = jobDescription ) )
self.assert_( res['OK'] )
jobID = res['Value']
res = JobStateUpdate.setJobStatus( jobID, 'Waiting', 'matching', 'source' )
self.assert_( res['OK'] )
tqDB = TaskQueueDB()
tqDefDict = {'OwnerDN': '/C=ch/O=DIRAC/OU=DIRAC CI/CN=ciuser/[email protected]',
'OwnerGroup':'prod', 'Setup':'dirac-JenkinsSetup', 'CPUTime':86400}
res = tqDB.insertJob( jobID, tqDefDict, 10 )
self.assert_( res['OK'] )
res = matcher.requestJob( resourceDescription )
print res
self.assert_( res['OK'] )
wmsClient.deleteJob( jobID )
def __init__(self,
pilotAgentsDB=None,
jobDB=None,
tqDB=None,
jlDB=None,
opsHelper=None):
""" c'tor
"""
if pilotAgentsDB:
self.pilotAgentsDB = pilotAgentsDB
else:
self.pilotAgentsDB = PilotAgentsDB()
if jobDB:
self.jobDB = jobDB
else:
self.jobDB = JobDB()
if tqDB:
self.tqDB = tqDB
else:
self.tqDB = TaskQueueDB()
if jlDB:
self.jlDB = jlDB
else:
self.jlDB = JobLoggingDB()
if opsHelper:
self.opsHelper = opsHelper
else:
self.opsHelper = Operations()
self.log = gLogger.getSubLogger("Matcher")
self.limiter = Limiter(jobDB=self.jobDB, opsHelper=self.opsHelper)
self.siteClient = SiteStatus()
def __init__( self, pilotAgentsDB = None, jobDB = None, tqDB = None, jlDB = None, opsHelper = None ):
""" c'tor
"""
if pilotAgentsDB:
self.pilotAgentsDB = pilotAgentsDB
else:
self.pilotAgentsDB = PilotAgentsDB()
if jobDB:
self.jobDB = jobDB
else:
self.jobDB = JobDB()
if tqDB:
self.tqDB = tqDB
else:
self.tqDB = TaskQueueDB()
if jlDB:
self.jlDB = jlDB
else:
self.jlDB = JobLoggingDB()
if opsHelper:
self.opsHelper = opsHelper
else:
self.opsHelper = Operations()
self.log = gLogger.getSubLogger( "Matcher" )
self.limiter = Limiter( jobDB = self.jobDB, opsHelper = self.opsHelper )
def initializeOptimizer(self):
"""Initialize specific parameters for TaskQueueAgent.
"""
self.waitingStatus = self.am_getOption('WaitingStatus', 'Waiting')
self.waitingMinorStatus = self.am_getOption('WaitingMinorStatus',
'Pilot Agent Submission')
try:
self.taskQueueDB = TaskQueueDB()
result = self.taskQueueDB.enableAllTaskQueues()
if not result['OK']:
raise Exception("Can't enable TaskQueues: %s" %
result['Message'])
except Exception, e:
self.log.exception()
return S_ERROR("Cannot initialize taskqueueDB: %s" % str(e))
def initializeJobMonitoringHandler(serviceInfo):
global jobDB, jobLoggingDB, taskQueueDB
jobDB = JobDB()
jobLoggingDB = JobLoggingDB()
taskQueueDB = TaskQueueDB()
return S_OK()
def checkDBAccess(cls):
# Init DB if there
if not JobState.__db.checked:
JobState.__db.jobDB = JobDB()
JobState.__db.logDB = JobLoggingDB()
JobState.__db.tqDB = TaskQueueDB()
JobState.__db.checked = True
def export_getPilots(cls, jobID):
""" Get pilot references and their states for :
- those pilots submitted for the TQ where job is sitting
- (or) the pilots executing/having executed the Job
"""
pilots = []
result = pilotDB.getPilotsForJobID(int(jobID))
if not result['OK']:
if result['Message'].find('not found') == -1:
return S_ERROR('Failed to get pilot: ' + result['Message'])
else:
pilots += result['Value']
if not pilots:
# Pilots were not found try to look in the Task Queue
taskQueueID = 0
result = TaskQueueDB().getTaskQueueForJob(int(jobID))
if result['OK'] and result['Value']:
taskQueueID = result['Value']
if taskQueueID:
result = pilotDB.getPilotsForTaskQueue(taskQueueID, limit=10)
if not result['OK']:
return S_ERROR('Failed to get pilot: ' + result['Message'])
pilots += result['Value']
if not pilots:
return S_ERROR('Failed to get pilot for Job %d' % int(jobID))
return pilotDB.getPilotInfo(pilotID=pilots)
def initializeJobMonitoringHandler(serviceInfo):
global gJobDB, gJobLoggingDB, gTaskQueueDB
gJobDB = JobDB()
gJobLoggingDB = JobLoggingDB()
gTaskQueueDB = TaskQueueDB()
return S_OK()
def initialize( self ):
""" Sets defaults
"""
self.am_setOption( "PollingTime", 120 )
self.jobDB = JobDB()
self.taskQueueDB = TaskQueueDB()
self.jobLoggingDB = JobLoggingDB()
# self.sandboxDB = SandboxDB( 'SandboxDB' )
agentTSTypes = self.am_getOption('ProductionTypes', [])
if agentTSTypes:
self.prod_types = agentTSTypes
else:
self.prod_types = Operations().getValue( 'Transformations/DataProcessing', ['MCSimulation', 'Merge'] )
gLogger.info( "Will exclude the following Production types from cleaning %s" % ( ', '.join( self.prod_types ) ) )
self.maxJobsAtOnce = self.am_getOption( 'MaxJobsAtOnce', 500 )
self.jobByJob = self.am_getOption( 'JobByJob', False )
self.throttlingPeriod = self.am_getOption('ThrottlingPeriod', 0.)
self.removeStatusDelay['Done'] = self.am_getOption( 'RemoveStatusDelay/Done', 7 )
self.removeStatusDelay['Killed'] = self.am_getOption( 'RemoveStatusDelay/Killed', 7 )
self.removeStatusDelay['Failed'] = self.am_getOption( 'RemoveStatusDelay/Failed', 7 )
self.removeStatusDelay['Any'] = self.am_getOption( 'RemoveStatusDelay/Any', -1 )
return S_OK()
def initializeJobManagerHandler(serviceInfo):
global gJobDB, gJobLoggingDB, gtaskQueueDB
gJobDB = JobDB()
gJobLoggingDB = JobLoggingDB()
gtaskQueueDB = TaskQueueDB()
return S_OK()
class TaskQueueAgent(OptimizerModule):
"""
The specific Optimizer must provide the following methods:
- initializeOptimizer() before each execution cycle
- checkJob() - the main method called for each job
"""
#############################################################################
def initializeOptimizer(self):
"""Initialize specific parameters for TaskQueueAgent.
"""
self.waitingStatus = self.am_getOption('WaitingStatus', 'Waiting')
self.waitingMinorStatus = self.am_getOption('WaitingMinorStatus',
'Pilot Agent Submission')
try:
self.taskQueueDB = TaskQueueDB()
result = self.taskQueueDB.enableAllTaskQueues()
if not result['OK']:
raise Exception("Can't enable TaskQueues: %s" %
result['Message'])
except Exception, e:
self.log.exception()
return S_ERROR("Cannot initialize taskqueueDB: %s" % str(e))
return S_OK()
def test_matcher(self):
# insert a proper DN to run the test
resourceDescription = {
"OwnerGroup": "prod",
"OwnerDN": "/C=ch/O=DIRAC/OU=DIRAC CI/CN=ciuser",
"DIRACVersion": "pippo",
"GridCE": "some.grid.ce.org",
"ReleaseVersion": "blabla",
"VirtualOrganization": "LHCb",
"PilotInfoReportedFlag": "True",
"PilotBenchmark": "anotherPilot",
"Site": "DIRAC.Jenkins.ch",
"CPUTime": 86400,
}
wmsClient = WMSClient()
job = helloWorldJob()
job.setDestination("DIRAC.Jenkins.ch")
job.setInputData("/a/bbb")
job.setType("User")
jobDescription = createFile(job)
res = wmsClient.submitJob(job._toJDL(xmlFile=jobDescription))
self.assertTrue(res["OK"], res.get("Message"))
jobID = res["Value"]
# forcing the update
res = JobStateUpdateClient().setJobStatus(jobID, JobStatus.WAITING,
"matching", "source", None,
True)
self.assertTrue(res["OK"], res.get("Message"))
tqDB = TaskQueueDB()
tqDefDict = {
"OwnerDN": "/C=ch/O=DIRAC/OU=DIRAC CI/CN=ciuser",
"OwnerGroup": "prod",
"Setup": "dirac-JenkinsSetup",
"CPUTime": 86400,
}
res = tqDB.insertJob(jobID, tqDefDict, 10)
self.assertTrue(res["OK"], res.get("Message"))
res = MatcherClient().requestJob(resourceDescription)
print(res)
self.assertTrue(res["OK"], res.get("Message"))
wmsClient.deleteJob(jobID)
def initializeMatcherHandler(serviceInfo):
""" Matcher Service initialization
"""
global gJobDB
global gJobLoggingDB
global gTaskQueueDB
global gPilotAgentsDB
# Create JobDB object and initialize its tables.
gJobDB = JobDB()
res = gJobDB._checkTable()
if not res['OK']:
return res
# Create JobLoggingDB object and initialize its tables.
gJobLoggingDB = JobLoggingDB()
res = gJobLoggingDB._checkTable()
if not res['OK']:
return res
gTaskQueueDB = TaskQueueDB()
