def submitJob( self, jdl ):
""" Submit one job specified by its JDL to WMS
"""
if os.path.exists( jdl ):
fic = open ( jdl, "r" )
jdlString = fic.read()
fic.close()
else:
# If file JDL does not exist, assume that the JDL is passed as a string
jdlString = jdl
# Check the validity of the input JDL
jdlString = jdlString.strip()
if jdlString.find( "[" ) != 0:
jdlString = "[%s]" % jdlString
classAdJob = ClassAd( jdlString )
if not classAdJob.isOK():
return S_ERROR( 'Invalid job JDL' )
# Check the size and the contents of the input sandbox
result = self.__uploadInputSandbox( classAdJob )
if not result['OK']:
return result
# Submit the job now and get the new job ID
if not self.jobManager:
self.jobManager = RPCClient( 'WorkloadManagement/JobManager',
useCertificates = self.useCertificates,
timeout = self.timeout )
result = self.jobManager.submitJob( classAdJob.asJDL() )
if 'requireProxyUpload' in result and result['requireProxyUpload']:
gLogger.warn( "Need to upload the proxy" )
return result
def getJobDefinition(self, job, jobDef=False):
""" Retrieve JDL of the Job and return jobDef dictionary
"""
if not jobDef:
jobDef = {}
# If not jdl in jobinfo load it
if 'jdl' not in jobDef:
if self.requiredJobInfo == 'jdlOriginal':
result = self.jobDB.getJobJDL(job, original=True)
if not result['OK']:
self.log.error("No JDL for job", "%s" % job)
return S_ERROR("No JDL for job")
jobDef['jdl'] = result['Value']
if self.requiredJobInfo == 'jdl':
result = self.jobDB.getJobJDL(job)
if not result['OK']:
self.log.error("No JDL for job", "%s" % job)
return S_ERROR("No JDL for job")
jobDef['jdl'] = result['Value']
# Load the classad if needed
if 'jdl' in jobDef and 'classad' not in jobDef:
try:
classad = ClassAd(jobDef['jdl'])
except BaseException:
self.log.debug("Cannot load JDL")
return S_ERROR('Illegal Job JDL')
if not classad.isOK():
self.log.debug("Warning: illegal JDL for job %s, will be marked problematic" % (job))
return S_ERROR('Illegal Job JDL')
jobDef['classad'] = classad
return S_OK(jobDef)
def getJobDefinition(self, job, jobDef=False):
"""Retrieve JDL of the Job and return jobDef dictionary"""
if not jobDef:
jobDef = {}
# If not jdl in jobinfo load it
if "jdl" not in jobDef:
if self.requiredJobInfo == "jdlOriginal":
result = self.jobDB.getJobJDL(job, original=True)
if not result["OK"]:
self.log.error("No JDL for job", "%s" % job)
return S_ERROR("No JDL for job")
jobDef["jdl"] = result["Value"]
if self.requiredJobInfo == "jdl":
result = self.jobDB.getJobJDL(job)
if not result["OK"]:
self.log.error("No JDL for job", "%s" % job)
return S_ERROR("No JDL for job")
jobDef["jdl"] = result["Value"]
# Load the classad if needed
if "jdl" in jobDef and "classad" not in jobDef:
try:
classad = ClassAd(jobDef["jdl"])
except Exception:
self.log.debug("Cannot load JDL")
return S_ERROR(JobMinorStatus.ILLEGAL_JOB_JDL)
if not classad.isOK():
self.log.debug(
"Warning: illegal JDL for job %s, will be marked problematic"
% (job))
return S_ERROR(JobMinorStatus.ILLEGAL_JOB_JDL)
jobDef["classad"] = classad
return S_OK(jobDef)
def getJobDefinition(self, job, jobDef=False):
""" Retrieve JDL of the Job and return jobDef dictionary
"""
if not jobDef:
jobDef = {}
# If not jdl in jobinfo load it
if 'jdl' not in jobDef:
if self.requiredJobInfo == 'jdlOriginal':
result = self.jobDB.getJobJDL(job, original=True)
if not result['OK']:
self.log.error("No JDL for job", "%s" % job)
return S_ERROR("No JDL for job")
jobDef['jdl'] = result['Value']
if self.requiredJobInfo == 'jdl':
result = self.jobDB.getJobJDL(job)
if not result['OK']:
self.log.error("No JDL for job", "%s" % job)
return S_ERROR("No JDL for job")
jobDef['jdl'] = result['Value']
# Load the classad if needed
if 'jdl' in jobDef and 'classad' not in jobDef:
try:
classad = ClassAd(jobDef['jdl'])
except BaseException:
self.log.debug("Cannot load JDL")
return S_ERROR('Illegal Job JDL')
if not classad.isOK():
self.log.debug(
"Warning: illegal JDL for job %s, will be marked problematic"
% (job))
return S_ERROR('Illegal Job JDL')
jobDef['classad'] = classad
return S_OK(jobDef)
def submitJob(self, jdl):
""" Submit one job specified by its JDL to WMS
"""
if os.path.exists(jdl):
fic = open(jdl, "r")
jdlString = fic.read()
fic.close()
else:
# If file JDL does not exist, assume that the JDL is passed as a string
jdlString = jdl
# Check the validity of the input JDL
jdlString = jdlString.strip()
if jdlString.find("[") != 0:
jdlString = "[%s]" % jdlString
classAdJob = ClassAd(jdlString)
if not classAdJob.isOK():
return S_ERROR('Invalid job JDL')
# Check the size and the contents of the input sandbox
result = self.__uploadInputSandbox(classAdJob)
if not result['OK']:
return result
# Submit the job now and get the new job ID
if not self.jobManager:
self.jobManager = RPCClient('WorkloadManagement/JobManager',
useCertificates=self.useCertificates,
timeout=self.timeout)
result = self.jobManager.submitJob(classAdJob.asJDL())
if 'requireProxyUpload' in result and result['requireProxyUpload']:
gLogger.warn("Need to upload the proxy")
return result
def __processResourceDescription(self, resourceDescription):
# Check and form the resource description dictionary
resourceDict = {}
if type(resourceDescription) in StringTypes:
classAdAgent = ClassAd(resourceDescription)
if not classAdAgent.isOK():
