def submit_jobs():
"""
Submits jobs to the queue for processing.
***NOTE: Priority is given to jobs with status 'retrying'.
"""
jobs = []
special_query1 = "SELECT distinct j.* FROM jobs AS j JOIN job_files AS jf " \
"ON j.id=jf.job_id JOIN files AS f ON f.id=jf.file_id WHERE " \
"j.status in ('new','retrying') AND f.filename LIKE " \
"'%p2030.2019____.G__.%.fits' ORDER BY j.updated_at ASC"
special_query2 = "SELECT distinct j.* FROM jobs AS j JOIN job_files AS jf " \
"ON j.id=jf.job_id JOIN files AS f ON f.id=jf.file_id WHERE " \
"j.status in ('new','retrying') AND f.filename LIKE " \
"'%p2030.2018____.G__.%.fits' ORDER BY j.updated_at ASC"
special_query3 = "SELECT distinct j.* FROM jobs AS j JOIN job_files AS jf " \
"ON j.id=jf.job_id JOIN files AS f ON f.id=jf.file_id WHERE " \
"j.status in ('new','retrying') AND f.filename LIKE " \
"'%p2030.2019____.G___.%.fits' ORDER BY j.updated_at ASC"
special_query4 = "SELECT distinct j.* FROM jobs AS j JOIN job_files AS jf " \
"ON j.id=jf.job_id JOIN files AS f ON f.id=jf.file_id WHERE " \
"j.status in ('new','retrying') AND f.filename LIKE " \
"'%p2030.2018____.G___.%.fits' ORDER BY j.updated_at ASC"
if not len(jobs):
jobs.extend(jobtracker.query(special_query1))
print len(jobs),"in inner galaxy (2019 data)."
jobs.extend(jobtracker.query(special_query2))
print len(jobs),"in inner galaxy (2018 data)."
if len(jobs)<50:
jobs.extend(jobtracker.query(special_query3))
print len(jobs),"in outer galaxy (2019 data)."
jobs.extend(jobtracker.query(special_query4))
print len(jobs),"in outer galaxy (2018 data)."
if len(jobs)<50:
jobs.extend(jobtracker.query("SELECT * FROM jobs " \
"WHERE status='retrying' " \
"ORDER BY updated_at ASC"))
jobs.extend(jobtracker.query("SELECT * FROM jobs " \
"WHERE status='new'" \
"ORDER BY updated_at ASC"))
for job in jobs[:500]:
if config.jobpooler.queue_manager.can_submit():
try:
submit(job)
if config.jobpooler.submit_sleep:
time.sleep(config.jobpooler.submit_sleep)
#time.sleep(5)
except:
fn = pipeline_utils.get_fns_for_jobid(job['id'])[0]
fn = fn.split('results/')[-1]
print "Skipping job-id: %d (%s)"%(job['id'],fn)
else:
break
def submit(job_row):
"""
Submits a job to QueueManager, if successful will store returned queue id.
Input:
job_row: A row from the jobs table. The datafiles associated
with this job will be submitted to be processed.
Outputs:
None
"""
fns = pipeline_utils.get_fns_for_jobid(job_row['id'])
try:
outdir = get_output_dir(fns)
# Submit job
queue_id = config.jobpooler.queue_manager.submit(fns, outdir)
msg = "Submitted job to process:\n"
msg += "\tJob ID: %d, Queue ID: %s\n" % (job_row['id'], queue_id)
msg += "\tData file(s):\n"
for fn in fns:
msg += "\t%s\n" % fn
jobpool_cout.outs(msg)
queries = []
queries.append("INSERT INTO job_submits (" \
"job_id, " \
"queue_id, " \
"output_dir, " \
"status, " \
"created_at, " \
"updated_at, " \
"details) " \
"VALUES (%d,'%s','%s','%s','%s','%s','%s')" % \
(job_row['id'], queue_id, outdir, 'running', \
jobtracker.nowstr(), jobtracker.nowstr(), \
'Job submitted to queue'))
queries.append("UPDATE jobs " \
"SET status='submitted', " \
"details='Job submitted to queue', " \
"updated_at='%s' " \
"WHERE id=%d" % \
(jobtracker.nowstr(), job_row['id']))
jobtracker.query(queries)
except pipeline_utils.PipelineError:
