cocluto/ClusterController/SunGridEngine.py

import Util
from QstatParser import *

class SunGridEngine:

	def getCurrentJobsState( self ):
		bBUG_00000009_IS_STILL_ALIVE = True
		if bBUG_00000009_IS_STILL_ALIVE:
			logDebug('Querying the current state of jobs')
		returnCode = -1
		delayBetweenAttemps = 5 # in seconds
		while returnCode != 0:
			command = ['qstat', '-f', '-u', '*']
			(returnCode, qstatOutput, stderr) = executeProgram( command )
			if returnCode != 0:
				logWarning('command "%s" failed (returnCode = %d, stdout="%s", stderr="%s"). Retrying in %d seconds' % (' '.join(command), returnCode, qstatOutput, stderr, delayBetweenAttemps))
				time.sleep(delayBetweenAttemps)
		if bBUG_00000009_IS_STILL_ALIVE:
			logDebug('Just got current state of jobs')

		jobsState = QstatParser().parseQstatOutput( qstatOutput )
		jobsState.setTime( time.time() )
		
		
		# read the requirements for pending jobs (which parallel environment, which queue, which architecture) from sge
		if False: # no need for job details at the moment and since it's very slow, it's been disabled
			for unused_jobId, job in jobsState.getPendingJobs().iteritems():
				(returnCode, stdout, stderr) = executeProgram( ['qstat', '-j', job.getId().asStr()] )
				assert returnCode != 0, 'prout'
				QstatParser().parseJobDetails( stdout, job )
		
		return jobsState

	def setQueueInstanceActivation( self, strQueueInstanceName, bEnable ):
		argument = 'd'
		if bEnable:
			argument = 'e'
		bBUG_00000269_IS_STILL_ALIVE = True # for some reason, qmod -d (and maybe any sge command) could fail with error: commlib error: can't connect to service (Address already in use)
		delayBetweenAttemps = 5 # in seconds
		while True:
			errorCode, unused_stdout, unused_stderr = executeProgram(['qmod', '-'+argument, strQueueInstanceName])
			if bBUG_00000269_IS_STILL_ALIVE:
				# if the command failed, try again
				if errorCode == 0:
					break
				time.sleep(delayBetweenAttemps)
			else:
				break
		return (errorCode == 0)
		
	def queueIsEmpty( self, strMachineName ):			
		(returnCode, qstatOutput, unused_stderr) = executeProgram( ['qstat', '-f', '-u', '*'] )
		assert( returnCode == 0 )
		jobsState = QstatParser().parseQstatOutput( qstatOutput )
		jobs = jobsState.getJobsOnMachine( strMachineName )
		return (len(jobs) == 0)