return S_ERROR('Illegal Resource JDL')
gLogger.verbose(classAdAgent.asJDL())
for name in gTaskQueueDB.getSingleValueTQDefFields():
if classAdAgent.lookupAttribute(name):
if name == 'CPUTime':
resourceDict[name] = classAdAgent.getAttributeInt(name)
else:
resourceDict[name] = classAdAgent.getAttributeString(
name)
for name in gTaskQueueDB.getMultiValueMatchFields():
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 gTaskQueueDB.getSingleValueTQDefFields():
if resourceDescription.has_key(name):
resourceDict[name] = resourceDescription[name]
for name in gTaskQueueDB.getMultiValueMatchFields():
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',
'PilotInfoReportedFlag', 'PilotBenchmark',
'LHCbPlatform'):
if k in resourceDescription:
resourceDict[k] = resourceDescription[k]
return resourceDict
def __processResourceDescription(self, resourceDescription):
# Check and form the resource description dictionary
resourceDict = {}
if type(resourceDescription) in StringTypes:
classAdAgent = ClassAd(resourceDescription)
if not classAdAgent.isOK():
return S_ERROR('Illegal Resource JDL')
gLogger.verbose(classAdAgent.asJDL())
for name in gTaskQueueDB.getSingleValueTQDefFields():
if classAdAgent.lookupAttribute(name):
if name == 'CPUTime':
resourceDict[name] = classAdAgent.getAttributeInt(name)
else:
resourceDict[name] = classAdAgent.getAttributeString(
name)
for name in gTaskQueueDB.getMultiValueMatchFields():
if classAdAgent.lookupAttribute(name):
resourceDict[name] = classAdAgent.getAttributeString(name)
# Check if a JobID is requested
if classAdAgent.lookupAttribute('JobID'):
resourceDict['JobID'] = classAdAgent.getAttributeInt('JobID')
if classAdAgent.lookupAttribute('DIRACVersion'):
resourceDict['DIRACVersion'] = classAdAgent.getAttributeString(
'DIRACVersion')
if classAdAgent.lookupAttribute('VirtualOrganization'):
resourceDict[
'VirtualOrganization'] = classAdAgent.getAttributeString(
'VirtualOrganization')
else:
for name in gTaskQueueDB.getSingleValueTQDefFields():
if resourceDescription.has_key(name):
resourceDict[name] = resourceDescription[name]
for name in gTaskQueueDB.getMultiValueMatchFields():
if resourceDescription.has_key(name):
resourceDict[name] = resourceDescription[name]
if resourceDescription.has_key('JobID'):
resourceDict['JobID'] = resourceDescription['JobID']
if resourceDescription.has_key('DIRACVersion'):
resourceDict['DIRACVersion'] = resourceDescription[
'DIRACVersion']
if resourceDescription.has_key('VirtualOrganization'):
resourceDict['VirtualOrganization'] = resourceDescription[
'VirtualOrganization']
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 _processResourceDescription( self, resourceDescription ):
""" Check and form the resource description dictionary
resourceDescription is a ceDict coming from a JobAgent, for example.
"""
resourceDict = {}
if type( resourceDescription ) in StringTypes:
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 __processResourceDescription( self, resourceDescription ):
# Check and form the resource description dictionary
resourceDict = {}
if type( resourceDescription ) in StringTypes:
classAdAgent = ClassAd( resourceDescription )
if not classAdAgent.isOK():
return S_ERROR( 'Illegal Resource JDL' )
gLogger.verbose( classAdAgent.asJDL() )
for name in gTaskQueueDB.getSingleValueTQDefFields():
if classAdAgent.lookupAttribute( name ):
if name == 'CPUTime':
resourceDict[name] = classAdAgent.getAttributeInt( name )
else:
resourceDict[name] = classAdAgent.getAttributeString( name )
for name in gTaskQueueDB.getMultiValueMatchFields():
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 gTaskQueueDB.getSingleValueTQDefFields():
if resourceDescription.has_key( name ):
resourceDict[name] = resourceDescription[name]
for name in gTaskQueueDB.getMultiValueMatchFields():
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', 'PilotInfoReportedFlag', 'PilotBenchmark' ):
if k in resourceDescription:
resourceDict[ k ] = resourceDescription[ k ]
return resourceDict
def __processResourceDescription( self, resourceDescription ):
# Check and form the resource description dictionary
resourceDict = {}
if type( resourceDescription ) in StringTypes:
classAdAgent = ClassAd( resourceDescription )
if not classAdAgent.isOK():
return S_ERROR( 'Illegal Resource JDL' )
gLogger.verbose( classAdAgent.asJDL() )
for name in gTaskQueueDB.getSingleValueTQDefFields():
if classAdAgent.lookupAttribute( name ):
if name == 'CPUTime':
resourceDict[name] = classAdAgent.getAttributeInt( name )
else:
resourceDict[name] = classAdAgent.getAttributeString( name )
for name in gTaskQueueDB.getMultiValueMatchFields():
if classAdAgent.lookupAttribute( name ):
resourceDict[name] = classAdAgent.getAttributeString( name )
# Check if a JobID is requested
if classAdAgent.lookupAttribute( 'JobID' ):
resourceDict['JobID'] = classAdAgent.getAttributeInt( 'JobID' )
if classAdAgent.lookupAttribute( 'DIRACVersion' ):
resourceDict['DIRACVersion'] = classAdAgent.getAttributeString( 'DIRACVersion' )
if classAdAgent.lookupAttribute( 'VirtualOrganization' ):
resourceDict['VirtualOrganization'] = classAdAgent.getAttributeString( 'VirtualOrganization' )
else:
for name in gTaskQueueDB.getSingleValueTQDefFields():
if resourceDescription.has_key( name ):
resourceDict[name] = resourceDescription[name]
