def testC_Profile(self):
"""
_Profile_
DON'T RUN THIS!
"""
return
import cProfile, pstats
myThread = threading.currentThread()
name = makeUUID()
config = self.getConfig()
jobList = self.createGiantJobSet(name = name, config = config,
nSubs = 10, nJobs = 1000, nFiles = 10)
cleanCouch = CleanCouchPoller(config = config)
cleanCouch.setup()
cProfile.runctx("cleanCouch.algorithm()", globals(), locals(), filename = "testStats.stat")
p = pstats.Stats('testStats.stat')
p.sort_stats('cumulative')
p.print_stats()
return
def testC_Profile(self):
"""
_Profile_
DON'T RUN THIS!
"""
import cProfile
import pstats
name = makeUUID()
config = self.getConfig()
jobList = self.createGiantJobSet(name=name,
config=config,
nSubs=10,
nJobs=1000,
nFiles=10)
cleanCouch = CleanCouchPoller(config=config)
cleanCouch.setup()
cProfile.runctx("cleanCouch.algorithm()",
globals(),
locals(),
filename="testStats.stat")
p = pstats.Stats('testStats.stat')
p.sort_stats('cumulative')
p.print_stats()
return
def preInitialization(self):
print "TaskArchiver.preInitialization"
# Add event loop to worker manager
myThread = threading.currentThread()
pollInterval = self.config.TaskArchiver.pollInterval
logging.info("Setting task archiver poll interval to %s seconds" %
pollInterval)
myThread.workerThreadManager.addWorker(TaskArchiverPoller(self.config),
pollInterval)
couchInterval = self.config.TaskArchiver.cleanCouchInterval
logging.info(
"Setting poll interval for cleanup old couch doc to %s seconds" %
couchInterval)
myThread.workerThreadManager.addWorker(CleanCouchPoller(self.config),
couchInterval)
return
def testB_testErrors(self):
"""
_testErrors_
Test with a failed FWJR
"""
myThread = threading.currentThread()
config = self.getConfig()
workloadPath = os.path.join(self.testDir, 'specDir', 'spec.pkl')
workload = self.createWorkload(workloadName = workloadPath)
testJobGroup = self.createTestJobGroup(config = config,
name = workload.name(),
specLocation = workloadPath,
error = True)
# Create second workload
testJobGroup2 = self.createTestJobGroup(config = config,
name = workload.name(),
filesetName = "TestFileset_2",
specLocation = workloadPath,
task = "/TestWorkload/ReReco/LogCollect",
type = "LogCollect")
cachePath = os.path.join(config.JobCreator.jobCacheDir,
"TestWorkload", "ReReco")
os.makedirs(cachePath)
self.assertTrue(os.path.exists(cachePath))
couchdb = CouchServer(config.JobStateMachine.couchurl)
jobdb = couchdb.connectDatabase("%s/jobs" % self.databaseName)
fwjrdb = couchdb.connectDatabase("%s/fwjrs" % self.databaseName)
jobdb.loadView("JobDump", "jobsByWorkflowName",
options = {"startkey": [workload.name()],
"endkey": [workload.name(), {}]})['rows']
fwjrdb.loadView("FWJRDump", "fwjrsByWorkflowName",
options = {"startkey": [workload.name()],
"endkey": [workload.name(), {}]})['rows']
self.populateWorkflowWithCompleteStatus()
testTaskArchiver = TaskArchiverPoller(config = config)
testTaskArchiver.algorithm()
cleanCouch = CleanCouchPoller(config = config)
cleanCouch.setup()
cleanCouch.algorithm()
dbname = getattr(config.JobStateMachine, "couchDBName")
workdatabase = couchdb.connectDatabase("%s/workloadsummary" % dbname)