# Error caught during job submission.
exceptionmsgs = traceback.format_exception(*sys.exc_info())
errormsg = "Error while submitting job!\n"
errormsg += "\tJob ID: %d\n\n" % job_row['id']
errormsg += "".join(exceptionmsgs)
jobpool_cout.outs("Error while submitting job!\n" \
"\tJob ID: %d\n\t%s\n" % \
(job_row['id'], exceptionmsgs[-1]))
queries = []
queries.append("INSERT INTO job_submits (" \
"job_id, " \
"status, " \
"created_at, " \
"updated_at, " \
"details) " \
"VALUES (%d,'%s','%s','%s','%s')" % \
(job_row['id'], 'submission_failed', \
jobtracker.nowstr(), jobtracker.nowstr(), \
errormsg))
queries.append("UPDATE jobs " \
"SET status='failed', " \
"details='Error while submitting job', " \
"updated_at='%s' " \
"WHERE id=%d" % \
(jobtracker.nowstr(), job_row['id']))
jobtracker.query(queries)
def submit(job_row):
"""
Submits a job to QueueManager, if successful will store returned queue id.
Input:
job_row: A row from the jobs table. The datafiles associated
with this job will be submitted to be processed.
Outputs:
None
"""
fns = pipeline_utils.get_fns_for_jobid(job_row['id'])
bad_days = [
'20170414', '20170419', '20170420', '20170423', '20170423', '20170427',
'20170429', '20170503', '20170510', '20170516'
]
bad_beams = ['b5', 'b6']
for bad_day in bad_days:
if bad_day in fns[0]:
if (bad_beams[0] in fns[0]) or (bad_beams[1] in fns[0]):
print "Files affected by the bad beams 5, 6 60Hz signal: ", fns
print "Will delete the raw data files."
queries = []
arglist = []
queries.append("INSERT INTO job_submits (" \
"job_id, " \
"status, " \
"created_at, " \
"updated_at, " \
"details) " \
"VALUES (?, ?, ?, ?, ?)" )
arglist.append( ( job_row['id'], 'Beams 5 and 6', \
jobtracker.nowstr(), jobtracker.nowstr(), \
"Affected by 60Hz signal") )
queries.append("UPDATE jobs " \
"SET status='terminal_failure', " \
"details='Beams 5 and 6 affected by 60Hz signal', " \
"updated_at=? " \
"WHERE id=?" )
arglist.append((jobtracker.nowstr(), job_row['id']))
jobtracker.execute(queries, arglist)
return
try:
presubmission_check(fns)
print "\t file has been checked"
outdir, created = get_output_dir(fns)
if not created:
#current_usr = getpass.getuser()
#errormsg = "Permission denied to %s when attempting to create output directory %s"%(current_usr,outdir)
#f = open('/scratch/eparent/eparent/PALFA4/results/missing_permission.out','a')
#f.write(errormsg+'\n')
#f.close()
print "Permission issues while creating output directory, skipping"
pass
# Attempt to submit the job
if job_row['status'] == 'retrying':
## ppn=2
ppn = 1
tfact = 1.25
else:
ppn = 1
tfact = 1.
if config.jobpooler.alternative_submit_script:
print "Submitting:", config.jobpooler.alternative_submit_script
queue_id = config.jobpooler.queue_manager.submit\
(fns, outdir, job_row['id'],\
script=config.jobpooler.alternative_submit_script,\
ppn=ppn)
else:
queue_id = config.jobpooler.queue_manager.submit(fns,
outdir,
job_row['id'],
ppn=ppn,
tfact=tfact)
msg = "Submitted job to process:\n"
msg += "\tJob ID: %d, Queue ID: %s\n" % (job_row['id'], queue_id)
msg += "\tData file(s):\n"
for fn in fns:
msg += "\t%s\n" % fn
jobpool_cout.outs(msg)
queries = []
queries.append("INSERT INTO job_submits (" \
"job_id, " \
"queue_id, " \
"output_dir, " \
"status, " \
"created_at, " \
"updated_at, " \
"details) " \
"VALUES (%d,'%s','%s','%s','%s','%s','%s')" % \
(job_row['id'], queue_id, outdir, 'running', \
jobtracker.nowstr(), jobtracker.nowstr(), \
'Job submitted to queue'))
queries.append("UPDATE jobs " \
"SET status='submitted', " \
"details='Job submitted to queue', updated_at='%s' WHERE id=%d" % (jobtracker.nowstr(), job_row['id']))
jobtracker.query(queries)
except (FailedPreCheckError):