for name in gTaskQueueDB.getMultiValueMatchFields():
if resourceDescription.has_key( name ):
resourceDict[name] = resourceDescription[name]
if resourceDescription.has_key( 'JobID' ):
resourceDict['JobID'] = resourceDescription['JobID']
if resourceDescription.has_key( 'DIRACVersion' ):
resourceDict['DIRACVersion'] = resourceDescription['DIRACVersion']
if resourceDescription.has_key( 'VirtualOrganization' ):
resourceDict['VirtualOrganization'] = resourceDescription['VirtualOrganization']
return resourceDict
def submitJob(self, jdl):
""" Submit one job specified by its JDL to WMS
"""
if not self.jobManagerClient:
jobManager = RPCClient(
"WorkloadManagement/JobManager", useCertificates=self.useCertificates, timeout=self.timeout
)
else:
jobManager = self.jobManagerClient
if os.path.exists(jdl):
fic = open(jdl, "r")
jdlString = fic.read()
fic.close()
else:
# If file JDL does not exist, assume that the JDL is
# passed as a string
jdlString = jdl
# Check the validity of the input JDL
jdlString = jdlString.strip()
if jdlString.find("[") != 0:
jdlString = "[%s]" % jdlString
classAdJob = ClassAd(jdlString)
if not classAdJob.isOK():
return S_ERROR("Invalid job JDL")
# Check the size and the contents of the input sandbox
result = self.__uploadInputSandbox(classAdJob)
if not result["OK"]:
return result
# Submit the job now and get the new job ID
result = jobManager.submitJob(classAdJob.asJDL())
if not result["OK"]:
return result
jobID = result["Value"]
if "requireProxyUpload" in result and result["requireProxyUpload"]:
gProxyManager.uploadProxy()
# print "Sandbox uploading"
return S_OK(jobID)
def submitJob(self, jdl, jobDescriptionObject=None):
""" Submit one job specified by its JDL to WMS
"""
if os.path.exists(jdl):
fic = open(jdl, "r")
jdlString = fic.read()
fic.close()
else:
# If file JDL does not exist, assume that the JDL is passed as a string
jdlString = jdl
jdlString = jdlString.strip()
# Strip of comments in the jdl string
newJdlList = []
for line in jdlString.split('\n'):
if not line.strip().startswith('#'):
newJdlList.append(line)
jdlString = '\n'.join(newJdlList)
# Check the validity of the input JDL
if jdlString.find("[") != 0:
jdlString = "[%s]" % jdlString
classAdJob = ClassAd(jdlString)
if not classAdJob.isOK():
return S_ERROR('Invalid job JDL')
# Check the size and the contents of the input sandbox
result = self.__uploadInputSandbox(classAdJob, jobDescriptionObject)
if not result['OK']:
return result
# Submit the job now and get the new job ID
if not self.jobManager:
self.jobManager = RPCClient('WorkloadManagement/JobManager',
useCertificates=self.useCertificates,
timeout=self.timeout)
result = self.jobManager.submitJob(classAdJob.asJDL())
if 'requireProxyUpload' in result and result['requireProxyUpload']:
gLogger.warn("Need to upload the proxy")
return result
def submitJob( self, jdl, jobDescriptionObject = None ):
""" Submit one job specified by its JDL to WMS
"""
if os.path.exists( jdl ):
fic = open ( jdl, "r" )
jdlString = fic.read()
fic.close()
else:
# If file JDL does not exist, assume that the JDL is passed as a string
jdlString = jdl
jdlString = jdlString.strip()
# Strip of comments in the jdl string
newJdlList = []
for line in jdlString.split('\n'):
if not line.strip().startswith( '#' ):
newJdlList.append( line )
jdlString = '\n'.join( newJdlList )
# Check the validity of the input JDL
if jdlString.find( "[" ) != 0:
jdlString = "[%s]" % jdlString
classAdJob = ClassAd( jdlString )
if not classAdJob.isOK():
return S_ERROR( 'Invalid job JDL' )
# Check the size and the contents of the input sandbox
result = self.__uploadInputSandbox( classAdJob, jobDescriptionObject )
if not result['OK']:
return result
# Submit the job now and get the new job ID
if not self.jobManager:
self.jobManager = RPCClient( 'WorkloadManagement/JobManager',
useCertificates = self.useCertificates,
timeout = self.timeout )
result = self.jobManager.submitJob( classAdJob.asJDL() )
if 'requireProxyUpload' in result and result['requireProxyUpload']:
gLogger.warn( "Need to upload the proxy" )
return result
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 'JobID' in resourceDescription:
resourceDict['JobID'] = resourceDescription['JobID']
# Convert MaxRAM and NumberOfCores parameters into a list of tags
maxRAM = resourceDescription.get('MaxRAM')
nCores = resourceDescription.get('NumberOfProcessors')
for param, key in [(maxRAM, 'GB'), (nCores, 'Cores')]:
if param:
try:
intValue = int(param) / 1000
if intValue <= 128:
paramList = range(1, intValue + 1)
paramTags = [
'%d%s' % (par, key) for par in paramList
]
resourceDict.setdefault("Tag",
[]).extend(paramTags)
except ValueError:
pass
if 'Tag' in resourceDict:
resourceDict['Tag'] = list(set(resourceDict['Tag']))
for k in ('DIRACVersion', 'ReleaseVersion', 'ReleaseProject',
'VirtualOrganization', 'PilotReference',
'PilotBenchmark', 'PilotInfoReportedFlag'):
if k in resourceDescription:
resourceDict[k] = resourceDescription[k]
return resourceDict
def matchQueue(jobJDL, queueDict, fullMatch=False):
"""
Match the job description to the queue definition
:param str job: JDL job description
:param bool fullMatch: test matching on all the criteria
:param dict queueDict: queue parameters dictionary
:return: S_OK/S_ERROR, Value - result of matching, S_OK if matched or