workloadSummary = workdatabase.document(id = workload.name())
self.assertEqual(workloadSummary['errors']['/TestWorkload/ReReco']['failureTime'], 500)
self.assertTrue('99999' in workloadSummary['errors']['/TestWorkload/ReReco']['cmsRun1'])
failedRunInfo = workloadSummary['errors']['/TestWorkload/ReReco']['cmsRun1']['99999']['runs']
self.assertEqual(failedRunInfo, {'10' : [[12312, 12312]]},
"Wrong lumi information in the summary for failed jobs")
# Check the failures by site histograms
self.assertEqual(workloadSummary['histograms']['workflowLevel']['failuresBySite']['data']['T1_IT_CNAF']['Failed Jobs'], 10)
self.assertEqual(workloadSummary['histograms']['stepLevel']['/TestWorkload/ReReco']['cmsRun1']['errorsBySite']['data']['T1_IT_CNAF']['99999'], 10)
self.assertEqual(workloadSummary['histograms']['stepLevel']['/TestWorkload/ReReco']['cmsRun1']['errorsBySite']['data']['T1_IT_CNAF']['8020'], 10)
self.assertEqual(workloadSummary['histograms']['workflowLevel']['failuresBySite']['average']['Failed Jobs'], 10)
self.assertEqual(workloadSummary['histograms']['stepLevel']['/TestWorkload/ReReco']['cmsRun1']['errorsBySite']['average']['99999'], 10)
self.assertEqual(workloadSummary['histograms']['stepLevel']['/TestWorkload/ReReco']['cmsRun1']['errorsBySite']['average']['8020'], 10)
self.assertEqual(workloadSummary['histograms']['workflowLevel']['failuresBySite']['stdDev']['Failed Jobs'], 0)
self.assertEqual(workloadSummary['histograms']['stepLevel']['/TestWorkload/ReReco']['cmsRun1']['errorsBySite']['stdDev']['99999'], 0)
self.assertEqual(workloadSummary['histograms']['stepLevel']['/TestWorkload/ReReco']['cmsRun1']['errorsBySite']['stdDev']['8020'], 0)
return
def testA_BasicFunctionTest(self):
"""
_BasicFunctionTest_
Tests the components, by seeing if they can process a simple set of closeouts
"""
myThread = threading.currentThread()
config = self.getConfig()
workloadPath = os.path.join(self.testDir, 'specDir', 'spec.pkl')
workload = self.createWorkload(workloadName = workloadPath)
testJobGroup = self.createTestJobGroup(config = config,
name = workload.name(),
specLocation = workloadPath,
error = False)
# Create second workload
testJobGroup2 = self.createTestJobGroup(config = config,
name = workload.name(),
filesetName = "TestFileset_2",
specLocation = workloadPath,
task = "/TestWorkload/ReReco/LogCollect",
type = "LogCollect")
cachePath = os.path.join(config.JobCreator.jobCacheDir,
"TestWorkload", "ReReco")
os.makedirs(cachePath)
self.assertTrue(os.path.exists(cachePath))
cachePath2 = os.path.join(config.JobCreator.jobCacheDir,
"TestWorkload", "LogCollect")
os.makedirs(cachePath2)
self.assertTrue(os.path.exists(cachePath2))
result = myThread.dbi.processData("SELECT * FROM wmbs_subscription")[0].fetchall()
self.assertEqual(len(result), 2)
workflowName = "TestWorkload"
dbname = config.TaskArchiver.workloadSummaryCouchDBName
couchdb = CouchServer(config.JobStateMachine.couchurl)
workdatabase = couchdb.connectDatabase(dbname)
jobdb = couchdb.connectDatabase("%s/jobs" % self.databaseName)
fwjrdb = couchdb.connectDatabase("%s/fwjrs" % self.databaseName)