# Error caught during presubmission check.
exceptionmsgs = traceback.format_exception(*sys.exc_info())
errormsg = "Job ID: %d " % job_row['id']
errormsg += "failed presubmission check!\n\n"
errormsg += "".join(exceptionmsgs)
jobpool_cout.outs("Job ID: %d failed presubmission check!\n\t%s\n" % \
(job_row['id'], exceptionmsgs[-1]))
#if config.basic.delete_rawdata:
# pipeline_utils.clean_up(job_row['id'])
queries = []
arglist = []
queries.append("INSERT INTO job_submits (" \
"job_id, " \
"status, " \
"created_at, " \
"updated_at, " \
"details) " \
"VALUES (?, ?, ?, ?, ?)" )
arglist.append( ( job_row['id'], 'precheck_failed', \
jobtracker.nowstr(), jobtracker.nowstr(), \
errormsg) )
queries.append("UPDATE jobs " \
"SET status='terminal_failure', " \
"details='Failed presubmission check', " \
"updated_at=? " \
"WHERE id=?" )
arglist.append((jobtracker.nowstr(), job_row['id']))
jobtracker.execute(queries, arglist)
except (queue_managers.QueueManagerJobFatalError,\
datafile.DataFileError):
# Error caught during job submission.
exceptionmsgs = traceback.format_exception(*sys.exc_info())
errormsg = "Error while submitting job!\n"
errormsg += "\tJob ID: %d\n\n" % job_row['id']
errormsg += "".join(exceptionmsgs)
jobpool_cout.outs("Error while submitting job!\n" \
"\tJob ID: %d\n\t%s\n" % \
(job_row['id'], exceptionmsgs[-1]))
queries = []
arglist = []
queries.append("INSERT INTO job_submits (" \
"job_id, " \
"status, " \
"created_at, " \
"updated_at, " \
"details) " \
"VALUES (?, ?, ?, ?, ?)" )
arglist.append( ( job_row['id'], 'submission_failed', \
jobtracker.nowstr(), jobtracker.nowstr(), \
errormsg) )
queries.append("UPDATE jobs " \
"SET status='failed', " \
"details='Error while submitting job', " \
"updated_at=? " \
"WHERE id=?" )
arglist.append((jobtracker.nowstr(), job_row['id']))
jobtracker.execute(queries, arglist)
except queue_managers.QueueManagerNonFatalError:
# Do nothing. Don't submit the job. Don't mark the job as 'submitted'.
# Don't mark the job as 'failed'. The job submission will be retried.
pass
except queue_managers.QueueManagerFatalError:
# A fatal error occurred. Re-raise!
raise
except (MissingFilesError):
# Unexpected error
exceptionmsgs = traceback.format_exception(*sys.exc_info())
errormsg = "Job ID: %d " % job_row['id']
errormsg += "Raw data files missing from /scratch/ area.!\n\n"
errormsg += "".join(exceptionmsgs)
queries = []
arglist = []
queries.append("INSERT INTO job_submits (" \
"job_id, " \
"status, " \
"created_at, " \
"updated_at, " \
"details) " \
"VALUES (?, ?, ?, ?, ?)" )
arglist.append( ( job_row['id'], 'submission_failed', \
jobtracker.nowstr(), jobtracker.nowstr(), \
errormsg) )
queries.append("UPDATE jobs " \
"SET status='failed',details='Error while submitting job',updated_at=? " \
"WHERE id=?" )
arglist.append((jobtracker.nowstr(), job_row['id']))
jobtracker.execute(queries, arglist)
print errormsg
def submit(job_row):
"""
Submits a job to QueueManager, if successful will store returned queue id.
Input:
job_row: A row from the jobs table. The datafiles associated
with this job will be submitted to be processed.
Outputs:
None
"""
fns = pipeline_utils.get_fns_for_jobid(job_row['id'])
try:
presubmission_check(fns)
outdir = get_output_dir(fns)
# Attempt to submit the job
queue_id = config.jobpooler.queue_manager.submit\
(fns, outdir, job_row['id'])
except (FailedPreCheckError):