S_ERROR with the reason for no match
"""
# Check the job description validity
job = ClassAd(jobJDL)
if not job.isOK():
return S_ERROR('Invalid job description')
noMatchReasons = []
# Check job requirements to resource
# 1. CPUTime
cpuTime = job.getAttributeInt('CPUTime')
if not cpuTime:
cpuTime = 84600
if cpuTime > queueDict.get('CPUTime', 0.):
noMatchReasons.append('Job CPUTime requirement not satisfied')
if not fullMatch:
return S_OK({'Match': False, 'Reason': noMatchReasons[0]})
# 2. Multi-value match requirements
for parameter in [
'Site', 'GridCE', 'Platform', 'GridMiddleware', 'PilotType',
'SubmitPool', 'JobType'
]:
if parameter in queueDict:
valueSet = set(job.getListFromExpression(parameter))
if not valueSet:
valueSet = set(job.getListFromExpression('%ss' % parameter))
queueSet = set(fromChar(queueDict[parameter]))
if valueSet and queueSet and not valueSet.intersection(queueSet):
valueToPrint = ','.join(valueSet)
if len(valueToPrint) > 20:
valueToPrint = "%s..." % valueToPrint[:20]
noMatchReasons.append('Job %s %s requirement not satisfied' %
(parameter, valueToPrint))
if not fullMatch:
return S_OK({'Match': False, 'Reason': noMatchReasons[0]})
# 3. Banned multi-value match requirements
for par in [
'Site', 'GridCE', 'Platform', 'GridMiddleware', 'PilotType',
'SubmitPool', 'JobType'
]:
parameter = "Banned%s" % par
if par in queueDict:
valueSet = set(job.getListFromExpression(parameter))
if not valueSet:
valueSet = set(job.getListFromExpression('%ss' % parameter))
queueSet = set(fromChar(queueDict[par]))
if valueSet and queueSet and valueSet.issubset(queueSet):
valueToPrint = ','.join(valueSet)
if len(valueToPrint) > 20:
valueToPrint = "%s..." % valueToPrint[:20]
noMatchReasons.append('Job %s %s requirement not satisfied' %
(parameter, valueToPrint))
if not fullMatch:
return S_OK({'Match': False, 'Reason': noMatchReasons[0]})
# 4. Tags
tags = set(job.getListFromExpression('Tag'))
nProc = job.getAttributeInt('NumberOfProcessors')
if nProc and nProc > 1:
tags.add('MultiProcessor')
wholeNode = job.getAttributeString('WholeNode')
if wholeNode:
tags.add('WholeNode')
queueTags = set(queueDict.get('Tags', []))
if not tags.issubset(queueTags):
noMatchReasons.append('Job Tag %s not satisfied' % ','.join(tags))
if not fullMatch:
return S_OK({'Match': False, 'Reason': noMatchReasons[0]})
# 4. MultiProcessor requirements
if nProc and nProc > int(queueDict.get('NumberOfProcessors', 1)):
noMatchReasons.append(
'Job NumberOfProcessors %d requirement not satisfied' % nProc)
if not fullMatch:
return S_OK({'Match': False, 'Reason': noMatchReasons[0]})
# 5. RAM
ram = job.getAttributeInt('RAM')
# If MaxRAM is not specified in the queue description, assume 2GB
if ram and ram > int(queueDict.get('MaxRAM', 2048)) / 1024:
noMatchReasons.append('Job RAM %d requirement not satisfied' % ram)
if not fullMatch:
return S_OK({'Match': False, 'Reason': noMatchReasons[0]})
# Check resource requirements to job
# 1. OwnerGroup - rare case but still
if "OwnerGroup" in queueDict:
result = getProxyInfo(disableVOMS=True)
if not result['OK']:
return S_ERROR('No valid proxy available')
ownerGroup = result['Value']['group']
if ownerGroup != queueDict['OwnerGroup']:
noMatchReasons.append(
'Resource OwnerGroup %s requirement not satisfied' %
queueDict['OwnerGroup'])
if not fullMatch:
return S_OK({'Match': False, 'Reason': noMatchReasons[0]})
# 2. Required tags
requiredTags = set(queueDict.get('RequiredTags', []))
if not requiredTags.issubset(tags):
noMatchReasons.append('Resource RequiredTags %s not satisfied' %
','.join(requiredTags))
if not fullMatch:
return S_OK({'Match': False, 'Reason': noMatchReasons[0]})
# 3. RunningLimit
site = queueDict['Site']
opsHelper = Operations()
result = opsHelper.getSections('JobScheduling/RunningLimit')
if result['OK'] and site in result['Value']:
result = opsHelper.getSections('JobScheduling/RunningLimit/%s' % site)
if result['OK']:
for parameter in result['Value']:
value = job.getAttributeString(parameter)
if value and opsHelper.getValue(
'JobScheduling/RunningLimit/%s/%s/%s' %
(site, parameter, value), 1) == 0:
noMatchReasons.append(
'Resource operational %s requirement not satisfied' %
parameter)
if not fullMatch:
return S_OK({
'Match': False,
'Reason': noMatchReasons[0]
})
return S_OK({'Match': not bool(noMatchReasons), 'Reason': noMatchReasons})
def selectJob( self, resourceDescription ):
""" Main job selection function to find the highest priority job
matching the resource capacity
"""
startTime = time.time()
# Check and form the resource description dictionary
resourceDict = {}
if type( resourceDescription ) in StringTypes:
classAdAgent = ClassAd( resourceDescription )
if not classAdAgent.isOK():
return S_ERROR( 'Illegal Resource JDL' )
gLogger.verbose( classAdAgent.asJDL() )
for name in taskQueueDB.getSingleValueTQDefFields():
if classAdAgent.lookupAttribute( name ):
if name == 'CPUTime':
resourceDict[name] = classAdAgent.getAttributeInt( name )
else:
resourceDict[name] = classAdAgent.getAttributeString( name )
for name in taskQueueDB.getMultiValueMatchFields():
if classAdAgent.lookupAttribute( name ):
resourceDict[name] = classAdAgent.getAttributeString( name )
# Check if a JobID is requested
if classAdAgent.lookupAttribute( 'JobID' ):
resourceDict['JobID'] = classAdAgent.getAttributeInt( 'JobID' )