# Create PilotAgentsDB object and initialize its tables.
gPilotAgentsDB = PilotAgentsDB()
res = gPilotAgentsDB._checkTable()
if not res['OK']:
return res
gMonitor.registerActivity('matchTime', "Job matching time", 'Matching',
"secs", gMonitor.OP_MEAN, 300)
gMonitor.registerActivity('matchesDone', "Job Match Request", 'Matching',
"matches", gMonitor.OP_RATE, 300)
gMonitor.registerActivity('matchesOK', "Matched jobs", 'Matching',
"matches", gMonitor.OP_RATE, 300)
gMonitor.registerActivity('numTQs', "Number of Task Queues", 'Matching',
"tqsk queues", gMonitor.OP_MEAN, 300)
gTaskQueueDB.recalculateTQSharesForAll()
gThreadScheduler.addPeriodicTask(120,
gTaskQueueDB.recalculateTQSharesForAll)
gThreadScheduler.addPeriodicTask(60, sendNumTaskQueues)
sendNumTaskQueues()
return S_OK()
def initialize( self ):
"""Sets defaults
"""
self.am_setOption( "PollingTime", 60 )
self.jobDB = JobDB()
self.taskQueueDB = TaskQueueDB()
self.jobLoggingDB = JobLoggingDB()
# self.sandboxDB = SandboxDB( 'SandboxDB' )
agentTSTypes = self.am_getOption('ProductionTypes', [])
if agentTSTypes:
self.prod_types = agentTSTypes
else:
self.prod_types = Operations().getValue( 'Transformations/DataProcessing', ['MCSimulation', 'Merge'] )
gLogger.info('Will exclude the following Production types from cleaning %s'%(string.join(self.prod_types,', ')))
self.maxJobsAtOnce = self.am_getOption('MaxJobsAtOnce',200)
self.jobByJob = self.am_getOption('JobByJob',True)
self.throttlingPeriod = self.am_getOption('ThrottlingPeriod',0.)
return S_OK()
def initialize(self):
"""Sets defaults
"""
self.am_setOption("PollingTime", 60)
self.jobDB = JobDB()
self.taskQueueDB = TaskQueueDB()
# self.sandboxDB = SandboxDB( 'SandboxDB' )
self.prod_types = self.am_getOption('ProductionTypes', [
'DataReconstruction', 'DataStripping', 'MCSimulation', 'Merge',
'production'
])
gLogger.info(
'Will exclude the following Production types from cleaning %s' %
(string.join(self.prod_types, ', ')))
self.maxJobsAtOnce = self.am_getOption('MaxJobsAtOnce', 200)
self.jobByJob = self.am_getOption('JobByJob', True)
self.throttlingPeriod = self.am_getOption('ThrottlingPeriod', 0.)
return S_OK()
def initializeHandler(cls, svcInfoDict):
cls.gJobDB = JobDB()
cls.gJobLoggingDB = JobLoggingDB()
cls.gTaskQueueDB = TaskQueueDB()
cls.gElasticJobParametersDB = None
useESForJobParametersFlag = Operations().getValue(
'/Services/JobMonitoring/useESForJobParametersFlag', False)
if useESForJobParametersFlag:
cls.gElasticJobParametersDB = ElasticJobParametersDB()
return S_OK()
def initializeWMSAdministratorHandler(serviceInfo):
""" WMS AdministratorService initialization
"""
global jobDB
global taskQueueDB
jobDB = JobDB()
taskQueueDB = TaskQueueDB()
return S_OK()
def initializeHandler(cls, svcInfoDict):
""" WMS AdministratorService initialization
"""
cls.jobDB = JobDB()
cls.taskQueueDB = TaskQueueDB()
cls.elasticJobParametersDB = None
useESForJobParametersFlag = Operations().getValue(
'/Services/JobMonitoring/useESForJobParametersFlag', False)
if useESForJobParametersFlag:
cls.elasticJobParametersDB = ElasticJobParametersDB()
return S_OK()
def initializeOptimizer( self ):
"""Initialize specific parameters for TaskQueueAgent.
"""
self.waitingStatus = self.am_getOption( 'WaitingStatus', 'Waiting' )
self.waitingMinorStatus = self.am_getOption( 'WaitingMinorStatus', 'Pilot Agent Submission' )
try:
self.taskQueueDB = TaskQueueDB()
result = self.taskQueueDB.enableAllTaskQueues()
if not result[ 'OK' ]:
raise Exception( "Can't enable TaskQueues: %s" % result[ 'Message' ] )
except Exception, e:
self.log.exception()
return S_ERROR( "Cannot initialize taskqueueDB: %s" % str( e ) )
def initializeWMSAdministratorHandler(serviceInfo):
""" WMS AdministratorService initialization
:param dict serviceInfo: service information dictionary
:return: S_OK()/S_ERROR()
"""
global jobDB
global taskQueueDB
jobDB = JobDB()
taskQueueDB = TaskQueueDB()
return S_OK()
def initializeMatcherHandler( serviceInfo ): """ Matcher Service initialization """ global jobDB global jobLoggingDB global taskQueueDB jobDB = JobDB() jobLoggingDB = JobLoggingDB() taskQueueDB = TaskQueueDB() gMonitor.registerActivity( 'matchTime', "Job matching time", 'Matching', "secs" , gMonitor.OP_MEAN, 300 ) gMonitor.registerActivity( 'matchTaskQueues', "Task queues checked per job", 'Matching', "task queues" , gMonitor.OP_MEAN, 300 ) gMonitor.registerActivity( 'matchesDone', "Job Matches", 'Matching', "matches" , gMonitor.OP_MEAN, 300 ) gMonitor.registerActivity( 'numTQs', "Number of Task Queues", 'Matching', "tqsk queues" , gMonitor.OP_MEAN, 300 ) taskQueueDB.recalculateTQSharesForAll() gThreadScheduler.addPeriodicTask( 120, taskQueueDB.recalculateTQSharesForAll ) gThreadScheduler.addPeriodicTask( 120, sendNumTaskQueues ) sendNumTaskQueues() return S_OK()
def initialize( self ):
"""Sets defaults
"""
self.am_setOption( "PollingTime", 60 )
self.jobDB = JobDB()
self.taskQueueDB = TaskQueueDB()
self.jobLoggingDB = JobLoggingDB()
# self.sandboxDB = SandboxDB( 'SandboxDB' )
self.prod_types = self.am_getOption('ProductionTypes',['DataReconstruction', 'DataStripping', 'MCSimulation', 'Merge', 'production'])
gLogger.info('Will exclude the following Production types from cleaning %s'%(string.join(self.prod_types,', ')))
self.maxJobsAtOnce = self.am_getOption('MaxJobsAtOnce',200)
self.jobByJob = self.am_getOption('JobByJob',True)
self.throttlingPeriod = self.am_getOption('ThrottlingPeriod',0.)
return S_OK()
def initializeJobManagerHandler(serviceInfo):
global gJobDB, gJobLoggingDB, gtaskQueueDB, enablePilotsLogging, gPilotAgentsDB, gPilotsLoggingDB
gJobDB = JobDB()
gJobLoggingDB = JobLoggingDB()
gtaskQueueDB = TaskQueueDB()
gPilotAgentsDB = PilotAgentsDB()
# there is a problem with accessing CS with shorter paths, so full path is extracted from serviceInfo dict
enablePilotsLogging = gConfig.getValue(
serviceInfo['serviceSectionPath'].replace('JobManager',
'PilotsLogging') + '/Enable',
'False').lower() in ('yes', 'true')
if enablePilotsLogging:
gPilotsLoggingDB = PilotsLoggingDB()
return S_OK()
def initializeHandler(cls, serviceInfoDict):
""" Initialization of DB objects and OptimizationMind
"""
cls.jobDB = JobDB()
cls.jobLoggingDB = JobLoggingDB()
cls.taskQueueDB = TaskQueueDB()
cls.pilotAgentsDB = PilotAgentsDB()
cls.pilotsLoggingDB = None
enablePilotsLogging = Operations().getValue(
'/Services/JobMonitoring/usePilotsLoggingFlag', False)
if enablePilotsLogging:
cls.pilotsLoggingDB = PilotsLoggingDB()
cls.msgClient = MessageClient("WorkloadManagement/OptimizationMind")
cls.__connectToOptMind()
gThreadScheduler.addPeriodicTask(60, cls.__connectToOptMind)
return S_OK()
def initialize(self):
"""Sets defaults
"""
self.am_setOption("PollingTime", 60)
self.jobDB = JobDB()
self.taskQueueDB = TaskQueueDB()
# self.sandboxDB = SandboxDB( 'SandboxDB' )
self.prod_types = self.am_getOption(
"ProductionTypes", ["DataReconstruction", "DataStripping", "MCSimulation", "Merge", "production"]
)
gLogger.info(
"Will exclude the following Production types from cleaning %s" % (string.join(self.prod_types, ", "))
)
self.maxJobsAtOnce = self.am_getOption("MaxJobsAtOnce", 200)
self.jobByJob = self.am_getOption("JobByJob", True)
self.throttlingPeriod = self.am_getOption("ThrottlingPeriod", 0.0)
return S_OK()
def initialize( self ):
"""Sets defaults
"""
self.am_setOption( "PollingTime", 60 )
self.jobDB = JobDB()
self.taskQueueDB = TaskQueueDB()
self.jobLoggingDB = JobLoggingDB()
# self.sandboxDB = SandboxDB( 'SandboxDB' )
agentTSTypes = self.am_getOption('ProductionTypes', [])
if agentTSTypes:
self.prod_types = agentTSTypes
else:
self.prod_types = Operations().getValue( 'Transformations/DataProcessing', ['MCSimulation', 'Merge'] )
gLogger.info('Will exclude the following Production types from cleaning %s'%(string.join(self.prod_types,', ')))
self.maxJobsAtOnce = self.am_getOption('MaxJobsAtOnce',100)
self.jobByJob = self.am_getOption('JobByJob',True)
self.throttlingPeriod = self.am_getOption('ThrottlingPeriod',0.)
return S_OK()
class TaskQueuesAgent(AgentModule):
"""Agent for recalculating TQ shares"""
def __init__(self, *args, **kwargs):
"""c'tor"""
AgentModule.__init__(self, *args, **kwargs)
# clients
self.tqDB = None
def initialize(self):
"""just initialize TQDB"""
self.tqDB = TaskQueueDB()
return S_OK()
def execute(self):
"""calls TQDB.recalculateTQSharesForAll"""
res = self.tqDB.recalculateTQSharesForAll()
if not res["OK"]:
self.log.error("Error recalculating TQ shares", res["Message"])
return S_OK()
class TaskQueueAgent( OptimizerModule ):
"""
The specific Optimizer must provide the following methods:
- initializeOptimizer() before each execution cycle
- checkJob() - the main method called for each job
"""
#############################################################################
def initializeOptimizer( self ):
"""Initialize specific parameters for TaskQueueAgent.