# Error caught during presubmission check.
exceptionmsgs = traceback.format_exception(*sys.exc_info())
errormsg = "Job ID: %d " % job_row['id']
errormsg += "failed presubmission check!\n\n"
errormsg += "".join(exceptionmsgs)
jobpool_cout.outs("Job ID: %d failed presubmission check!\n\t%s\n" % \
(job_row['id'], exceptionmsgs[-1]))
if config.email.send_on_terminal_failures:
# Send error email
msg = "Presubmission check failed!\n"
msg += "Job ID: %d\n\n" % \
(job_row['id'])
msg += errormsg
msg += "\n*** Job has been terminally failed. ***\n"
msg += "*** Job will NOT be re-submitted! ***\n"
if config.basic.delete_rawdata:
jobpool_cout.outs("Job #%d will NOT be retried. " \
"Data files will be deleted." % job_row['id'])
msg += "*** Raw data files will be deleted. ***\n"
else:
jobpool_cout.outs("Job #%d will NOT be retried. " % job_row['id'])
notification = mailer.ErrorMailer(msg, \
subject="Job failed presubmission check - Terminal")
notification.send()
if config.basic.delete_rawdata:
pipeline_utils.clean_up(job_row['id'])
queries = []
arglist = []
queries.append("INSERT INTO job_submits (" \
"job_id, " \
"status, " \
"created_at, " \
"updated_at, " \
"details) " \
"VALUES (?, ?, ?, ?, ?)" )
arglist.append( ( job_row['id'], 'precheck_failed', \
jobtracker.nowstr(), jobtracker.nowstr(), \
errormsg) )
queries.append("UPDATE jobs " \
"SET status='terminal_failure', " \
"details='Failed presubmission check', " \
"updated_at=? " \
"WHERE id=?" )
arglist.append( (jobtracker.nowstr(), job_row['id']) )
jobtracker.execute(queries, arglist)
except (queue_managers.QueueManagerJobFatalError,\
datafile.DataFileError):
# Error caught during job submission.
exceptionmsgs = traceback.format_exception(*sys.exc_info())
errormsg = "Error while submitting job!\n"
errormsg += "\tJob ID: %d\n\n" % job_row['id']
errormsg += "".join(exceptionmsgs)
jobpool_cout.outs("Error while submitting job!\n" \
"\tJob ID: %d\n\t%s\n" % \
(job_row['id'], exceptionmsgs[-1]))
queries = []
arglist = []
queries.append("INSERT INTO job_submits (" \
"job_id, " \
"status, " \
"created_at, " \
"updated_at, " \
"details) " \
"VALUES (?, ?, ?, ?, ?)" )
arglist.append( ( job_row['id'], 'submission_failed', \
jobtracker.nowstr(), jobtracker.nowstr(), \
errormsg) )
queries.append("UPDATE jobs " \
"SET status='failed', " \
"details='Error while submitting job', " \
"updated_at=? " \
"WHERE id=?" )
arglist.append( (jobtracker.nowstr(), job_row['id']) )
jobtracker.execute(queries, arglist)
except queue_managers.QueueManagerNonFatalError:
# Do nothing. Don't submit the job. Don't mark the job as 'submitted'.
# Don't mark the job as 'failed'. The job submission will be retried.
pass
except queue_managers.QueueManagerFatalError:
# A fatal error occurred. Re-raise!
raise
else:
# No error occurred
msg = "Submitted job to process:\n"
msg += "\tJob ID: %d, Queue ID: %s\n" % (job_row['id'], queue_id)
msg += "\tData file(s):\n"
for fn in fns:
msg += "\t%s\n" % fn
jobpool_cout.outs(msg)
queries = []
queries.append("INSERT INTO job_submits (" \
"job_id, " \
"queue_id, " \
"output_dir, " \
"status, " \
"created_at, " \
"updated_at, " \
"details) " \
"VALUES (%d,'%s','%s','%s','%s','%s','%s')" % \
(job_row['id'], queue_id, outdir, 'running', \
jobtracker.nowstr(), jobtracker.nowstr(), \
'Job submitted to queue'))
queries.append("UPDATE jobs " \
"SET status='submitted', " \
"details='Job submitted to queue', " \
"updated_at='%s' " \
"WHERE id=%d" % \
(jobtracker.nowstr(), job_row['id']))
jobtracker.query(queries)
def submit(job_row):
"""
Submits a job to QueueManager, if successful will store returned queue id.
Input:
job_row: A row from the jobs table. The datafiles associated
with this job will be submitted to be processed.