if classAdAgent.lookupAttribute( 'DIRACVersion' ):
resourceDict['DIRACVersion'] = classAdAgent.getAttributeString( 'DIRACVersion' )
else:
for name in taskQueueDB.getSingleValueTQDefFields():
if resourceDescription.has_key( name ):
resourceDict[name] = resourceDescription[name]
for name in taskQueueDB.getMultiValueMatchFields():
if resourceDescription.has_key( name ):
resourceDict[name] = resourceDescription[name]
if resourceDescription.has_key( 'JobID' ):
resourceDict['JobID'] = resourceDescription['JobID']
if resourceDescription.has_key( 'DIRACVersion' ):
resourceDict['DIRACVersion'] = resourceDescription['DIRACVersion']
# Check the pilot DIRAC version
if self.checkPilotVersion:
if 'DIRACVersion' in resourceDict:
if self.pilotVersion and resourceDict['DIRACVersion'] != self.pilotVersion:
return S_ERROR( 'Pilot version does not match the production version %s:%s' % \
( resourceDict['DIRACVersion'], self.pilotVersion ) )
# Get common site mask and check the agent site
result = jobDB.getSiteMask( siteState = 'Active' )
if result['OK']:
maskList = result['Value']
else:
return S_ERROR( 'Internal error: can not get site mask' )
if not 'Site' in resourceDict:
return S_ERROR( 'Missing Site Name in Resource JDL' )
siteName = resourceDict['Site']
if resourceDict['Site'] not in maskList:
if 'GridCE' in resourceDict:
del resourceDict['Site']
else:
return S_ERROR( 'Site not in mask and GridCE not specified' )
resourceDict['Setup'] = self.serviceInfoDict['clientSetup']
if DEBUG:
print "Resource description:"
for k, v in resourceDict.items():
print k.rjust( 20 ), v
# Check if Job Limits are imposed onto the site
extraConditions = {}
if self.siteJobLimits:
result = self.getExtraConditions( siteName )
if result['OK']:
extraConditions = result['Value']
if extraConditions:
gLogger.info( 'Job Limits for site %s are: %s' % ( siteName, str( extraConditions ) ) )
result = taskQueueDB.matchAndGetJob( resourceDict, extraConditions = extraConditions )
if DEBUG:
print result
if not result['OK']:
return result
result = result['Value']
if not result['matchFound']:
return S_ERROR( 'No match found' )
jobID = result['jobId']
resAtt = jobDB.getJobAttributes( jobID, ['OwnerDN', 'OwnerGroup', 'Status'] )
if not resAtt['OK']:
return S_ERROR( 'Could not retrieve job attributes' )
if not resAtt['Value']:
return S_ERROR( 'No attributes returned for job' )
if not resAtt['Value']['Status'] == 'Waiting':
gLogger.error( 'Job %s matched by the TQ is not in Waiting state' % str( jobID ) )
result = taskQueueDB.deleteJob( jobID )
result = jobDB.setJobStatus( jobID, status = 'Matched', minor = 'Assigned' )
result = jobLoggingDB.addLoggingRecord( jobID,
status = 'Matched',
minor = 'Assigned',
source = 'Matcher' )
result = jobDB.getJobJDL( jobID )
if not result['OK']:
return S_ERROR( 'Failed to get the job JDL' )
resultDict = {}
resultDict['JDL'] = result['Value']
resultDict['JobID'] = jobID
matchTime = time.time() - startTime
gLogger.info( "Match time: [%s]" % str( matchTime ) )
gMonitor.addMark( "matchTime", matchTime )
# Get some extra stuff into the response returned
resOpt = jobDB.getJobOptParameters( jobID )
if resOpt['OK']:
for key, value in resOpt['Value'].items():
resultDict[key] = value
resAtt = jobDB.getJobAttributes( jobID, ['OwnerDN', 'OwnerGroup'] )
if not resAtt['OK']:
return S_ERROR( 'Could not retrieve job attributes' )
if not resAtt['Value']:
return S_ERROR( 'No attributes returned for job' )
resultDict['DN'] = resAtt['Value']['OwnerDN']
resultDict['Group'] = resAtt['Value']['OwnerGroup']
return S_OK( resultDict )
def submitJob(self, jdl, jobDescriptionObject=None):
"""Submit one job specified by its JDL to WMS.
The JDL may actually be the desciption of a parametric job,
resulting in multiple DIRAC jobs submitted to the DIRAC WMS
"""
if os.path.exists(jdl):
with open(jdl, "r") as fic:
jdlString = fic.read()
else:
# If file JDL does not exist, assume that the JDL is passed as a string
jdlString = jdl
jdlString = jdlString.strip()
gLogger.debug("Submitting JDL", jdlString)
# Strip of comments in the jdl string
newJdlList = []
for line in jdlString.split("\n"):
if not line.strip().startswith("#"):
newJdlList.append(line)
jdlString = "\n".join(newJdlList)
# Check the validity of the input JDL
if jdlString.find("[") != 0:
jdlString = "[%s]" % jdlString
classAdJob = ClassAd(jdlString)
if not classAdJob.isOK():
return S_ERROR(EWMSJDL, "Invalid job JDL")
# Check the size and the contents of the input sandbox
result = self.__uploadInputSandbox(classAdJob, jobDescriptionObject)
if not result["OK"]:
return result
# Submit the job now and get the new job ID
result = getParameterVectorLength(classAdJob)
if not result["OK"]:
return result
nJobs = result["Value"]
result = self.jobManager.submitJob(classAdJob.asJDL())
if nJobs:
gLogger.debug("Applying transactional job submission")
# The server applies transactional bulk submission, we should confirm the jobs
if result["OK"]:
jobIDList = result["Value"]
if len(jobIDList) == nJobs:
# Confirm the submitted jobs
confirmed = False
for _attempt in range(3):
result = self.jobManager.confirmBulkSubmission(jobIDList)
if result["OK"]:
confirmed = True
break
time.sleep(1)
if not confirmed:
# The bulk submission failed, try to remove the created jobs
resultDelete = self.jobManager.removeJob(jobIDList)