"""
self.waitingStatus = self.am_getOption( 'WaitingStatus', 'Waiting' )
self.waitingMinorStatus = self.am_getOption( 'WaitingMinorStatus', 'Pilot Agent Submission' )
try:
self.taskQueueDB = TaskQueueDB()
result = self.taskQueueDB.enableAllTaskQueues()
if not result[ 'OK' ]:
raise Exception( "Can't enable TaskQueues: %s" % result[ 'Message' ] )
except Exception, e:
self.log.exception()
return S_ERROR( "Cannot initialize taskqueueDB: %s" % str( e ) )
return S_OK()
def initializeHandler(cls, serviceInfoDict):
cls.jobDB = JobDB()
cls.jobLoggingDB = JobLoggingDB()
cls.taskQueueDB = TaskQueueDB()
cls.pilotAgentsDB = PilotAgentsDB()
cls.limiter = Limiter(jobDB=cls.jobDB)
cls.taskQueueDB.recalculateTQSharesForAll()
gMonitor.registerActivity('matchTime', "Job matching time",
'Matching', "secs", gMonitor.OP_MEAN, 300)
gMonitor.registerActivity('matchesDone', "Job Match Request",
'Matching', "matches", gMonitor.OP_RATE, 300)
gMonitor.registerActivity('matchesOK', "Matched jobs",
'Matching', "matches", gMonitor.OP_RATE, 300)
gMonitor.registerActivity('numTQs', "Number of Task Queues",
'Matching', "tqsk queues", gMonitor.OP_MEAN, 300)
gThreadScheduler.addPeriodicTask(120, cls.taskQueueDB.recalculateTQSharesForAll)
gThreadScheduler.addPeriodicTask(60, cls.sendNumTaskQueues)
cls.sendNumTaskQueues()
return S_OK()
def initializeWMSAdministratorHandler(serviceInfo):
""" WMS AdministratorService initialization
"""
global jobDB
global pilotDB
global taskQueueDB
global enablePilotsLogging
# there is a problem with accessing CS with shorter paths, so full path is extracted from serviceInfo dict
enablePilotsLogging = gConfig.getValue(
serviceInfo['serviceSectionPath'].replace(
'WMSAdministrator',
'PilotsLogging') + '/Enable',
'False').lower() in (
'yes',
'true')
jobDB = JobDB()
pilotDB = PilotAgentsDB()
taskQueueDB = TaskQueueDB()
if enablePilotsLogging:
pilotsLoggingDB = PilotsLoggingDB()
return S_OK()
# File : dirac-admin-submit-pilot-for-job
# Author : Ricardo Graciani
########################################################################
__RCSID__ = "cd6b25c (2010-12-04 11:45:50 +0000) Ricardo Graciani <*****@*****.**>"
import sys
import DIRAC
from DIRAC.Core.Base import Script
Script.parseCommandLine( ignoreErrors = True )
args = Script.getPositionalArgs()
from DIRAC.WorkloadManagementSystem.DB.TaskQueueDB import TaskQueueDB
from DIRAC.WorkloadManagementSystem.DB.JobDB import JobDB
jobdb = JobDB()
tqdb = TaskQueueDB()
result = jobdb.selectJobs( { 'Status' : [ 'Received', 'Checking', 'Waiting' ] } )
if not result[ 'OK' ]:
print result[ 'Message' ]
sys.exit( 1 )
jobList = result[ 'Value' ]
print tqdb.forceRecreationOfTables()
for job in jobList:
result = jobdb.getJobAttribute( job, 'RescheduleCounter' )
if not result[ 'OK' ]:
print "Cannot get reschedule counter for job %s" % job
rC = 0
rC = result[ 'Value' ]
if rC >= jobdb.maxRescheduling:
jobdb.setJobAttribute( job, "RescheduleCounter", "0" )
class JobCleaningAgent( AgentModule ):
"""
The specific agents must provide the following methods:
- initialize() for initial settings
- beginExecution()
- execute() - the main method called in the agent cycle
- endExecution()
- finalize() - the graceful exit of the method, this one is usually used
for the agent restart
"""
#############################################################################
def initialize( self ):
"""Sets defaults
"""
self.am_setOption( "PollingTime", 60 )
self.jobDB = JobDB()
self.taskQueueDB = TaskQueueDB()
self.jobLoggingDB = JobLoggingDB()
# self.sandboxDB = SandboxDB( 'SandboxDB' )
agentTSTypes = self.am_getOption('ProductionTypes', [])
if agentTSTypes:
self.prod_types = agentTSTypes
else:
self.prod_types = Operations().getValue( 'Transformations/DataProcessing', ['MCSimulation', 'Merge'] )
gLogger.info('Will exclude the following Production types from cleaning %s'%(string.join(self.prod_types,', ')))
self.maxJobsAtOnce = self.am_getOption('MaxJobsAtOnce',100)
self.jobByJob = self.am_getOption('JobByJob',True)
self.throttlingPeriod = self.am_getOption('ThrottlingPeriod',0.)
return S_OK()
def __getAllowedJobTypes( self ):
#Get valid jobTypes
result = self.jobDB.getDistinctJobAttributes( 'JobType' )
if not result[ 'OK' ]:
return result
cleanJobTypes = []
for jobType in result[ 'Value' ]:
if jobType not in self.prod_types:
cleanJobTypes.append( jobType )
self.log.notice( "JobTypes to clean %s" % cleanJobTypes )
return S_OK( cleanJobTypes )
#############################################################################
def execute( self ):
"""The PilotAgent execution method.
"""
#Delete jobs in "Deleted" state
result = self.removeJobsByStatus( { 'Status' : 'Deleted' } )
if not result[ 'OK' ]:
return result
#Get all the Job types that can be cleaned
result = self.__getAllowedJobTypes()
if not result[ 'OK' ]:
return result
baseCond = { 'JobType' : result[ 'Value' ] }
# Remove jobs with final status
for status in REMOVE_STATUS_DELAY:
delay = REMOVE_STATUS_DELAY[ status ]
condDict = dict( baseCond )
condDict[ 'Status' ] = status
delTime = str( Time.dateTime() - delay * Time.day )
result = self.removeJobsByStatus( condDict, delTime )
if not result['OK']:
gLogger.warn( 'Failed to remove jobs in status %s' % status )
return S_OK()
def removeJobsByStatus( self, condDict, delay = False ):
""" Remove deleted jobs
"""
if delay:
gLogger.verbose( "Removing jobs with %s and older than %s" % ( condDict, delay ) )
result = self.jobDB.selectJobs( condDict, older = delay, limit = self.maxJobsAtOnce )
else:
gLogger.verbose( "Removing jobs with %s " % condDict )
result = self.jobDB.selectJobs( condDict, limit = self.maxJobsAtOnce )
if not result['OK']:
return result
jobList = result['Value']
if len(jobList) > self.maxJobsAtOnce:
jobList = jobList[:self.maxJobsAtOnce]
if not jobList:
return S_OK()
self.log.notice( "Deleting %s jobs for %s" % ( len( jobList ), condDict ) )
count = 0
error_count = 0
result = SandboxStoreClient( useCertificates = True ).unassignJobs( jobList )
if not result[ 'OK' ]:
gLogger.warn( "Cannot unassign jobs to sandboxes", result[ 'Message' ] )
result = self.deleteJobOversizedSandbox( jobList )
if not result[ 'OK' ]:
gLogger.warn( "Cannot schedle removal of oversized sandboxes", result[ 'Message' ] )
return result
failedJobs = result['Value']['Failed']
for job in failedJobs:
jobList.pop( jobList.index( job ) )
if self.jobByJob:
for jobID in jobList:
resultJobDB = self.jobDB.removeJobFromDB( jobID )
resultTQ = self.taskQueueDB.deleteJob( jobID )
resultLogDB = self.jobLoggingDB.deleteJob( jobID )
errorFlag = False
if not resultJobDB['OK']:
gLogger.warn( 'Failed to remove job %d from JobDB' % jobID, result['Message'] )
errorFlag = True
if not resultTQ['OK']:
gLogger.warn( 'Failed to remove job %d from TaskQueueDB' % jobID, result['Message'] )
errorFlag = True
if not resultLogDB['OK']:
gLogger.warn( 'Failed to remove job %d from JobLoggingDB' % jobID, result['Message'] )
errorFlag = True
if errorFlag:
error_count += 1
else:
count += 1
if self.throttlingPeriod:
time.sleep(self.throttlingPeriod)
else:
result = self.jobDB.removeJobFromDB( jobList )
if not result['OK']:
gLogger.error('Failed to delete %d jobs from JobDB' % len(jobList) )
else:
gLogger.info('Deleted %d jobs from JobDB' % len(jobList) )
for jobID in jobList:
resultTQ = self.taskQueueDB.deleteJob( jobID )
if not resultTQ['OK']:
gLogger.warn( 'Failed to remove job %d from TaskQueueDB' % jobID, resultTQ['Message'] )
error_count += 1
else:
count += 1
result = self.jobLoggingDB.deleteJob( jobList )
if not result['OK']:
gLogger.error('Failed to delete %d jobs from JobLoggingDB' % len(jobList) )
else:
gLogger.info('Deleted %d jobs from JobLoggingDB' % len(jobList) )
if count > 0 or error_count > 0 :
gLogger.info( 'Deleted %d jobs from JobDB, %d errors' % ( count, error_count ) )
return S_OK()
def deleteJobOversizedSandbox( self, jobIDList ):
""" Delete the job oversized sandbox files from storage elements
"""
failed = {}
successful = {}
lfnDict = {}
for jobID in jobIDList:
result = self.jobDB.getJobParameter( jobID, 'OutputSandboxLFN' )
if result['OK']:
lfn = result['Value']
if lfn:
lfnDict[lfn] = jobID
else:
successful[jobID] = 'No oversized sandbox found'
else:
gLogger.warn( 'Error interrogting JobDB: %s' % result['Message'] )
if not lfnDict:
return S_OK( {'Successful':successful, 'Failed':failed} )
# Schedule removal of the LFNs now
for lfn,jobID in lfnDict.items():
result = self.jobDB.getJobAttributes( jobID, ['OwnerDN', 'OwnerGroup'] )
if not result['OK']:
failed[jobID] = lfn
continue
if not result['Value']:
failed[jobID] = lfn
continue
ownerDN = result['Value']['OwnerDN']
ownerGroup = result['Value']['OwnerGroup']
result = self.__setRemovalRequest( lfn, ownerDN, ownerGroup )
if not result['OK']:
failed[jobID] = lfn
else:
successful[jobID] = lfn
result = {'Successful':successful, 'Failed':failed}
return S_OK( result )
def __setRemovalRequest( self, lfn, ownerDN, ownerGroup ):
""" Set removal request with the given credentials
"""
oRequest = Request()
oRequest.OwnerDN = ownerDN
oRequest.OwnerGroup = ownerGroup
oRequest.RequestName = os.path.basename( lfn ).strip() + '_removal_request.xml'
oRequest.SourceComponent = 'JobCleaningAgent'
removeFile = Operation()
removeFile.Type = 'RemoveFile'
removedFile = File()
removedFile.LFN = lfn
removeFile.addFile( removedFile )
oRequest.addOperation( removeFile )
return ReqClient().putRequest( oRequest )
Suggestion: for local testing, run this with::
python -m pytest -c ../pytest.ini -vv tests/Integration/WorkloadManagementSystem/Test_TaskQueueDB.py
"""
from DIRAC import gLogger
from DIRAC.Core.Base.Script import parseCommandLine
parseCommandLine()
from DIRAC.WorkloadManagementSystem.DB.TaskQueueDB import TaskQueueDB
gLogger.setLevel('DEBUG')
tqDB = TaskQueueDB()
def test_basicChain():
""" a basic put - remove
"""
tqDefDict = {
'OwnerDN': '/my/DN',
'OwnerGroup': 'myGroup',
'Setup': 'aSetup',
'CPUTime': 50000
}
result = tqDB.insertJob(123, tqDefDict, 10)
assert result['OK'] is True
result = tqDB.getTaskQueueForJobs([123])
assert result['OK'] is True
class JobCleaningAgent( AgentModule ):
"""
The specific agents must provide the following methods:
- initialize() for initial settings
- beginExecution()
- execute() - the main method called in the agent cycle
- endExecution()
- finalize() - the graceful exit of the method, this one is usually used
for the agent restart
"""
def __init__( self, *args, **kwargs ):
""" c'tor
"""
AgentModule.__init__( self, *args, **kwargs )
#clients
# FIXME: shouldn't we avoid using the DBs directly, and instead go through the service?
self.jobDB = None
self.taskQueueDB = None
self.jobLoggingDB = None
self.maxJobsAtOnce = 100
self.jobByJob = False
self.throttlingPeriod = 0.
self.removeStatusDelay = {'Done':7,
'Killed':1,
'Failed':7 }
#############################################################################
def initialize( self ):
""" Sets defaults
"""
self.am_setOption( "PollingTime", 120 )
self.jobDB = JobDB()
self.taskQueueDB = TaskQueueDB()
self.jobLoggingDB = JobLoggingDB()
# self.sandboxDB = SandboxDB( 'SandboxDB' )
agentTSTypes = self.am_getOption('ProductionTypes', [])
if agentTSTypes:
self.prod_types = agentTSTypes
else:
self.prod_types = Operations().getValue( 'Transformations/DataProcessing', ['MCSimulation', 'Merge'] )
gLogger.info( "Will exclude the following Production types from cleaning %s" % ( ', '.join( self.prod_types ) ) )
self.maxJobsAtOnce = self.am_getOption( 'MaxJobsAtOnce', 500 )
self.jobByJob = self.am_getOption( 'JobByJob', False )
self.throttlingPeriod = self.am_getOption('ThrottlingPeriod', 0.)
self.removeStatusDelay['Done'] = self.am_getOption( 'RemoveStatusDelay/Done', 7 )
self.removeStatusDelay['Killed'] = self.am_getOption( 'RemoveStatusDelay/Killed', 7 )
self.removeStatusDelay['Failed'] = self.am_getOption( 'RemoveStatusDelay/Failed', 7 )
return S_OK()
def __getAllowedJobTypes( self ):
""" Get valid jobTypes
"""
result = self.jobDB.getDistinctJobAttributes( 'JobType' )
if not result[ 'OK' ]:
return result
cleanJobTypes = []
for jobType in result[ 'Value' ]:
if jobType not in self.prod_types:
cleanJobTypes.append( jobType )
self.log.notice( "JobTypes to clean %s" % cleanJobTypes )
return S_OK( cleanJobTypes )
#############################################################################
def execute( self ):
""" Remove jobs in various status
"""
#Delete jobs in "Deleted" state
result = self.removeJobsByStatus( { 'Status' : 'Deleted' } )
if not result[ 'OK' ]:
return result
#Get all the Job types that can be cleaned
result = self.__getAllowedJobTypes()
if not result[ 'OK' ]:
return result
# No jobs in the system subject to removal
if not result['Value']:
return S_OK()
baseCond = { 'JobType' : result[ 'Value' ] }
# Remove jobs with final status
for status in self.removeStatusDelay:
delay = self.removeStatusDelay[ status ]
condDict = dict( baseCond )
condDict[ 'Status' ] = status
delTime = str( Time.dateTime() - delay * Time.day )
result = self.removeJobsByStatus( condDict, delTime )
if not result['OK']:
gLogger.warn( 'Failed to remove jobs in status %s' % status )
return S_OK()
def removeJobsByStatus( self, condDict, delay = False ):
""" Remove deleted jobs
"""
if delay:
gLogger.verbose( "Removing jobs with %s and older than %s" % ( condDict, delay ) )
result = self.jobDB.selectJobs( condDict, older = delay, limit = self.maxJobsAtOnce )
else:
gLogger.verbose( "Removing jobs with %s " % condDict )
result = self.jobDB.selectJobs( condDict, limit = self.maxJobsAtOnce )
if not result['OK']:
return result
jobList = result['Value']
if len(jobList) > self.maxJobsAtOnce:
jobList = jobList[:self.maxJobsAtOnce]
if not jobList:
return S_OK()
self.log.notice( "Deleting %s jobs for %s" % ( len( jobList ), condDict ) )
count = 0
error_count = 0
result = SandboxStoreClient( useCertificates = True ).unassignJobs( jobList )
if not result[ 'OK' ]:
gLogger.warn( "Cannot unassign jobs to sandboxes", result[ 'Message' ] )
result = self.deleteJobOversizedSandbox( jobList )
if not result[ 'OK' ]:
gLogger.warn( "Cannot schedule removal of oversized sandboxes", result[ 'Message' ] )
return result
failedJobs = result['Value']['Failed']
for job in failedJobs:
jobList.pop( jobList.index( job ) )