Outputs:
None
"""
fns = pipeline_utils.get_fns_for_jobid(job_row['id'])
try:
outdir = get_output_dir(fns)
# Submit job
queue_id = config.jobpooler.queue_manager.submit(fns, outdir)
msg = "Submitted job to process:\n"
msg += "\tJob ID: %d, Queue ID: %s\n" % (job_row['id'], queue_id)
msg += "\tData file(s):\n"
for fn in fns:
msg += "\t%s\n" % fn
jobpool_cout.outs(msg)
queries = []
queries.append("INSERT INTO job_submits (" \
"job_id, " \
"queue_id, " \
"output_dir, " \
"status, " \
"created_at, " \
"updated_at, " \
"details) " \
"VALUES (%d,'%s','%s','%s','%s','%s','%s')" % \
(job_row['id'], queue_id, outdir, 'running', \
jobtracker.nowstr(), jobtracker.nowstr(), \
'Job submitted to queue'))
queries.append("UPDATE jobs " \
"SET status='submitted', " \
"details='Job submitted to queue', " \
"updated_at='%s' " \
"WHERE id=%d" % \
(jobtracker.nowstr(), job_row['id']))
jobtracker.query(queries)
except pipeline_utils.PipelineError:
# Error caught during job submission.
exceptionmsgs = traceback.format_exception(*sys.exc_info())
errormsg = "Error while submitting job!\n"
errormsg += "\tJob ID: %d\n\n" % job_row['id']
errormsg += "".join(exceptionmsgs)
jobpool_cout.outs("Error while submitting job!\n" \
"\tJob ID: %d\n\t%s\n" % \
(job_row['id'], exceptionmsgs[-1]))
queries = []
queries.append("INSERT INTO job_submits (" \
"job_id, " \
"status, " \
"created_at, " \
"updated_at, " \
"details) " \
"VALUES (%d,'%s','%s','%s','%s')" % \
(job_row['id'], 'submission_failed', \
jobtracker.nowstr(), jobtracker.nowstr(), \
errormsg))
queries.append("UPDATE jobs " \
"SET status='failed', " \
"details='Error while submitting job', " \
"updated_at='%s' " \
"WHERE id=%d" % \
(jobtracker.nowstr(), job_row['id']))
jobtracker.query(queries)
def submit(job_row):
"""
Submits a job to QueueManager, if successful will store returned queue id.
Input:
job_row: A row from the jobs table. The datafiles associated
with this job will be submitted to be processed.
Outputs:
None
"""
fns = pipeline_utils.get_fns_for_jobid(job_row['id'])
try:
presubmission_check(fns)
outdir = get_output_dir(fns)
# Attempt to submit the job
queue_id = config.jobpooler.queue_manager.submit\
(fns, outdir, job_row['id'])
except (FailedPreCheckError):
# Error caught during presubmission check.
exceptionmsgs = traceback.format_exception(*sys.exc_info())
errormsg = "Job ID: %d " % job_row['id']
errormsg += "failed presubmission check!\n\n"
errormsg += "".join(exceptionmsgs)
jobpool_cout.outs("Job ID: %d failed presubmission check!\n\t%s\n" % \
(job_row['id'], exceptionmsgs[-1]))
if config.email.send_on_terminal_failures:
# Send error email
msg = "Presubmission check failed!\n"
msg += "Job ID: %d\n\n" % \
(job_row['id'])
msg += errormsg
msg += "\n*** Job has been terminally failed. ***\n"
msg += "*** Job will NOT be re-submitted! ***\n"
if config.basic.delete_rawdata:
jobpool_cout.outs("Job #%d will NOT be retried. " \
"Data files will be deleted." % job_row['id'])
msg += "*** Raw data files will be deleted. ***\n"
else:
jobpool_cout.outs("Job #%d will NOT be retried. " %
job_row['id'])
notification = mailer.ErrorMailer(msg, \
subject="Job failed presubmission check - Terminal")
notification.send()
if config.basic.delete_rawdata:
pipeline_utils.clean_up(job_row['id'])
queries = []
arglist = []
queries.append("INSERT INTO job_submits (" \
"job_id, " \
"status, " \
"created_at, " \
"updated_at, " \
"details) " \
"VALUES (?, ?, ?, ?, ?)" )
arglist.append( ( job_row['id'], 'precheck_failed', \
jobtracker.nowstr(), jobtracker.nowstr(), \
errormsg) )
queries.append("UPDATE jobs " \
"SET status='terminal_failure', " \
"details='Failed presubmission check', " \
"updated_at=? " \
"WHERE id=?" )
arglist.append((jobtracker.nowstr(), job_row['id']))
jobtracker.execute(queries, arglist)
except (queue_managers.QueueManagerJobFatalError,\
datafile.DataFileError):