error = "Job submission failed to confirm bulk transaction"
if not resultDelete["OK"]:
error += "; removal of created jobs failed"
return S_ERROR(EWMSSUBM, error)
else:
return S_ERROR(EWMSSUBM, "The number of submitted jobs does not match job description")
if result.get("requireProxyUpload"):
gLogger.warn("Need to upload the proxy")
return result
def matchQueue(jobJDL, queueDict, fullMatch=False):
"""
Match the job description to the queue definition
:param str job: JDL job description
:param bool fullMatch: test matching on all the criteria
:param dict queueDict: queue parameters dictionary
:return: S_OK/S_ERROR, Value - result of matching, S_OK if matched or
S_ERROR with the reason for no match
"""
# Check the job description validity
job = ClassAd(jobJDL)
if not job.isOK():
return S_ERROR("Invalid job description")
noMatchReasons = []
# Check job requirements to resource
# 1. CPUTime
cpuTime = job.getAttributeInt("CPUTime")
if not cpuTime:
cpuTime = 84600
if cpuTime > int(queueDict.get("CPUTime", 0)):
noMatchReasons.append("Job CPUTime requirement not satisfied")
if not fullMatch:
return S_OK({"Match": False, "Reason": noMatchReasons[0]})
# 2. Multi-value match requirements
for parameter in ["Site", "GridCE", "Platform", "JobType"]:
if parameter in queueDict:
valueSet = set(job.getListFromExpression(parameter))
if not valueSet:
valueSet = set(job.getListFromExpression("%ss" % parameter))
queueSet = set(fromChar(queueDict[parameter]))
if valueSet and queueSet and not valueSet.intersection(queueSet):
valueToPrint = ",".join(valueSet)
if len(valueToPrint) > 20:
valueToPrint = "%s..." % valueToPrint[:20]
noMatchReasons.append("Job %s %s requirement not satisfied" % (parameter, valueToPrint))
if not fullMatch:
return S_OK({"Match": False, "Reason": noMatchReasons[0]})
# 3. Banned multi-value match requirements
for par in ["Site", "GridCE", "Platform", "JobType"]:
parameter = "Banned%s" % par
if par in queueDict:
valueSet = set(job.getListFromExpression(parameter))
if not valueSet:
valueSet = set(job.getListFromExpression("%ss" % parameter))
queueSet = set(fromChar(queueDict[par]))
if valueSet and queueSet and valueSet.issubset(queueSet):
valueToPrint = ",".join(valueSet)
if len(valueToPrint) > 20:
valueToPrint = "%s..." % valueToPrint[:20]
noMatchReasons.append("Job %s %s requirement not satisfied" % (parameter, valueToPrint))
if not fullMatch:
return S_OK({"Match": False, "Reason": noMatchReasons[0]})
# 4. Tags
tags = set(job.getListFromExpression("Tag"))
nProc = job.getAttributeInt("NumberOfProcessors")
if nProc and nProc > 1:
tags.add("MultiProcessor")
wholeNode = job.getAttributeString("WholeNode")
if wholeNode:
tags.add("WholeNode")
queueTags = set(queueDict.get("Tag", []))
if not tags.issubset(queueTags):
noMatchReasons.append("Job Tag %s not satisfied" % ",".join(tags))
if not fullMatch:
return S_OK({"Match": False, "Reason": noMatchReasons[0]})
# 4. MultiProcessor requirements
if nProc and nProc > int(queueDict.get("NumberOfProcessors", 1)):
noMatchReasons.append("Job NumberOfProcessors %d requirement not satisfied" % nProc)
if not fullMatch:
return S_OK({"Match": False, "Reason": noMatchReasons[0]})
# 5. RAM
ram = job.getAttributeInt("RAM")
# If MaxRAM is not specified in the queue description, assume 2GB
if ram and ram > int(queueDict.get("MaxRAM", 2048) / 1024):
noMatchReasons.append("Job RAM %d requirement not satisfied" % ram)
if not fullMatch:
return S_OK({"Match": False, "Reason": noMatchReasons[0]})
# Check resource requirements to job
# 1. OwnerGroup - rare case but still
if "OwnerGroup" in queueDict:
result = getProxyInfo(disableVOMS=True)
if not result["OK"]:
return S_ERROR("No valid proxy available")
ownerGroup = result["Value"]["group"]
if ownerGroup != queueDict["OwnerGroup"]:
noMatchReasons.append("Resource OwnerGroup %s requirement not satisfied" % queueDict["OwnerGroup"])
if not fullMatch:
return S_OK({"Match": False, "Reason": noMatchReasons[0]})
# 2. Required tags
requiredTags = set(queueDict.get("RequiredTags", []))
if not requiredTags.issubset(tags):
noMatchReasons.append("Resource RequiredTags %s not satisfied" % ",".join(requiredTags))
if not fullMatch:
return S_OK({"Match": False, "Reason": noMatchReasons[0]})
# 3. RunningLimit
site = queueDict["Site"]
ce = queueDict.get("GridCE")
opsHelper = Operations()
result = opsHelper.getSections("JobScheduling/RunningLimit")
if result["OK"] and site in result["Value"]:
result = opsHelper.getSections("JobScheduling/RunningLimit/%s" % site)
if result["OK"]:
for parameter in result["Value"]:
value = job.getAttributeString(parameter)
if (
value
and (
opsHelper.getValue("JobScheduling/RunningLimit/%s/%s/%s" % (site, parameter, value), 1)
or opsHelper.getValue(
"JobScheduling/RunningLimit/%s/CEs/%s/%s/%s" % (site, ce, parameter, value), 1
)
)
== 0
):
noMatchReasons.append("Resource operational %s requirement not satisfied" % parameter)
if not fullMatch:
return S_OK({"Match": False, "Reason": noMatchReasons[0]})
return S_OK({"Match": not bool(noMatchReasons), "Reason": noMatchReasons})
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 'JobID' in resourceDescription:
resourceDict['JobID'] = resourceDescription['JobID']
# Convert MaxRAM and NumberOfCores parameters into a list of tags
maxRAM = resourceDescription.get( 'MaxRAM' )
