def stop_job(self, job):
"""Attempts to delete a job from the PBS queue"""
job_id = job.get_job_runner_external_id().encode('utf-8')
job_tag = "(%s/%s)" % (job.get_id_tag(), job_id)
log.debug("%s Stopping PBS job" % job_tag)
# Declare the connection handle c so that it can be cleaned up:
c = None
try:
pbs_server_name = self.__get_pbs_server(job.destination_params)
if pbs_server_name is None:
log.debug(
"(%s) Job queued but no destination stored in job params, cannot delete"
% job_tag)
return
c = pbs.pbs_connect(util.smart_str(pbs_server_name))
if c <= 0:
log.debug(
"(%s) Connection to PBS server for job delete failed" %
job_tag)
return
pbs.pbs_deljob(c, job_id, '')
log.debug("%s Removed from PBS queue before job completion" %
job_tag)
except:
e = traceback.format_exc()
log.debug("%s Unable to stop job: %s" % (job_tag, e))
finally:
# Cleanup: disconnect from the server.
if (None is not c):
pbs.pbs_disconnect(c)
def stop_job( self, job ):
"""Attempts to delete a job from the PBS queue"""
job_id = job.get_job_runner_external_id().encode('utf-8')
job_tag = "(%s/%s)" % ( job.get_id_tag(), job_id )
log.debug( "%s Stopping PBS job" % job_tag )
# Declare the connection handle c so that it can be cleaned up:
c = None
try:
pbs_server_name = self.__get_pbs_server( job.destination_params )
if pbs_server_name is None:
log.debug("(%s) Job queued but no destination stored in job params, cannot delete"
% job_tag )
return
c = pbs.pbs_connect( util.smart_str( pbs_server_name ) )
if c <= 0:
log.debug("(%s) Connection to PBS server for job delete failed"
% job_tag )
return
pbs.pbs_deljob( c, job_id, '' )
log.debug( "%s Removed from PBS queue before job completion"
% job_tag )
except:
e = traceback.format_exc()
log.debug( "%s Unable to stop job: %s" % ( job_tag, e ) )
finally:
# Cleanup: disconnect from the server.
if ( None is not c ):
pbs.pbs_disconnect( c )
def alive(self, process_id):
alive = False
try:
status = self.pbsquery.getjob(str(process_id))['job_state'][0]
except:
# job not found
status = -1
sys.stderr.write("EXC: %s\n" % str(sys.exc_info()[0]))
sys.stderr.write("Could not find job for process id %d\n" % process_id)
if status == 'Q':
sys.stderr.write("Job %d waiting in queue.\n" % (process_id))
alive = True
elif status == 'R':
sys.stderr.write("Job %d is running.\n" % (process_id))
alive = True
elif status in ['H','S']:
sys.stderr.write("Job %d is held or suspended.\n" % (process_id))
alive = False
if not alive:
try:
# Kill the job.
c = pbs.pbs_connect(pbs.pbs_default())
result = pbs.pbs_deljob(c, str(process_id))
sys.stderr.write("Killed job %d.\n" % (process_id))
except:
sys.stderr.write("Failed to kill job %d.\n" % (process_id))
return False
else:
return True
def remove(self):
"""Remove the job with id jobid"""
result = pbs.pbs_deljob(self.pbsconn, self.jobid, '') # use empty string, not NULL
if result:
self.log.error("Failed to delete job %s: error %s" % (self.jobid, result))
else:
self.log.debug("Succesfully deleted job %s" % self.jobid)
def remove(self):
"""Remove the job with id jobid"""
result = pbs.pbs_deljob(self.pbsconn, self.jobid, '') # use empty string, not NULL
if result:
raise EasyBuildError("Failed to delete job %s: error %s", self.jobid, result)
else:
self.log.debug("Succesfully deleted job %s" % self.jobid)
def del_job(self, job_id, server):
c = pbs.pbs_connect(str(
pbs.pbs_default())) # Create new connection for the child process
if server is None:
server = pbs.pbs_default()
job_full_id = job_id + '.' + server
result = pbs.pbs_deljob(c, job_full_id, 'NULL')
return result # If operation is successfull, result == 0
def remove(self, jobid=None):
"""Remove the job with id jobid."""
if jobid is None:
jobid = self.jobid
result = pbs.pbs_deljob(self.pbsconn, self.jobid, '') # use empty string, not NULL (one can pass the deldelay=nnnn option)
if result:
self.log.error("Failed to delete job %s: error %s", jobid, result)
else:
self.log.debug("Succesfully deleted job %s", jobid)
def delete_job(self, job_id):
"""
Sends job delete request to pbs_server for job
:param job_id: job id to delete
:return: pbs_deljob return value (0 on success)
"""
connection = _connect_to_server(self.pbs_server)
rval = pbs.pbs_deljob(connection, job_id, '')
pbs.pbs_disconnect(connection)
return rval
def delete_all_jobs(self, ids):
"""
delete all jobs in a list of jobs
:param ids: list of all jobs
:return: zero on success, otherwise return value of failed pbs_deljob
"""
for job_id in ids:
connection = _connect_to_server(self.pbs_server)
rval = pbs.pbs_deljob(connection, job_id, '')
pbs.pbs_disconnect(connection)
if rval and (rval != self.E_UNKNOWN or rval != self.E_STATE):
return rval
return 0
def kill(self, obj):
conn=self.pbs_conn()
for job in obj.jobs :
if not self.valid( job.runningJob ): continue
id=str(job.runningJob['schedulerId']).strip()
res=pbs.pbs_deljob(conn, id, '')
if res!=0:
err, err_text=pbs.error()
self.logging.error('Error in job kill for '+id)
self.logging.error('PBS error code '+str(err)+': '+err_text)
self.pbs_disconn(conn)
raise SchedulerError('PBS error', str(err)+': '+err_text)
self.pbs_disconn(conn)
def kill(self, obj):
conn = self.pbs_conn()
for job in obj.jobs:
if not self.valid(job.runningJob): continue
id = str(job.runningJob['schedulerId']).strip()
res = pbs.pbs_deljob(conn, id, '')
if res != 0:
err, err_text = pbs.error()
self.logging.error('Error in job kill for ' + id)
self.logging.error('PBS error code ' + str(err) + ': ' +
err_text)
self.pbs_disconn(conn)
raise SchedulerError('PBS error', str(err) + ': ' + err_text)
self.pbs_disconn(conn)