# TODO: we should not remove a job if it still has requests in the RequestManager.
# But this logic should go in the client or in the service, and right now no service expose jobDB.removeJobFromDB
if self.jobByJob:
for jobID in jobList:
resultJobDB = self.jobDB.removeJobFromDB( jobID )
resultTQ = self.taskQueueDB.deleteJob( jobID )
resultLogDB = self.jobLoggingDB.deleteJob( jobID )
errorFlag = False
if not resultJobDB['OK']:
gLogger.warn( 'Failed to remove job %d from JobDB' % jobID, result['Message'] )
errorFlag = True
if not resultTQ['OK']:
gLogger.warn( 'Failed to remove job %d from TaskQueueDB' % jobID, result['Message'] )
errorFlag = True
if not resultLogDB['OK']:
gLogger.warn( 'Failed to remove job %d from JobLoggingDB' % jobID, result['Message'] )
errorFlag = True
if errorFlag:
error_count += 1
else:
count += 1
if self.throttlingPeriod:
time.sleep(self.throttlingPeriod)
else:
result = self.jobDB.removeJobFromDB( jobList )
if not result['OK']:
gLogger.error('Failed to delete %d jobs from JobDB' % len(jobList) )
else:
gLogger.info('Deleted %d jobs from JobDB' % len(jobList) )
for jobID in jobList:
resultTQ = self.taskQueueDB.deleteJob( jobID )
if not resultTQ['OK']:
gLogger.warn( 'Failed to remove job %d from TaskQueueDB' % jobID, resultTQ['Message'] )
error_count += 1
else:
count += 1
result = self.jobLoggingDB.deleteJob( jobList )
if not result['OK']:
gLogger.error('Failed to delete %d jobs from JobLoggingDB' % len(jobList) )
else:
gLogger.info('Deleted %d jobs from JobLoggingDB' % len(jobList) )
if count > 0 or error_count > 0 :
gLogger.info( 'Deleted %d jobs from JobDB, %d errors' % ( count, error_count ) )
return S_OK()
def deleteJobOversizedSandbox( self, jobIDList ):
""" Delete the job oversized sandbox files from storage elements
"""
failed = {}
successful = {}
lfnDict = {}
for jobID in jobIDList:
result = self.jobDB.getJobParameter( jobID, 'OutputSandboxLFN' )
if result['OK']:
lfn = result['Value']
if lfn:
lfnDict[lfn] = jobID
else:
successful[jobID] = 'No oversized sandbox found'
else:
gLogger.warn( 'Error interrogating JobDB: %s' % result['Message'] )
if not lfnDict:
return S_OK( {'Successful':successful, 'Failed':failed} )
# Schedule removal of the LFNs now
for lfn,jobID in lfnDict.items():
result = self.jobDB.getJobAttributes( jobID, ['OwnerDN', 'OwnerGroup'] )
if not result['OK']:
failed[jobID] = lfn
continue
if not result['Value']:
failed[jobID] = lfn
continue
ownerDN = result['Value']['OwnerDN']
ownerGroup = result['Value']['OwnerGroup']
result = self.__setRemovalRequest( lfn, ownerDN, ownerGroup )
if not result['OK']:
failed[jobID] = lfn
else:
successful[jobID] = lfn
result = {'Successful':successful, 'Failed':failed}
return S_OK( result )
def __setRemovalRequest( self, lfn, ownerDN, ownerGroup ):
""" Set removal request with the given credentials
"""
oRequest = Request()
oRequest.OwnerDN = ownerDN
oRequest.OwnerGroup = ownerGroup
oRequest.RequestName = os.path.basename( lfn ).strip() + '_removal_request.xml'
oRequest.SourceComponent = 'JobCleaningAgent'
removeFile = Operation()
removeFile.Type = 'RemoveFile'
removedFile = File()
removedFile.LFN = lfn
removeFile.addFile( removedFile )
oRequest.addOperation( removeFile )
return ReqClient().putRequest( oRequest )
class JobCleaningAgent( AgentModule ):
"""
The specific agents must provide the following methods:
- initialize() for initial settings
- beginExecution()
- execute() - the main method called in the agent cycle
- endExecution()
- finalize() - the graceful exit of the method, this one is usually used
for the agent restart
"""
#############################################################################
def initialize( self ):
"""Sets defaults
"""
self.am_setOption( "PollingTime", 60 )
self.jobDB = JobDB()
self.taskQueueDB = TaskQueueDB()
self.jobLoggingDB = JobLoggingDB()
# self.sandboxDB = SandboxDB( 'SandboxDB' )
agentTSTypes = self.am_getOption('ProductionTypes', [])
if agentTSTypes:
self.prod_types = agentTSTypes
else:
self.prod_types = Operations().getValue( 'Transformations/DataProcessing', ['MCSimulation', 'Merge'] )
gLogger.info('Will exclude the following Production types from cleaning %s'%(string.join(self.prod_types,', ')))
self.maxJobsAtOnce = self.am_getOption('MaxJobsAtOnce',200)
self.jobByJob = self.am_getOption('JobByJob',True)
self.throttlingPeriod = self.am_getOption('ThrottlingPeriod',0.)
return S_OK()
def __getAllowedJobTypes( self ):
#Get valid jobTypes
result = self.jobDB.getDistinctJobAttributes( 'JobType' )
if not result[ 'OK' ]:
return result
cleanJobTypes = []
for jobType in result[ 'Value' ]:
if jobType not in self.prod_types:
cleanJobTypes.append( jobType )
self.log.notice( "JobTypes to clean %s" % cleanJobTypes )
return S_OK( cleanJobTypes )
#############################################################################
def execute( self ):
"""The PilotAgent execution method.
"""
#Delete jobs in "Deleted" state
result = self.removeJobsByStatus( { 'Status' : 'Deleted' } )
if not result[ 'OK' ]:
return result
#Get all the Job types that can be cleaned
result = self.__getAllowedJobTypes()
if not result[ 'OK' ]:
return result
baseCond = { 'JobType' : result[ 'Value' ] }
# Remove jobs with final status
for status in REMOVE_STATUS_DELAY:
delay = REMOVE_STATUS_DELAY[ status ]
condDict = dict( baseCond )
condDict[ 'Status' ] = status
delTime = str( Time.dateTime() - delay * Time.day )
result = self.removeJobsByStatus( condDict, delTime )
if not result['OK']:
gLogger.warn( 'Failed to remove jobs in status %s' % status )
return S_OK()
def removeJobsByStatus( self, condDict, delay = False ):
""" Remove deleted jobs
"""
if delay:
gLogger.verbose( "Removing jobs with %s and older than %s" % ( condDict, delay ) )
result = self.jobDB.selectJobs( condDict, older = delay, limit = self.maxJobsAtOnce )
else:
gLogger.verbose( "Removing jobs with %s " % condDict )
result = self.jobDB.selectJobs( condDict, limit = self.maxJobsAtOnce )
if not result['OK']:
return result
jobList = result['Value']
if len(jobList) > self.maxJobsAtOnce:
jobList = jobList[:self.maxJobsAtOnce]
if not jobList:
return S_OK()
self.log.notice( "Deleting %s jobs for %s" % ( len( jobList ), condDict ) )
count = 0
error_count = 0
result = SandboxStoreClient( useCertificates = True ).unassignJobs( jobList )
if not result[ 'OK' ]:
gLogger.warn( "Cannot unassign jobs to sandboxes", result[ 'Message' ] )
if self.jobByJob:
for jobID in jobList:
resultJobDB = self.jobDB.removeJobFromDB( jobID )
resultTQ = self.taskQueueDB.deleteJob( jobID )
resultLogDB = self.jobLoggingDB.deleteJob( jobID )
errorFlag = False
if not resultJobDB['OK']:
gLogger.warn( 'Failed to remove job %d from JobDB' % jobID, result['Message'] )
errorFlag = True
if not resultTQ['OK']:
gLogger.warn( 'Failed to remove job %d from TaskQueueDB' % jobID, result['Message'] )
errorFlag = True
if not resultLogDB['OK']:
gLogger.warn( 'Failed to remove job %d from JobLoggingDB' % jobID, result['Message'] )
errorFlag = True
if errorFlag:
error_count += 1
else:
count += 1
if self.throttlingPeriod:
time.sleep(self.throttlingPeriod)
else:
result = self.jobDB.removeJobFromDB( jobList )
if not result['OK']:
gLogger.error('Failed to delete %d jobs from JobDB' % len(jobList) )
else:
gLogger.info('Deleted %d jobs from JobDB' % len(jobList) )
for jobID in jobList:
resultTQ = self.taskQueueDB.deleteJob( jobID )
if not resultTQ['OK']:
gLogger.warn( 'Failed to remove job %d from TaskQueueDB' % jobID, resultTQ['Message'] )
error_count += 1
else:
count += 1
result = self.jobLoggingDB.deleteJob( jobList )
if not result['OK']:
gLogger.error('Failed to delete %d jobs from JobLoggingDB' % len(jobList) )
else:
gLogger.info('Deleted %d jobs from JobLoggingDB' % len(jobList) )
if count > 0 or error_count > 0 :
gLogger.info( 'Deleted %d jobs from JobDB, %d errors' % ( count, error_count ) )
return S_OK()
def setUp(self):
gLogger.setLevel('DEBUG')
self.tqDB = TaskQueueDB()
class Matcher( object ):
""" Logic for matching
"""
def __init__( self, pilotAgentsDB = None, jobDB = None, tqDB = None, jlDB = None, opsHelper = None ):
""" c'tor
"""
if pilotAgentsDB:
self.pilotAgentsDB = pilotAgentsDB
else:
self.pilotAgentsDB = PilotAgentsDB()
if jobDB:
self.jobDB = jobDB
else:
self.jobDB = JobDB()
if tqDB:
self.tqDB = tqDB
else:
self.tqDB = TaskQueueDB()
if jlDB:
self.jlDB = jlDB
else:
self.jlDB = JobLoggingDB()
if opsHelper:
self.opsHelper = opsHelper
else:
self.opsHelper = Operations()
self.log = gLogger.getSubLogger( "Matcher" )
self.limiter = Limiter( jobDB = self.jobDB, opsHelper = self.opsHelper )
def selectJob( self, resourceDescription, credDict ):
""" Main job selection function to find the highest priority job matching the resource capacity
"""
startTime = time.time()
resourceDict = self._getResourceDict( resourceDescription, credDict )
negativeCond = self.limiter.getNegativeCondForSite( resourceDict['Site'] )
result = self.tqDB.matchAndGetJob( resourceDict, negativeCond = negativeCond )
if not result['OK']:
return result
result = result['Value']
if not result['matchFound']:
self.log.info( "No match found" )
raise RuntimeError( "No match found" )
jobID = result['jobId']
resAtt = self.jobDB.getJobAttributes( jobID, ['OwnerDN', 'OwnerGroup', 'Status'] )
if not resAtt['OK']:
raise RuntimeError( 'Could not retrieve job attributes' )
if not resAtt['Value']:
raise RuntimeError( "No attributes returned for job" )
if not resAtt['Value']['Status'] == 'Waiting':
self.log.error( 'Job matched by the TQ is not in Waiting state', str( jobID ) )
result = self.tqDB.deleteJob( jobID )
if not result[ 'OK' ]:
return result
raise RuntimeError( "Job %s is not in Waiting state" % str( jobID ) )
self._reportStatus( resourceDict, jobID )
result = self.jobDB.getJobJDL( jobID )
if not result['OK']:
raise RuntimeError( "Failed to get the job JDL" )
resultDict = {}
resultDict['JDL'] = result['Value']
resultDict['JobID'] = jobID
matchTime = time.time() - startTime
self.log.info( "Match time: [%s]" % str( matchTime ) )
gMonitor.addMark( "matchTime", matchTime )
# Get some extra stuff into the response returned
resOpt = self.jobDB.getJobOptParameters( jobID )
if resOpt['OK']:
for key, value in resOpt['Value'].items():
resultDict[key] = value
resAtt = self.jobDB.getJobAttributes( jobID, ['OwnerDN', 'OwnerGroup'] )
if not resAtt['OK']:
raise RuntimeError( 'Could not retrieve job attributes' )
if not resAtt['Value']:
raise RuntimeError( 'No attributes returned for job' )
if self.opsHelper.getValue( "JobScheduling/CheckMatchingDelay", True ):
self.limiter.updateDelayCounters( resourceDict['Site'], jobID )
pilotInfoReportedFlag = resourceDict.get( 'PilotInfoReportedFlag', False )
if not pilotInfoReportedFlag:
self._updatePilotInfo( resourceDict )
self._updatePilotJobMapping( resourceDict, jobID )
resultDict['DN'] = resAtt['Value']['OwnerDN']
resultDict['Group'] = resAtt['Value']['OwnerGroup']
resultDict['PilotInfoReportedFlag'] = True
return resultDict
def _getResourceDict( self, resourceDescription, credDict ):
""" from resourceDescription to resourceDict (just various mods)
"""
resourceDict = self._processResourceDescription( resourceDescription )
resourceDict = self._checkCredentials( resourceDict, credDict )
self._checkPilotVersion( resourceDict )
if not self._checkMask( resourceDict ):
# Banned destinations can only take Test jobs
resourceDict['JobType'] = 'Test'
self.log.verbose( "Resource description:" )
for key in resourceDict:
self.log.verbose( "%s : %s" % ( key.rjust( 20 ), resourceDict[ key ] ) )
return resourceDict
def _processResourceDescription( self, resourceDescription ):
""" Check and form the resource description dictionary
resourceDescription is a ceDict coming from a JobAgent, for example.