nCores = resourceDescription.get( 'NumberOfProcessors' )
for param, key in [ ( maxRAM, 'GB' ), ( nCores, 'Cores' ) ]:
if param:
try:
intValue = int( param )/1000
if intValue <= 128:
paramList = range( 1, intValue + 1 )
paramTags = [ '%d%s' % ( par, key ) for par in paramList ]
resourceDict.setdefault( "Tag", [] ).extend( paramTags )
except ValueError:
pass
if 'Tag' in resourceDict:
resourceDict['Tag'] = list( set( resourceDict['Tag'] ) )
for k in ( 'DIRACVersion', 'ReleaseVersion', 'ReleaseProject', 'VirtualOrganization',
'PilotReference', 'PilotBenchmark', 'PilotInfoReportedFlag' ):
if k in resourceDescription:
resourceDict[ k ] = resourceDescription[ k ]
return resourceDict
def selectJob(self, resourceDescription):
""" Main job selection function to find the highest priority job
matching the resource capacity
"""
startTime = time.time()
# Check and form the resource description dictionary
resourceDict = {}
if type(resourceDescription) in StringTypes:
classAdAgent = ClassAd(resourceDescription)
if not classAdAgent.isOK():
return S_ERROR("Illegal Resource JDL")
gLogger.verbose(classAdAgent.asJDL())
for name in taskQueueDB.getSingleValueTQDefFields():
if classAdAgent.lookupAttribute(name):
if name == "CPUTime":
resourceDict[name] = classAdAgent.getAttributeInt(name)
else:
resourceDict[name] = classAdAgent.getAttributeString(name)
for name in taskQueueDB.getMultiValueMatchFields():
if classAdAgent.lookupAttribute(name):
resourceDict[name] = classAdAgent.getAttributeString(name)
# Check if a JobID is requested
if classAdAgent.lookupAttribute("JobID"):
resourceDict["JobID"] = classAdAgent.getAttributeInt("JobID")
if classAdAgent.lookupAttribute("DIRACVersion"):
resourceDict["DIRACVersion"] = classAdAgent.getAttributeString("DIRACVersion")
else:
for name in taskQueueDB.getSingleValueTQDefFields():
if resourceDescription.has_key(name):
resourceDict[name] = resourceDescription[name]
for name in taskQueueDB.getMultiValueMatchFields():
if resourceDescription.has_key(name):
resourceDict[name] = resourceDescription[name]
if resourceDescription.has_key("JobID"):
resourceDict["JobID"] = resourceDescription["JobID"]
if resourceDescription.has_key("DIRACVersion"):
resourceDict["DIRACVersion"] = resourceDescription["DIRACVersion"]
# Check the pilot DIRAC version
if self.checkPilotVersion:
if "DIRACVersion" in resourceDict:
if self.pilotVersion and resourceDict["DIRACVersion"] != self.pilotVersion:
return S_ERROR(
"Pilot version does not match the production version %s:%s"
% (resourceDict["DIRACVersion"], self.pilotVersion)
)
# Get common site mask and check the agent site
result = jobDB.getSiteMask(siteState="Active")
if result["OK"]:
maskList = result["Value"]
else:
return S_ERROR("Internal error: can not get site mask")
if not "Site" in resourceDict:
return S_ERROR("Missing Site Name in Resource JDL")
siteName = resourceDict["Site"]
if resourceDict["Site"] not in maskList:
if "GridCE" in resourceDict:
del resourceDict["Site"]
else:
return S_ERROR("Site not in mask and GridCE not specified")
resourceDict["Setup"] = self.serviceInfoDict["clientSetup"]
if DEBUG:
print "Resource description:"
for k, v in resourceDict.items():
print k.rjust(20), v
# Check if Job Limits are imposed onto the site
extraConditions = {}
if self.siteJobLimits:
result = self.getExtraConditions(siteName)
if result["OK"]:
extraConditions = result["Value"]
if extraConditions:
gLogger.info("Job Limits for site %s are: %s" % (siteName, str(extraConditions)))
result = taskQueueDB.matchAndGetJob(resourceDict, extraConditions=extraConditions)
if DEBUG:
print result
if not result["OK"]:
return result
result = result["Value"]
if not result["matchFound"]:
return S_ERROR("No match found")
jobID = result["jobId"]
resAtt = jobDB.getJobAttributes(jobID, ["OwnerDN", "OwnerGroup", "Status"])
if not resAtt["OK"]:
return S_ERROR("Could not retrieve job attributes")
if not resAtt["Value"]:
return S_ERROR("No attributes returned for job")
if not resAtt["Value"]["Status"] == "Waiting":
gLogger.error("Job %s matched by the TQ is not in Waiting state" % str(jobID))
result = taskQueueDB.deleteJob(jobID)
result = jobDB.setJobStatus(jobID, status="Matched", minor="Assigned")
result = jobLoggingDB.addLoggingRecord(jobID, status="Matched", minor="Assigned", source="Matcher")
result = jobDB.getJobJDL(jobID)
if not result["OK"]:
return S_ERROR("Failed to get the job JDL")
resultDict = {}
resultDict["JDL"] = result["Value"]
resultDict["JobID"] = jobID
matchTime = time.time() - startTime
gLogger.info("Match time: [%s]" % str(matchTime))
gMonitor.addMark("matchTime", matchTime)
# Get some extra stuff into the response returned
resOpt = jobDB.getJobOptParameters(jobID)
if resOpt["OK"]:
for key, value in resOpt["Value"].items():
resultDict[key] = value
resAtt = jobDB.getJobAttributes(jobID, ["OwnerDN", "OwnerGroup"])
if not resAtt["OK"]:
return S_ERROR("Could not retrieve job attributes")
if not resAtt["Value"]:
return S_ERROR("No attributes returned for job")
resultDict["DN"] = resAtt["Value"]["OwnerDN"]
resultDict["Group"] = resAtt["Value"]["OwnerGroup"]
return S_OK(resultDict)
def selectJob( self, resourceDescription ):
""" Main job selection function to find the highest priority job
matching the resource capacity
