def _performWallClockChecks(self):
"""Watchdog performs the wall clock checks based on MJF. Signals are sent
to processes if we need to stop, but function always returns S_OK()
"""
mjf = MJF.MJF()
try:
wallClockSecondsLeft = mjf.getWallClockSecondsLeft()
except Exception as e:
# Just stop if we can't get the wall clock seconds left
return S_OK()
jobstartSeconds = mjf.getIntJobFeature('jobstart_secs')
if jobstartSeconds is None:
# Just stop if we don't know when the job started
return S_OK()
if (int(time.time()) > jobstartSeconds + self.stopSigStartSeconds) and \
(wallClockSecondsLeft < self.stopSigFinishSeconds + self.wallClockCheckSeconds):
# Need to send the signal! Assume it works to avoid sending the signal more than once
self.log.info('Sending signal %d to JobWrapper children' %
self.stopSigNumber)
self.stopSigSent = True
try:
for childPid in getChildrenPIDs(self.wrapperPID):
try:
cmdline = open('/proc/%d/cmdline' % childPid,
'r').read().replace('\0', ' ').strip()
except IOError:
# Process gone away? Not running on Linux? Skip anyway
continue
if re.search(self.stopSigRegex, cmdline) is not None:
self.log.info(
'Sending signal %d to process ID %d, cmdline = "%s"'
% (self.stopSigNumber, childPid, cmdline))
os.kill(childPid, self.stopSigNumber)
except Exception as e:
self.log.error(
'Failed to send signals to JobWrapper children! (%s)' %
str(e))
return S_OK()
configFile = None
for unprocSw in Script.getUnprocessedSwitches():
if unprocSw[0] in ("U", "Update"):
update = True
elif unprocSw[0] in ("R", "Reconfig"):
configFile = unprocSw[1]
if __name__ == "__main__":
from DIRAC import gLogger, gConfig
from DIRAC.WorkloadManagementSystem.Client.DIRACbenchmark import singleDiracBenchmark
from DIRAC.ConfigurationSystem.Client.Helpers.Operations import Operations
from DIRAC.Core.Utilities import MJF
mjf = MJF.MJF()
mjf.updateConfig()
db12JobFeature = mjf.getJobFeature('db12')
hs06JobFeature = mjf.getJobFeature('hs06')
result = singleDiracBenchmark(1)
if result is None:
gLogger.error('Cannot make benchmark measurements')
DIRAC.exit(1)
db12Measured = round(result['NORM'], 1)
corr = Operations().getValue('JobScheduling/CPUNormalizationCorrection',
1.)
norm = round(result['NORM'] / corr, 1)