"""
resourceDict = {}
if isinstance( resourceDescription, basestring ):
classAdAgent = ClassAd( resourceDescription )
if not classAdAgent.isOK():
raise ValueError( 'Illegal Resource JDL' )
self.log.verbose( classAdAgent.asJDL() )
for name in singleValueDefFields:
if classAdAgent.lookupAttribute( name ):
if name == 'CPUTime':
resourceDict[name] = classAdAgent.getAttributeInt( name )
else:
resourceDict[name] = classAdAgent.getAttributeString( name )
for name in multiValueMatchFields:
if classAdAgent.lookupAttribute( name ):
if name == 'SubmitPool':
resourceDict[name] = classAdAgent.getListFromExpression( name )
else:
resourceDict[name] = classAdAgent.getAttributeString( name )
# Check if a JobID is requested
if classAdAgent.lookupAttribute( 'JobID' ):
resourceDict['JobID'] = classAdAgent.getAttributeInt( 'JobID' )
for k in ( 'DIRACVersion', 'ReleaseVersion', 'ReleaseProject', 'VirtualOrganization' ):
if classAdAgent.lookupAttribute( k ):
resourceDict[ k ] = classAdAgent.getAttributeString( k )
else:
for name in singleValueDefFields:
if resourceDescription.has_key( name ):
resourceDict[name] = resourceDescription[name]
for name in multiValueMatchFields:
if resourceDescription.has_key( name ):
resourceDict[name] = resourceDescription[name]
if resourceDescription.has_key( 'JobID' ):
resourceDict['JobID'] = resourceDescription['JobID']
for k in ( 'DIRACVersion', 'ReleaseVersion', 'ReleaseProject', 'VirtualOrganization',
'PilotReference', 'PilotBenchmark', 'PilotInfoReportedFlag' ):
if k in resourceDescription:
resourceDict[ k ] = resourceDescription[ k ]
return resourceDict
def _reportStatus( self, resourceDict, jobID ):
""" Reports the status of the matched job in jobDB and jobLoggingDB
Do not fail if errors happen here
"""
attNames = ['Status', 'MinorStatus', 'ApplicationStatus', 'Site']
attValues = ['Matched', 'Assigned', 'Unknown', resourceDict['Site']]
result = self.jobDB.setJobAttributes( jobID, attNames, attValues )
if not result['OK']:
self.log.error( "Problem reporting job status", "setJobAttributes, jobID = %s: %s" % ( jobID, result['Message'] ) )
else:
self.log.verbose( "Set job attributes for jobID %s" % jobID )
result = self.jlDB.addLoggingRecord( jobID,
status = 'Matched',
minor = 'Assigned',
source = 'Matcher' )
if not result['OK']:
self.log.error( "Problem reporting job status", "addLoggingRecord, jobID = %s: %s" % ( jobID, result['Message'] ) )
else:
self.log.verbose( "Added logging record for jobID %s" % jobID )
def _checkMask( self, resourceDict ):
""" Check the mask: are we allowed to run normal jobs?
FIXME: should we move to site OR SE?
"""
if not 'Site' in resourceDict:
self.log.error( "Missing Site Name in Resource JDL" )
raise RuntimeError( "Missing Site Name in Resource JDL" )
# Get common site mask and check the agent site
result = self.jobDB.getSiteMask( siteState = 'Active' )
if not result['OK']:
self.log.error( "Internal error", "getSiteMask: %s" % result['Message'] )
raise RuntimeError( "Internal error" )
maskList = result['Value']
if resourceDict['Site'] not in maskList:
return False
return True
def _updatePilotInfo( self, resourceDict ):
""" Update pilot information - do not fail if we don't manage to do it
"""
pilotReference = resourceDict.get( 'PilotReference', '' )
if pilotReference:
gridCE = resourceDict.get( 'GridCE', 'Unknown' )
site = resourceDict.get( 'Site', 'Unknown' )
benchmark = resourceDict.get( 'PilotBenchmark', 0.0 )
self.log.verbose( 'Reporting pilot info for %s: gridCE=%s, site=%s, benchmark=%f' % ( pilotReference, gridCE, site, benchmark ) )
result = self.pilotAgentsDB.setPilotStatus( pilotReference, status = 'Running', gridSite = site,
destination = gridCE, benchmark = benchmark )
if not result['OK']:
self.log.error( "Problem updating pilot information",
"; setPilotStatus. pilotReference: %s; %s" % ( pilotReference, result['Message'] ) )
def _updatePilotJobMapping( self, resourceDict, jobID ):
""" Update pilot to job mapping information
"""
pilotReference = resourceDict.get( 'PilotReference', '' )
if pilotReference:
result = self.pilotAgentsDB.setCurrentJobID( pilotReference, jobID )
if not result['OK']:
self.log.error( "Problem updating pilot information",
";setCurrentJobID. pilotReference: %s; %s" % ( pilotReference, result['Message'] ) )
result = self.pilotAgentsDB.setJobForPilot( jobID, pilotReference, updateStatus = False )
if not result['OK']:
self.log.error( "Problem updating pilot information",
"; setJobForPilot. pilotReference: %s; %s" % ( pilotReference, result['Message'] ) )
def _checkCredentials( self, resourceDict, credDict ):
""" Check if we can get a job given the passed credentials
"""
if Properties.GENERIC_PILOT in credDict[ 'properties' ]:
# You can only match groups in the same VO
if credDict[ 'group' ] == "hosts":
# for the host case the VirtualOrganization parameter
# is mandatory in resourceDict
vo = resourceDict.get( 'VirtualOrganization', '' )
else:
vo = Registry.getVOForGroup( credDict[ 'group' ] )
result = Registry.getGroupsForVO( vo )
if result[ 'OK' ]:
resourceDict[ 'OwnerGroup' ] = result[ 'Value' ]
else:
raise RuntimeError( result['Message'] )
else:
# If it's a private pilot, the DN has to be the same
if Properties.PILOT in credDict[ 'properties' ]:
self.log.notice( "Setting the resource DN to the credentials DN" )
resourceDict[ 'OwnerDN' ] = credDict[ 'DN' ]
# If it's a job sharing. The group has to be the same and just check that the DN (if any)
# belongs to the same group
elif Properties.JOB_SHARING in credDict[ 'properties' ]:
resourceDict[ 'OwnerGroup' ] = credDict[ 'group' ]
self.log.notice( "Setting the resource group to the credentials group" )
if 'OwnerDN' in resourceDict and resourceDict[ 'OwnerDN' ] != credDict[ 'DN' ]:
ownerDN = resourceDict[ 'OwnerDN' ]
result = Registry.getGroupsForDN( resourceDict[ 'OwnerDN' ] )
if not result[ 'OK' ]:
raise RuntimeError( result['Message'] )
if credDict[ 'group' ] not in result[ 'Value' ]:
# DN is not in the same group! bad boy.
self.log.notice( "You cannot request jobs from DN %s. It does not belong to your group!" % ownerDN )
resourceDict[ 'OwnerDN' ] = credDict[ 'DN' ]
# Nothing special, group and DN have to be the same
else:
resourceDict[ 'OwnerDN' ] = credDict[ 'DN' ]
resourceDict[ 'OwnerGroup' ] = credDict[ 'group' ]
return resourceDict
def _checkPilotVersion( self, resourceDict ):
""" Check the pilot DIRAC version
"""
if self.opsHelper.getValue( "Pilot/CheckVersion", True ):
if 'ReleaseVersion' not in resourceDict:
if not 'DIRACVersion' in resourceDict:
raise RuntimeError( 'Version check requested and not provided by Pilot' )
else:
pilotVersion = resourceDict['DIRACVersion']
else:
pilotVersion = resourceDict['ReleaseVersion']
validVersions = self.opsHelper.getValue( "Pilot/Version", [] )
if validVersions and pilotVersion not in validVersions:
raise RuntimeError( 'Pilot version does not match the production version %s not in ( %s )' % \
( pilotVersion, ",".join( validVersions ) ) )
# Check project if requested
validProject = self.opsHelper.getValue( "Pilot/Project", "" )
if validProject:
if 'ReleaseProject' not in resourceDict:
raise RuntimeError( "Version check requested but expected project %s not received" % validProject )
if resourceDict[ 'ReleaseProject' ] != validProject:
raise RuntimeError( "Version check requested but expected project %s != received %s" % ( validProject,
resourceDict[ 'ReleaseProject' ] ) )
class JobCleaningAgent( AgentModule ):
"""
The specific agents must provide the following methods:
- initialize() for initial settings
- beginExecution()
- execute() - the main method called in the agent cycle
- endExecution()
- finalize() - the graceful exit of the method, this one is usually used
for the agent restart
"""
#############################################################################
def initialize( self ):
"""Sets defaults
"""
self.am_setOption( "PollingTime", 60 )
self.jobDB = JobDB()
self.taskQueueDB = TaskQueueDB()
self.jobLoggingDB = JobLoggingDB()
# self.sandboxDB = SandboxDB( 'SandboxDB' )
agentTSTypes = self.am_getOption('ProductionTypes', [])
if agentTSTypes:
self.prod_types = agentTSTypes
else:
self.prod_types = Operations().getValue( 'Transformations/DataProcessing', ['MCSimulation', 'Merge'] )
gLogger.info('Will exclude the following Production types from cleaning %s'%(string.join(self.prod_types,', ')))
self.maxJobsAtOnce = self.am_getOption('MaxJobsAtOnce',200)
self.jobByJob = self.am_getOption('JobByJob',True)
self.throttlingPeriod = self.am_getOption('ThrottlingPeriod',0.)