jobs = jobdb.loadView("JobDump", "jobsByWorkflowName",
options = {"startkey": [workflowName],
"endkey": [workflowName, {}]})['rows']
fwjrdb.loadView("FWJRDump", "fwjrsByWorkflowName",
options = {"startkey": [workflowName],
"endkey": [workflowName, {}]})['rows']
self.assertEqual(len(jobs), 2*self.nJobs)
from WMCore.WMBS.CreateWMBSBase import CreateWMBSBase
create = CreateWMBSBase()
tables = []
for x in create.requiredTables:
tables.append(x[2:])
self.populateWorkflowWithCompleteStatus()
testTaskArchiver = TaskArchiverPoller(config = config)
testTaskArchiver.algorithm()
cleanCouch = CleanCouchPoller(config = config)
cleanCouch.setup()
cleanCouch.algorithm()
result = myThread.dbi.processData("SELECT * FROM wmbs_job")[0].fetchall()
self.assertEqual(len(result), 0)
result = myThread.dbi.processData("SELECT * FROM wmbs_subscription")[0].fetchall()
self.assertEqual(len(result), 0)
result = myThread.dbi.processData("SELECT * FROM wmbs_jobgroup")[0].fetchall()
self.assertEqual(len(result), 0)
result = myThread.dbi.processData("SELECT * FROM wmbs_fileset")[0].fetchall()
self.assertEqual(len(result), 0)
result = myThread.dbi.processData("SELECT * FROM wmbs_file_details")[0].fetchall()
self.assertEqual(len(result), 0)
# Make sure we deleted the directory
self.assertFalse(os.path.exists(cachePath))
self.assertFalse(os.path.exists(os.path.join(self.testDir, 'workloadTest/TestWorkload')))
testWMBSFileset = Fileset(id = 1)
self.assertEqual(testWMBSFileset.exists(), False)
workloadSummary = workdatabase.document(id = "TestWorkload")
# Check ACDC
self.assertEqual(workloadSummary['ACDCServer'], sanitizeURL(config.ACDC.couchurl)['url'])
# Check the output
self.assertEqual(workloadSummary['output'].keys(), ['/Electron/MorePenguins-v0/RECO'])
self.assertEqual(sorted(workloadSummary['output']['/Electron/MorePenguins-v0/RECO']['tasks']),
['/TestWorkload/ReReco', '/TestWorkload/ReReco/LogCollect'])
# Check performance
# Check histograms
self.assertAlmostEquals(workloadSummary['performance']['/TestWorkload/ReReco']['cmsRun1']['AvgEventTime']['histogram'][0]['average'],
0.89405199999999996, places = 2)
self.assertEqual(workloadSummary['performance']['/TestWorkload/ReReco']['cmsRun1']['AvgEventTime']['histogram'][0]['nEvents'],
10)
# Check standard performance
self.assertAlmostEquals(workloadSummary['performance']['/TestWorkload/ReReco']['cmsRun1']['TotalJobCPU']['average'], 17.786300000000001,
places = 2)
self.assertAlmostEquals(workloadSummary['performance']['/TestWorkload/ReReco']['cmsRun1']['TotalJobCPU']['stdDev'], 0.0,
places = 2)
# Check worstOffenders
self.assertEqual(workloadSummary['performance']['/TestWorkload/ReReco']['cmsRun1']['AvgEventTime']['worstOffenders'],
[{'logCollect': None, 'log': None, 'value': '0.894052', 'jobID': 1},
{'logCollect': None, 'log': None, 'value': '0.894052', 'jobID': 1},
{'logCollect': None, 'log': None, 'value': '0.894052', 'jobID': 2}])
# Check retryData
self.assertEqual(workloadSummary['retryData']['/TestWorkload/ReReco'], {'1': 10})
