def _submitJob(self, jobNum, module):
activity = Activity('submitting job %d' % jobNum)
try:
sandbox = tempfile.mkdtemp('', '%s.%04d.' % (module.taskID, jobNum), self._sandbox_helper.get_path())
except Exception:
raise BackendError('Unable to create sandbox directory "%s"!' % sandbox)
sbPrefix = sandbox.replace(self._sandbox_helper.get_path(), '').lstrip('/')
def translateTarget(d, s, t):
return (d, s, os.path.join(sbPrefix, t))
self.smSBIn.doTransfer(ismap(translateTarget, self._getSandboxFilesIn(module)))
self._writeJobConfig(os.path.join(sandbox, '_jobconfig.sh'), jobNum, module, {
'GC_SANDBOX': sandbox, 'GC_SCRATCH_SEARCH': str.join(' ', self.scratchPath)})
reqs = self.brokerSite.brokerAdd(module.getRequirements(jobNum), WMS.SITES)
reqs = dict(self.brokerQueue.brokerAdd(reqs, WMS.QUEUES))
if (self.memory > 0) and (reqs.get(WMS.MEMORY, 0) < self.memory):
reqs[WMS.MEMORY] = self.memory # local jobs need higher (more realistic) memory requirements
(stdout, stderr) = (os.path.join(sandbox, 'gc.stdout'), os.path.join(sandbox, 'gc.stderr'))
jobName = module.getDescription(jobNum).jobName
proc = LoggedProcess(self.submitExec, '%s %s "%s" %s' % (self.submitOpts,
self.getSubmitArguments(jobNum, jobName, reqs, sandbox, stdout, stderr),
utils.pathShare('gc-local.sh'), self.getJobArguments(jobNum, sandbox)))
retCode = proc.wait()
gcIDText = proc.getOutput().strip().strip('\n')
try:
gcID = self.parseSubmitOutput(gcIDText)
except Exception:
gcID = None
activity.finish()
if retCode != 0:
self._log.warning('%s failed:', self.submitExec)
elif gcID is None:
self._log.warning('%s did not yield job id:\n%s', self.submitExec, gcIDText)
if gcID:
gcID = self._createId(gcID)
open(os.path.join(sandbox, gcID), 'w')
else:
proc.logError(self.errorLog)
return (jobNum, utils.QM(gcID, gcID, None), {'sandbox': sandbox})
def _getJobsOutput(self, allIds):
if len(allIds) == 0:
raise StopIteration
basePath = os.path.join(self._outputPath, 'tmp')
try:
if len(allIds) == 1:
# For single jobs create single subdir
basePath = os.path.join(basePath, md5(allIds[0][0]).hexdigest())
utils.ensureDirExists(basePath)
except Exception:
raise BackendError('Temporary path "%s" could not be created.' % basePath, BackendError)
activity = Activity('retrieving %d job outputs' % len(allIds))
for ids in imap(lambda x: allIds[x:x+self._nJobsPerChunk], irange(0, len(allIds), self._nJobsPerChunk)):
jobNumMap = dict(ids)
jobs = ' '.join(self._getRawIDs(ids))
log = tempfile.mktemp('.log')
proc = LoggedProcess(self._outputExec,
'--noint --logfile "%s" --dir "%s" %s' % (log, basePath, jobs))
# yield output dirs
todo = jobNumMap.values()
done = []
currentJobNum = None
for line in imap(str.strip, proc.iter()):
match = re.match(self._outputRegex, line)
if match:
currentJobNum = jobNumMap.get(self._createId(match.groupdict()['rawId']))
todo.remove(currentJobNum)
done.append(match.groupdict()['rawId'])
outputDir = match.groupdict()['outputDir']
if os.path.exists(outputDir):
if 'GC_WC.tar.gz' in os.listdir(outputDir):
wildcardTar = os.path.join(outputDir, 'GC_WC.tar.gz')
try:
tarfile.TarFile.open(wildcardTar, 'r:gz').extractall(outputDir)
os.unlink(wildcardTar)
except Exception:
self._log.error('Can\'t unpack output files contained in %s', wildcardTar)
yield (currentJobNum, outputDir)
currentJobNum = None
retCode = proc.wait()
if retCode != 0:
if 'Keyboard interrupt raised by user' in proc.getError():
utils.removeFiles([log, basePath])
raise StopIteration
else:
proc.logError(self.errorLog, log = log)
self._log.error('Trying to recover from error ...')