"""
startTime = time.time()
# Check and form the resource description dictionary
resourceDict = {}
if type( resourceDescription ) in StringTypes:
classAdAgent = ClassAd( resourceDescription )
if not classAdAgent.isOK():
return S_ERROR( 'Illegal Resource JDL' )
gLogger.verbose( classAdAgent.asJDL() )
for name in gTaskQueueDB.getSingleValueTQDefFields():
if classAdAgent.lookupAttribute( name ):
if name == 'CPUTime':
resourceDict[name] = classAdAgent.getAttributeInt( name )
else:
resourceDict[name] = classAdAgent.getAttributeString( name )
for name in gTaskQueueDB.getMultiValueMatchFields():
if classAdAgent.lookupAttribute( name ):
resourceDict[name] = classAdAgent.getAttributeString( name )
# Check if a JobID is requested
if classAdAgent.lookupAttribute( 'JobID' ):
resourceDict['JobID'] = classAdAgent.getAttributeInt( 'JobID' )
if classAdAgent.lookupAttribute( 'DIRACVersion' ):
resourceDict['DIRACVersion'] = classAdAgent.getAttributeString( 'DIRACVersion' )
if classAdAgent.lookupAttribute( 'VirtualOrganization' ):
resourceDict['VirtualOrganization'] = classAdAgent.getAttributeString( 'VirtualOrganization' )
else:
for name in gTaskQueueDB.getSingleValueTQDefFields():
if resourceDescription.has_key( name ):
resourceDict[name] = resourceDescription[name]
for name in gTaskQueueDB.getMultiValueMatchFields():
if resourceDescription.has_key( name ):
resourceDict[name] = resourceDescription[name]
if resourceDescription.has_key( 'JobID' ):
resourceDict['JobID'] = resourceDescription['JobID']
if resourceDescription.has_key( 'DIRACVersion' ):
resourceDict['DIRACVersion'] = resourceDescription['DIRACVersion']
if resourceDescription.has_key( 'VirtualOrganization' ):
resourceDict['VirtualOrganization'] = resourceDescription['VirtualOrganization']
# Check the pilot DIRAC version
if self.checkPilotVersion:
if not 'DIRACVersion' in resourceDict:
return S_ERROR( 'Version check requested and not provided by Pilot' )
# Check if the matching Request provides a VirtualOrganization
if 'VirtualOrganization' in resourceDict:
voName = resourceDict['VirtualOrganization']
# Check if the matching Request provides an OwnerGroup
elif 'OwnerGroup' in resourceDict:
voName = getVOForGroup( resourceDict['OwnerGroup'] )
# else take the default VirtualOrganization for the installation
else:
voName = getVOForGroup( '' )
self.pilotVersion = gConfig.getValue( '/Operations/%s/%s/Versions/PilotVersion' % ( voName, self.setup ), '' )
if self.pilotVersion and resourceDict['DIRACVersion'] != self.pilotVersion:
return S_ERROR( 'Pilot version does not match the production version %s:%s' % \
( resourceDict['DIRACVersion'], self.pilotVersion ) )
# Get common site mask and check the agent site
result = gJobDB.getSiteMask( siteState = 'Active' )
if result['OK']:
maskList = result['Value']
else:
return S_ERROR( 'Internal error: can not get site mask' )
if not 'Site' in resourceDict:
return S_ERROR( 'Missing Site Name in Resource JDL' )
siteName = resourceDict['Site']
if resourceDict['Site'] not in maskList:
if 'GridCE' in resourceDict:
del resourceDict['Site']
else:
return S_ERROR( 'Site not in mask and GridCE not specified' )
resourceDict['Setup'] = self.serviceInfoDict['clientSetup']
if DEBUG:
print "Resource description:"
for key, value in resourceDict.items():
print key.rjust( 20 ), value
# Check if Job Limits are imposed onto the site
extraConditions = {}
if self.siteJobLimits:
result = self.getExtraConditions( siteName )
if result['OK']:
extraConditions = result['Value']
if extraConditions:
gLogger.info( 'Job Limits for site %s are: %s' % ( siteName, str( extraConditions ) ) )
result = gTaskQueueDB.matchAndGetJob( resourceDict, extraConditions = extraConditions )
if DEBUG:
print result
if not result['OK']:
return result
result = result['Value']
if not result['matchFound']:
return S_ERROR( 'No match found' )
jobID = result['jobId']
resAtt = gJobDB.getJobAttributes( jobID, ['OwnerDN', 'OwnerGroup', 'Status'] )
if not resAtt['OK']:
return S_ERROR( 'Could not retrieve job attributes' )
if not resAtt['Value']:
return S_ERROR( 'No attributes returned for job' )
if not resAtt['Value']['Status'] == 'Waiting':
gLogger.error( 'Job %s matched by the TQ is not in Waiting state' % str( jobID ) )
result = gTaskQueueDB.deleteJob( jobID )
result = gJobDB.setJobStatus( jobID, status = 'Matched', minor = 'Assigned' )
result = gJobLoggingDB.addLoggingRecord( jobID,
status = 'Matched',
minor = 'Assigned',
source = 'Matcher' )
result = gJobDB.getJobJDL( jobID )
if not result['OK']:
return S_ERROR( 'Failed to get the job JDL' )
resultDict = {}
resultDict['JDL'] = result['Value']
resultDict['JobID'] = jobID
matchTime = time.time() - startTime
gLogger.info( "Match time: [%s]" % str( matchTime ) )
gMonitor.addMark( "matchTime", matchTime )
# Get some extra stuff into the response returned
resOpt = gJobDB.getJobOptParameters( jobID )
if resOpt['OK']:
for key, value in resOpt['Value'].items():
resultDict[key] = value
resAtt = gJobDB.getJobAttributes( jobID, ['OwnerDN', 'OwnerGroup'] )
if not resAtt['OK']:
return S_ERROR( 'Could not retrieve job attributes' )
if not resAtt['Value']:
return S_ERROR( 'No attributes returned for job' )
resultDict['DN'] = resAtt['Value']['OwnerDN']
resultDict['Group'] = resAtt['Value']['OwnerGroup']
return S_OK( resultDict )