logCollectPFN = 'srm://srm-cms.cern.ch:8443/srm/managerv2?SFN=/castor/cern.ch/cms/store/logs/prod/2012/11/WMAgent/Run206446-MinimumBias-Run2012D-v1-Tier1PromptReco-4af7e658-23a4-11e2-96c7-842b2b4671d8/Run206446-MinimumBias-Run2012D-v1-Tier1PromptReco-4af7e658-23a4-11e2-96c7-842b2b4671d8-AlcaSkimLogCollect-1-logs.tar'
self.assertEqual(workloadSummary['logArchives'], {'/TestWorkload/ReReco/LogCollect' : [logCollectPFN for _ in range(10)]})
# LogCollect task is made out of identical FWJRs
# assert that it is identical
for x in workloadSummary['performance']['/TestWorkload/ReReco/LogCollect']['cmsRun1'].keys():
if x in config.TaskArchiver.histogramKeys:
continue
for y in ['average', 'stdDev']:
self.assertAlmostEquals(workloadSummary['performance']['/TestWorkload/ReReco/LogCollect']['cmsRun1'][x][y],
workloadSummary['performance']['/TestWorkload/ReReco']['cmsRun1'][x][y],
places = 2)
return
def testB_testErrors(self):
"""
_testErrors_
Test with a failed FWJR
"""
config = self.getConfig()
workloadPath = os.path.join(self.testDir, 'specDir', 'spec.pkl')
workload = self.createWorkload(workloadName=workloadPath)
testJobGroup = self.createTestJobGroup(config=config,
name=workload.name(),
specLocation=workloadPath,
error=True)
# Create second workload
testJobGroup2 = self.createTestJobGroup(
config=config,
name=workload.name(),
filesetName="TestFileset_2",
specLocation=workloadPath,
task="/TestWorkload/ReReco/LogCollect",
jobType="LogCollect")
cachePath = os.path.join(config.JobCreator.jobCacheDir, "TestWorkload",
"ReReco")
os.makedirs(cachePath)
self.assertTrue(os.path.exists(cachePath))
couchdb = CouchServer(config.JobStateMachine.couchurl)
jobdb = couchdb.connectDatabase("%s/jobs" % self.databaseName)
fwjrdb = couchdb.connectDatabase("%s/fwjrs" % self.databaseName)
jobdb.loadView("JobDump",
"jobsByWorkflowName",
options={
"startkey": [workload.name()],
"endkey": [workload.name(), {}]
})['rows']
fwjrdb.loadView("FWJRDump",
"fwjrsByWorkflowName",
options={
"startkey": [workload.name()],
"endkey": [workload.name(), {}]
})['rows']
self.populateWorkflowWithCompleteStatus()
testTaskArchiver = TaskArchiverPoller(config=config)
testTaskArchiver.algorithm()
cleanCouch = CleanCouchPoller(config=config)
cleanCouch.setup()
cleanCouch.algorithm()
dbname = getattr(config.JobStateMachine, "couchDBName")
workdatabase = couchdb.connectDatabase("%s/workloadsummary" % dbname)
workloadSummary = workdatabase.document(id=workload.name())
self.assertEqual(
workloadSummary['errors']['/TestWorkload/ReReco']['failureTime'],
500)
self.assertTrue('99999' in workloadSummary['errors']
['/TestWorkload/ReReco']['cmsRun1'])
failedRunInfo = workloadSummary['errors']['/TestWorkload/ReReco'][
'cmsRun1']['99999']['runs']
self.assertEqual(
failedRunInfo, {'10': [[12312, 12312]]},
"Wrong lumi information in the summary for failed jobs")
# Check the failures by site histograms
self.assertEqual(
workloadSummary['histograms']['workflowLevel']['failuresBySite']
['data']['T1_IT_CNAF']['Failed Jobs'], 10)
self.assertEqual(
workloadSummary['histograms']['stepLevel']['/TestWorkload/ReReco']
['cmsRun1']['errorsBySite']['data']['T1_IT_CNAF']['99999'], 10)
self.assertEqual(