for dirName in os.listdir(basePath):
yield (None, os.path.join(basePath, dirName))
activity.finish()
# return unretrievable jobs
for jobNum in todo:
yield (jobNum, None)
purgeLog = tempfile.mktemp('.log')
purgeProc = LoggedProcess(utils.resolveInstallPath('glite-ce-job-purge'),
'--noint --logfile "%s" %s' % (purgeLog, str.join(' ', done)))
retCode = purgeProc.wait()
if retCode != 0:
if self.explainError(purgeProc, retCode):
pass
else:
proc.logError(self.errorLog, log = purgeLog, jobs = done)
utils.removeFiles([log, purgeLog, basePath])
def _getJobsOutput(self, allIds):
if len(allIds) == 0:
raise StopIteration
basePath = os.path.join(self._outputPath, 'tmp')
try:
if len(allIds) == 1:
# For single jobs create single subdir
basePath = os.path.join(basePath,
md5(allIds[0][0]).hexdigest())
utils.ensureDirExists(basePath)
except Exception:
raise BackendError(
'Temporary path "%s" could not be created.' % basePath,
BackendError)
activity = Activity('retrieving %d job outputs' % len(allIds))
for ids in imap(lambda x: allIds[x:x + self._nJobsPerChunk],
irange(0, len(allIds), self._nJobsPerChunk)):
jobNumMap = dict(ids)
jobs = ' '.join(self._getRawIDs(ids))
log = tempfile.mktemp('.log')
proc = LoggedProcess(
self._outputExec,
'--noint --logfile "%s" --dir "%s" %s' % (log, basePath, jobs))
# yield output dirs
todo = jobNumMap.values()
done = []
currentJobNum = None
for line in imap(str.strip, proc.iter()):
match = re.match(self._outputRegex, line)
if match:
currentJobNum = jobNumMap.get(
self._createId(match.groupdict()['rawId']))
todo.remove(currentJobNum)
done.append(match.groupdict()['rawId'])
outputDir = match.groupdict()['outputDir']
if os.path.exists(outputDir):
if 'GC_WC.tar.gz' in os.listdir(outputDir):
wildcardTar = os.path.join(outputDir,
'GC_WC.tar.gz')
try:
tarfile.TarFile.open(
wildcardTar, 'r:gz').extractall(outputDir)
os.unlink(wildcardTar)
except Exception:
self._log.error(
'Can\'t unpack output files contained in %s',
wildcardTar)
yield (currentJobNum, outputDir)
currentJobNum = None
retCode = proc.wait()
if retCode != 0:
if 'Keyboard interrupt raised by user' in proc.getError():
utils.removeFiles([log, basePath])
raise StopIteration
else:
proc.logError(self.errorLog, log=log)
self._log.error('Trying to recover from error ...')
for dirName in os.listdir(basePath):
yield (None, os.path.join(basePath, dirName))
activity.finish()
# return unretrievable jobs
for jobNum in todo:
yield (jobNum, None)
purgeLog = tempfile.mktemp('.log')
purgeProc = LoggedProcess(
utils.resolveInstallPath('glite-ce-job-purge'),
'--noint --logfile "%s" %s' % (purgeLog, str.join(' ', done)))
retCode = purgeProc.wait()
if retCode != 0:
if self.explainError(purgeProc, retCode):
pass
else:
proc.logError(self.errorLog, log=purgeLog, jobs=done)
utils.removeFiles([log, purgeLog, basePath])