return S_OK()
def __getAllowedJobTypes( self ):
#Get valid jobTypes
result = self.jobDB.getDistinctJobAttributes( 'JobType' )
if not result[ 'OK' ]:
return result
cleanJobTypes = []
for jobType in result[ 'Value' ]:
if jobType not in self.prod_types:
cleanJobTypes.append( jobType )
self.log.notice( "JobTypes to clean %s" % cleanJobTypes )
return S_OK( cleanJobTypes )
#############################################################################
def execute( self ):
"""The PilotAgent execution method.
"""
#Delete jobs in "Deleted" state
result = self.removeJobsByStatus( { 'Status' : 'Deleted' } )
if not result[ 'OK' ]:
return result
#Get all the Job types that can be cleaned
result = self.__getAllowedJobTypes()
if not result[ 'OK' ]:
return result
baseCond = { 'JobType' : result[ 'Value' ] }
# Remove jobs with final status
for status in REMOVE_STATUS_DELAY:
delay = REMOVE_STATUS_DELAY[ status ]
condDict = dict( baseCond )
condDict[ 'Status' ] = status
delTime = str( Time.dateTime() - delay * Time.day )
result = self.removeJobsByStatus( condDict, delTime )
if not result['OK']:
gLogger.warn( 'Failed to remove jobs in status %s' % status )
return S_OK()
def removeJobsByStatus( self, condDict, delay = False ):
""" Remove deleted jobs
"""
if delay:
gLogger.verbose( "Removing jobs with %s and older than %s" % ( condDict, delay ) )
result = self.jobDB.selectJobs( condDict, older = delay, limit = self.maxJobsAtOnce )
else:
gLogger.verbose( "Removing jobs with %s " % condDict )
result = self.jobDB.selectJobs( condDict, limit = self.maxJobsAtOnce )
if not result['OK']:
return result
jobList = result['Value']
if len(jobList) > self.maxJobsAtOnce:
jobList = jobList[:self.maxJobsAtOnce]
if not jobList:
return S_OK()
self.log.notice( "Deleting %s jobs for %s" % ( len( jobList ), condDict ) )
count = 0
error_count = 0
result = SandboxStoreClient( useCertificates = True ).unassignJobs( jobList )
if not result[ 'OK' ]:
gLogger.warn( "Cannot unassign jobs to sandboxes", result[ 'Message' ] )
if self.jobByJob:
for jobID in jobList:
resultJobDB = self.jobDB.removeJobFromDB( jobID )
resultTQ = self.taskQueueDB.deleteJob( jobID )
resultLogDB = self.jobLoggingDB.deleteJob( jobID )
errorFlag = False
if not resultJobDB['OK']:
gLogger.warn( 'Failed to remove job %d from JobDB' % jobID, result['Message'] )
errorFlag = True
if not resultTQ['OK']:
gLogger.warn( 'Failed to remove job %d from TaskQueueDB' % jobID, result['Message'] )
errorFlag = True
if not resultLogDB['OK']:
gLogger.warn( 'Failed to remove job %d from JobLoggingDB' % jobID, result['Message'] )
errorFlag = True
if errorFlag:
error_count += 1
else:
count += 1
if self.throttlingPeriod:
time.sleep(self.throttlingPeriod)
else:
result = self.jobDB.removeJobFromDB( jobList )
if not result['OK']:
gLogger.error('Failed to delete %d jobs from JobDB' % len(jobList) )
else:
gLogger.info('Deleted %d jobs from JobDB' % len(jobList) )
for jobID in jobList:
resultTQ = self.taskQueueDB.deleteJob( jobID )
if not resultTQ['OK']:
gLogger.warn( 'Failed to remove job %d from TaskQueueDB' % jobID, resultTQ['Message'] )
error_count += 1
else:
count += 1
result = self.jobLoggingDB.deleteJob( jobList )
if not result['OK']:
gLogger.error('Failed to delete %d jobs from JobLoggingDB' % len(jobList) )
else:
gLogger.info('Deleted %d jobs from JobLoggingDB' % len(jobList) )
if count > 0 or error_count > 0 :
gLogger.info( 'Deleted %d jobs from JobDB, %d errors' % ( count, error_count ) )
return S_OK()
class JobCleaningAgent(AgentModule):
"""
The specific agents must provide the following methods:
- initialize() for initial settings
- beginExecution()
- execute() - the main method called in the agent cycle
- endExecution()
- finalize() - the graceful exit of the method, this one is usually used
for the agent restart
"""
#############################################################################
def initialize(self):
"""Sets defaults
"""
self.am_setOption("PollingTime", 60)
self.jobDB = JobDB()
self.taskQueueDB = TaskQueueDB()
# self.sandboxDB = SandboxDB( 'SandboxDB' )
self.prod_types = self.am_getOption(
"ProductionTypes", ["DataReconstruction", "DataStripping", "MCSimulation", "Merge", "production"]
)
gLogger.info(
"Will exclude the following Production types from cleaning %s" % (string.join(self.prod_types, ", "))
)
self.maxJobsAtOnce = self.am_getOption("MaxJobsAtOnce", 200)
self.jobByJob = self.am_getOption("JobByJob", True)
self.throttlingPeriod = self.am_getOption("ThrottlingPeriod", 0.0)
return S_OK()
def __getAllowedJobTypes(self):
# Get valid jobTypes
result = self.jobDB.getDistinctJobAttributes("JobType")
if not result["OK"]:
return result
cleanJobTypes = []
for jobType in result["Value"]:
if jobType not in self.prod_types:
cleanJobTypes.append(jobType)
self.log.notice("JobTypes to clean %s" % cleanJobTypes)
return S_OK(cleanJobTypes)
#############################################################################
def execute(self):
"""The PilotAgent execution method.
"""
# Delete jobs in "Deleted" state
result = self.removeJobsByStatus({"Status": "Deleted"})
if not result["OK"]:
return result
# Get all the Job types that can be cleaned
result = self.__getAllowedJobTypes()
if not result["OK"]:
return result
baseCond = {"JobType": result["Value"]}
# Remove jobs with final status
for status in REMOVE_STATUS_DELAY:
delay = REMOVE_STATUS_DELAY[status]
condDict = dict(baseCond)
condDict["Status"] = status
delTime = str(Time.dateTime() - delay * Time.day)
result = self.removeJobsByStatus(condDict, delTime)
if not result["OK"]:
gLogger.warn("Failed to remove jobs in status %s" % status)
return S_OK()
def removeJobsByStatus(self, condDict, delay=False):
""" Remove deleted jobs
"""
if delay:
gLogger.verbose("Removing jobs with %s and older than %s" % (condDict, delay))
result = self.jobDB.selectJobs(condDict, older=delay)
else:
gLogger.verbose("Removing jobs with %s " % condDict)
result = self.jobDB.selectJobs(condDict)
if not result["OK"]:
return result
jobList = result["Value"]
if len(jobList) > self.maxJobsAtOnce:
jobList = jobList[: self.maxJobsAtOnce]
if not jobList:
return S_OK()
self.log.notice("Deleting %s jobs for %s" % (len(jobList), condDict))
count = 0
error_count = 0
result = SandboxStoreClient(useCertificates=True).unassignJobs(jobList)
if not result["OK"]:
gLogger.warn("Cannot unassign jobs to sandboxes", result["Message"])
if self.jobByJob:
for jobID in jobList:
resultJobDB = self.jobDB.removeJobFromDB(jobID)
resultTQ = self.taskQueueDB.deleteJob(jobID)
if not resultJobDB["OK"]:
gLogger.warn("Failed to remove job %d from JobDB" % jobID, result["Message"])
error_count += 1
elif not resultTQ["OK"]:
gLogger.warn("Failed to remove job %d from TaskQueueDB" % jobID, result["Message"])
error_count += 1
else:
count += 1
if self.throttlingPeriod:
time.sleep(self.throttlingPeriod)
else:
result = self.jobDB.removeJobFromDB(jobList)
if not result["OK"]:
gLogger.error("Failed to delete %d jobs from JobDB" % len(jobList))
else:
gLogger.info("Deleted %d jobs from JobDB" % len(jobList))
for jobID in jobList:
resultTQ = self.taskQueueDB.deleteJob(jobID)
if not resultTQ["OK"]:
gLogger.warn("Failed to remove job %d from TaskQueueDB" % jobID, resultTQ["Message"])
error_count += 1
else:
count += 1
if count > 0 or error_count > 0:
gLogger.info("Deleted %d jobs from JobDB, %d errors" % (count, error_count))
return S_OK()
class JobCleaningAgent( AgentModule ):
"""
The specific agents must provide the following methods:
* initialize() for initial settings
* beginExecution()
* execute() - the main method called in the agent cycle
* endExecution()
* finalize() - the graceful exit of the method, this one is usually used for the agent restart
"""
def __init__( self, *args, **kwargs ):
""" c'tor
"""
AgentModule.__init__( self, *args, **kwargs )
#clients
# FIXME: shouldn't we avoid using the DBs directly, and instead go through the service?
self.jobDB = None
self.taskQueueDB = None
self.jobLoggingDB = None
self.maxJobsAtOnce = 100
self.jobByJob = False
self.throttlingPeriod = 0.