workloadSummary['histograms']['stepLevel']['/TestWorkload/ReReco']
['cmsRun1']['errorsBySite']['data']['T1_IT_CNAF']['8020'], 10)
self.assertEqual(
workloadSummary['histograms']['workflowLevel']['failuresBySite']
['average']['Failed Jobs'], 10)
self.assertEqual(
workloadSummary['histograms']['stepLevel']['/TestWorkload/ReReco']
['cmsRun1']['errorsBySite']['average']['99999'], 10)
self.assertEqual(
workloadSummary['histograms']['stepLevel']['/TestWorkload/ReReco']
['cmsRun1']['errorsBySite']['average']['8020'], 10)
self.assertEqual(
workloadSummary['histograms']['workflowLevel']['failuresBySite']
['stdDev']['Failed Jobs'], 0)
self.assertEqual(
workloadSummary['histograms']['stepLevel']['/TestWorkload/ReReco']
['cmsRun1']['errorsBySite']['stdDev']['99999'], 0)
self.assertEqual(
workloadSummary['histograms']['stepLevel']['/TestWorkload/ReReco']
['cmsRun1']['errorsBySite']['stdDev']['8020'], 0)
return
def testA_BasicFunctionTest(self):
"""
_BasicFunctionTest_
Tests the components, by seeing if they can process a simple set of closeouts
"""
myThread = threading.currentThread()
config = self.getConfig()
workloadPath = os.path.join(self.testDir, 'specDir', 'spec.pkl')
workload = self.createWorkload(workloadName=workloadPath)
testJobGroup = self.createTestJobGroup(config=config,
name=workload.name(),
specLocation=workloadPath,
error=False)
# Create second workload
testJobGroup2 = self.createTestJobGroup(
config=config,
name=workload.name(),
filesetName="TestFileset_2",
specLocation=workloadPath,
task="/TestWorkload/ReReco/LogCollect",
jobType="LogCollect")
cachePath = os.path.join(config.JobCreator.jobCacheDir, "TestWorkload",
"ReReco")
os.makedirs(cachePath)
self.assertTrue(os.path.exists(cachePath))
cachePath2 = os.path.join(config.JobCreator.jobCacheDir,
"TestWorkload", "LogCollect")
os.makedirs(cachePath2)
self.assertTrue(os.path.exists(cachePath2))
result = myThread.dbi.processData(
"SELECT * FROM wmbs_subscription")[0].fetchall()
self.assertEqual(len(result), 2)
workflowName = "TestWorkload"
dbname = config.TaskArchiver.workloadSummaryCouchDBName
couchdb = CouchServer(config.JobStateMachine.couchurl)
workdatabase = couchdb.connectDatabase(dbname)
jobdb = couchdb.connectDatabase("%s/jobs" % self.databaseName)
fwjrdb = couchdb.connectDatabase("%s/fwjrs" % self.databaseName)
jobs = jobdb.loadView("JobDump",
"jobsByWorkflowName",
options={
"startkey": [workflowName],
"endkey": [workflowName, {}]
})['rows']
fwjrdb.loadView("FWJRDump",
"fwjrsByWorkflowName",
options={
"startkey": [workflowName],
"endkey": [workflowName, {}]
})['rows']
self.assertEqual(len(jobs), 2 * self.nJobs)
from WMCore.WMBS.CreateWMBSBase import CreateWMBSBase
create = CreateWMBSBase()
tables = []
for x in create.requiredTables:
tables.append(x[2:])
self.populateWorkflowWithCompleteStatus()
testTaskArchiver = TaskArchiverPoller(config=config)
testTaskArchiver.algorithm()
cleanCouch = CleanCouchPoller(config=config)
cleanCouch.setup()
cleanCouch.algorithm()
result = myThread.dbi.processData(
"SELECT * FROM wmbs_job")[0].fetchall()
self.assertEqual(len(result), 0)
result = myThread.dbi.processData(
"SELECT * FROM wmbs_subscription")[0].fetchall()
self.assertEqual(len(result), 0)
result = myThread.dbi.processData(
"SELECT * FROM wmbs_jobgroup")[0].fetchall()
self.assertEqual(len(result), 0)
result = myThread.dbi.processData(
"SELECT * FROM wmbs_fileset")[0].fetchall()
self.assertEqual(len(result), 0)
result = myThread.dbi.processData(
"SELECT * FROM wmbs_file_details")[0].fetchall()
self.assertEqual(len(result), 0)
# Make sure we deleted the directory
self.assertFalse(os.path.exists(cachePath))
self.assertFalse(
os.path.exists(
os.path.join(self.testDir, 'workloadTest/TestWorkload')))
testWMBSFileset = Fileset(id=1)
self.assertEqual(testWMBSFileset.exists(), False)
workloadSummary = workdatabase.document(id="TestWorkload")
# Check ACDC
self.assertEqual(workloadSummary['ACDCServer'],
sanitizeURL(config.ACDC.couchurl)['url'])
# Check the output
self.assertEqual(workloadSummary['output'].keys(),
['/Electron/MorePenguins-v0/RECO'])
self.assertEqual(
sorted(workloadSummary['output']['/Electron/MorePenguins-v0/RECO']
['tasks']),
['/TestWorkload/ReReco', '/TestWorkload/ReReco/LogCollect'])
# Check performance
# Check histograms
self.assertAlmostEqual(
workloadSummary['performance']['/TestWorkload/ReReco']['cmsRun1']
['AvgEventTime']['histogram'][0]['average'],
0.89405199999999996,
places=2)
self.assertEqual(
workloadSummary['performance']['/TestWorkload/ReReco']['cmsRun1']
['AvgEventTime']['histogram'][0]['nEvents'], 10)
# Check standard performance
self.assertAlmostEqual(
workloadSummary['performance']['/TestWorkload/ReReco']['cmsRun1']
['TotalJobCPU']['average'],
17.786300000000001,
places=2)
self.assertAlmostEqual(
workloadSummary['performance']['/TestWorkload/ReReco']['cmsRun1']
['TotalJobCPU']['stdDev'],
0.0,
places=2)
# Check worstOffenders
self.assertEqual(
workloadSummary['performance']['/TestWorkload/ReReco']['cmsRun1']
['AvgEventTime']['worstOffenders'], [{
'logCollect': None,
'log': None,
'value': '0.894052',
'jobID': 1
}, {
'logCollect': None,
'log': None,
'value': '0.894052',
'jobID': 1
}, {
'logCollect': None,
'log': None,
'value': '0.894052',
'jobID': 2
}])
# Check retryData
self.assertEqual(workloadSummary['retryData']['/TestWorkload/ReReco'],
{'1': 10})
logCollectPFN = 'srm://srm-cms.cern.ch:8443/srm/managerv2?SFN=/castor/cern.ch/cms/store/logs/prod/2012/11/WMAgent/Run206446-MinimumBias-Run2012D-v1-Tier1PromptReco-4af7e658-23a4-11e2-96c7-842b2b4671d8/Run206446-MinimumBias-Run2012D-v1-Tier1PromptReco-4af7e658-23a4-11e2-96c7-842b2b4671d8-AlcaSkimLogCollect-1-logs.tar'
self.assertEqual(workloadSummary['logArchives'], {
'/TestWorkload/ReReco/LogCollect':
[logCollectPFN for _ in range(10)]
})
# LogCollect task is made out of identical FWJRs
# assert that it is identical
for x in workloadSummary['performance'][
'/TestWorkload/ReReco/LogCollect']['cmsRun1'].keys():
if x in config.TaskArchiver.histogramKeys:
continue
for y in ['average', 'stdDev']:
self.assertAlmostEqual(
workloadSummary['performance']
['/TestWorkload/ReReco/LogCollect']['cmsRun1'][x][y],
workloadSummary['performance']['/TestWorkload/ReReco']
['cmsRun1'][x][y],
places=2)
return