self.prodTypes = []
self.removeStatusDelay = {}
#############################################################################
def initialize( self ):
""" Sets defaults
"""
self.am_setOption( "PollingTime", 120 )
self.jobDB = JobDB()
self.taskQueueDB = TaskQueueDB()
self.jobLoggingDB = JobLoggingDB()
# self.sandboxDB = SandboxDB( 'SandboxDB' )
agentTSTypes = self.am_getOption('ProductionTypes', [])
if agentTSTypes:
self.prodTypes = agentTSTypes
else:
self.prodTypes = Operations().getValue(
'Transformations/DataProcessing', ['MCSimulation', 'Merge'])
gLogger.info("Will exclude the following Production types from cleaning %s" % (
', '.join(self.prodTypes)))
self.maxJobsAtOnce = self.am_getOption( 'MaxJobsAtOnce', 500 )
self.jobByJob = self.am_getOption( 'JobByJob', False )
self.throttlingPeriod = self.am_getOption('ThrottlingPeriod', 0.)
self.removeStatusDelay['Done'] = self.am_getOption( 'RemoveStatusDelay/Done', 7 )
self.removeStatusDelay['Killed'] = self.am_getOption( 'RemoveStatusDelay/Killed', 7 )
self.removeStatusDelay['Failed'] = self.am_getOption( 'RemoveStatusDelay/Failed', 7 )
self.removeStatusDelay['Any'] = self.am_getOption( 'RemoveStatusDelay/Any', -1 )
return S_OK()
def __getAllowedJobTypes( self ):
""" Get valid jobTypes
"""
result = self.jobDB.getDistinctJobAttributes( 'JobType' )
if not result[ 'OK' ]:
return result
cleanJobTypes = []
for jobType in result[ 'Value' ]:
if jobType not in self.prodTypes:
cleanJobTypes.append( jobType )
self.log.notice( "JobTypes to clean %s" % cleanJobTypes )
return S_OK( cleanJobTypes )
#############################################################################
def execute( self ):
""" Remove jobs in various status
"""
#Delete jobs in "Deleted" state
result = self.removeJobsByStatus( { 'Status' : 'Deleted' } )
if not result[ 'OK' ]:
return result
#Get all the Job types that can be cleaned
result = self.__getAllowedJobTypes()
if not result[ 'OK' ]:
return result
# No jobs in the system subject to removal
if not result['Value']:
return S_OK()
baseCond = { 'JobType' : result[ 'Value' ] }
# Remove jobs with final status
for status in self.removeStatusDelay:
delay = self.removeStatusDelay[ status ]
if delay < 0:
# Negative delay means don't delete anything...
continue
condDict = dict( baseCond )
if status != 'Any':
condDict[ 'Status' ] = status
delTime = str( Time.dateTime() - delay * Time.day )
result = self.removeJobsByStatus( condDict, delTime )
if not result['OK']:
gLogger.warn( 'Failed to remove jobs in status %s' % status )
return S_OK()
def removeJobsByStatus( self, condDict, delay = False ):
""" Remove deleted jobs
"""
if delay:
gLogger.verbose( "Removing jobs with %s and older than %s day(s)" % ( condDict, delay ) )
result = self.jobDB.selectJobs( condDict, older = delay, limit = self.maxJobsAtOnce )
else:
gLogger.verbose( "Removing jobs with %s " % condDict )
result = self.jobDB.selectJobs( condDict, limit = self.maxJobsAtOnce )
if not result['OK']:
return result
jobList = result['Value']
if len(jobList) > self.maxJobsAtOnce:
jobList = jobList[:self.maxJobsAtOnce]
if not jobList:
return S_OK()
self.log.notice( "Deleting %s jobs for %s" % ( len( jobList ), condDict ) )
count = 0
error_count = 0
result = SandboxStoreClient( useCertificates = True ).unassignJobs( jobList )
if not result[ 'OK' ]:
gLogger.error("Cannot unassign jobs to sandboxes", result['Message'])
return result
result = self.deleteJobOversizedSandbox(jobList)
if not result[ 'OK' ]:
gLogger.error(
"Cannot schedule removal of oversized sandboxes", result['Message'])
return result
failedJobs = result['Value']['Failed']
for job in failedJobs:
jobList.pop(jobList.index(job))
# TODO: we should not remove a job if it still has requests in the RequestManager.
# But this logic should go in the client or in the service, and right now no service expose jobDB.removeJobFromDB
if self.jobByJob:
for jobID in jobList:
resultJobDB = self.jobDB.removeJobFromDB( jobID )
resultTQ = self.taskQueueDB.deleteJob( jobID )
resultLogDB = self.jobLoggingDB.deleteJob( jobID )
errorFlag = False
if not resultJobDB['OK']:
gLogger.warn( 'Failed to remove job %d from JobDB' % jobID, result['Message'] )
errorFlag = True
if not resultTQ['OK']:
gLogger.warn( 'Failed to remove job %d from TaskQueueDB' % jobID, result['Message'] )
errorFlag = True
if not resultLogDB['OK']:
gLogger.warn( 'Failed to remove job %d from JobLoggingDB' % jobID, result['Message'] )
errorFlag = True
if errorFlag:
error_count += 1
else:
count += 1
if self.throttlingPeriod:
time.sleep(self.throttlingPeriod)
else:
result = self.jobDB.removeJobFromDB( jobList )
if not result['OK']:
gLogger.error('Failed to delete %d jobs from JobDB' % len(jobList) )
else:
gLogger.info('Deleted %d jobs from JobDB' % len(jobList) )
for jobID in jobList:
resultTQ = self.taskQueueDB.deleteJob( jobID )
if not resultTQ['OK']:
gLogger.warn( 'Failed to remove job %d from TaskQueueDB' % jobID, resultTQ['Message'] )
error_count += 1
else:
count += 1
result = self.jobLoggingDB.deleteJob( jobList )
if not result['OK']:
gLogger.error('Failed to delete %d jobs from JobLoggingDB' % len(jobList) )
else:
gLogger.info('Deleted %d jobs from JobLoggingDB' % len(jobList) )
if count > 0 or error_count > 0 :
gLogger.info( 'Deleted %d jobs from JobDB, %d errors' % ( count, error_count ) )
return S_OK()
def deleteJobOversizedSandbox( self, jobIDList ):
""" Delete the job oversized sandbox files from storage elements
"""
failed = {}
successful = {}
lfnDict = {}
for jobID in jobIDList:
result = self.jobDB.getJobParameter( jobID, 'OutputSandboxLFN' )
if result['OK']:
lfn = result['Value']
if lfn:
lfnDict[lfn] = jobID
else:
successful[jobID] = 'No oversized sandbox found'
else:
gLogger.warn( 'Error interrogating JobDB: %s' % result['Message'] )
if not lfnDict:
return S_OK({'Successful': successful, 'Failed': failed})
# Schedule removal of the LFNs now
for lfn,jobID in lfnDict.items():
result = self.jobDB.getJobAttributes( jobID, ['OwnerDN', 'OwnerGroup'] )
if not result['OK']:
failed[jobID] = lfn
continue
if not result['Value']:
failed[jobID] = lfn
continue
ownerDN = result['Value']['OwnerDN']
ownerGroup = result['Value']['OwnerGroup']
result = self.__setRemovalRequest( lfn, ownerDN, ownerGroup )
if not result['OK']:
failed[jobID] = lfn
else:
successful[jobID] = lfn
result = {'Successful':successful, 'Failed':failed}
return S_OK(result)
def __setRemovalRequest( self, lfn, ownerDN, ownerGroup ):
""" Set removal request with the given credentials
"""
oRequest = Request()
oRequest.OwnerDN = ownerDN
oRequest.OwnerGroup = ownerGroup
oRequest.RequestName = os.path.basename( lfn ).strip() + '_removal_request.xml'
oRequest.SourceComponent = 'JobCleaningAgent'
removeFile = Operation()
removeFile.Type = 'RemoveFile'
removedFile = File()
removedFile.LFN = lfn
removeFile.addFile( removedFile )
oRequest.addOperation( removeFile )
return ReqClient().putRequest( oRequest )
# print db.checkImageStatus( 'name', 'flavor'*10, 'requirements' ) ret = db.insertInstance( 'Image3', 'instance' ) print "insertInstance ", ret ret = db.insertInstance( 'Image2', 'instance' ) print "insertInstance ", ret if not ret['OK']: DIRAC.exit() print type( ret['Value'] ) print "declareInstanceSubmitted", db.declareInstanceSubmitted( ret['Value'] ) id1 = DIRAC.Time.toString() print "declareInstanceRunning ", db.declareInstanceRunning( 'Image3', id1, 'IP', 'ip' ) id2 = DIRAC.Time.toString() print "declareInstanceRunning ", db.declareInstanceRunning( 'Image2', id2, 'IP', 'ip' ) print "declareInstanceRunning ", db.instanceIDHeartBeat( id2, 1.0 ) for status in validStates: print "get%10sInstances " % status, db.getInstancesByStatus( status ) print "declareInstanceHalting ", db.declareInstanceHalting( id1, 0.0 ) print "declareInstanceHalting ", db.declareInstanceHalting( id2, 0.0 ) print "declareStalledInstances ", db.declareStalledInstances() print "declareStalledInstances ", db.declareStalledInstances() from DIRAC.WorkloadManagementSystem.DB.TaskQueueDB import TaskQueueDB tq = TaskQueueDB() print tq.retrieveTaskQueues()