# Error caught during job submission.
exceptionmsgs = traceback.format_exception(*sys.exc_info())
errormsg = "Error while submitting job!\n"
errormsg += "\tJob ID: %d\n\n" % job_row['id']
errormsg += "".join(exceptionmsgs)
jobpool_cout.outs("Error while submitting job!\n" \
"\tJob ID: %d\n\t%s\n" % \
(job_row['id'], exceptionmsgs[-1]))
queries = []
arglist = []
queries.append("INSERT INTO job_submits (" \
"job_id, " \
"status, " \
"created_at, " \
"updated_at, " \
"details) " \
"VALUES (?, ?, ?, ?, ?)" )
arglist.append( ( job_row['id'], 'submission_failed', \
jobtracker.nowstr(), jobtracker.nowstr(), \
errormsg) )
queries.append("UPDATE jobs " \
"SET status='failed', " \
"details='Error while submitting job', " \
"updated_at=? " \
"WHERE id=?" )
arglist.append((jobtracker.nowstr(), job_row['id']))
jobtracker.execute(queries, arglist)
except queue_managers.QueueManagerNonFatalError:
# Do nothing. Don't submit the job. Don't mark the job as 'submitted'.
# Don't mark the job as 'failed'. The job submission will be retried.
pass
except queue_managers.QueueManagerFatalError:
# A fatal error occurred. Re-raise!
raise
else:
# No error occurred
msg = "Submitted job to process:\n"
msg += "\tJob ID: %d, Queue ID: %s\n" % (job_row['id'], queue_id)
msg += "\tData file(s):\n"
for fn in fns:
msg += "\t%s\n" % fn
jobpool_cout.outs(msg)
queries = []
queries.append("INSERT INTO job_submits (" \
"job_id, " \
"queue_id, " \
"output_dir, " \
"status, " \
"created_at, " \
"updated_at, " \
"details) " \
"VALUES (%d,'%s','%s','%s','%s','%s','%s')" % \
(job_row['id'], queue_id, outdir, 'running', \
jobtracker.nowstr(), jobtracker.nowstr(), \
'Job submitted to queue'))
queries.append("UPDATE jobs " \
"SET status='submitted', " \
"details='Job submitted to queue', " \
"updated_at='%s' " \
"WHERE id=%d" % \
(jobtracker.nowstr(), job_row['id']))
jobtracker.query(queries)
def submit(job_row):
"""
Submits a job to QueueManager, if successful will store returned queue id.
Input:
job_row: A row from the jobs table. The datafiles associated
with this job will be submitted to be processed.
Outputs:
None
"""
fns = pipeline_utils.get_fns_for_jobid(job_row['id'])
try:
outdir = get_output_dir(fns)
# Attempt to submit the job
queue_id = config.jobpooler.queue_manager.submit\
(fns, outdir, job_row['id'])
except (queue_managers.QueueManagerJobFatalError,\
datafile.DataFileError):
# Error caught during job submission.
exceptionmsgs = traceback.format_exception(*sys.exc_info())
errormsg = "Error while submitting job!\n"
errormsg += "\tJob ID: %d\n\n" % job_row['id']
errormsg += "".join(exceptionmsgs)
jobpool_cout.outs("Error while submitting job!\n" \
"\tJob ID: %d\n\t%s\n" % \
(job_row['id'], exceptionmsgs[-1]))
queries = []
arglist = []
queries.append("INSERT INTO job_submits (" \
"job_id, " \
"status, " \
"created_at, " \
"updated_at, " \
"details) " \
"VALUES (?, ?, ?, ?, ?)" )
arglist.append( ( job_row['id'], 'submission_failed', \
jobtracker.nowstr(), jobtracker.nowstr(), \
errormsg) )
queries.append("UPDATE jobs " \
"SET status='failed', " \
"details='Error while submitting job', " \
"updated_at=? " \
"WHERE id=?" )
arglist.append((jobtracker.nowstr(), job_row['id']))
jobtracker.execute(queries, arglist)
except queue_managers.QueueManagerNonFatalError:
# Do nothing. Don't submit the job. Don't mark the job as 'submitted'.
# Don't mark the job as 'failed'. The job submission will be retried.
pass
except queue_managers.QueueManagerFatalError:
# A fatal error occurred. Re-raise!
raise
else:
# No error occurred
msg = "Submitted job to process:\n"
msg += "\tJob ID: %d, Queue ID: %s\n" % (job_row['id'], queue_id)
msg += "\tData file(s):\n"
for fn in fns:
msg += "\t%s\n" % fn
jobpool_cout.outs(msg)
queries = []
queries.append("INSERT INTO job_submits (" \
"job_id, " \
"queue_id, " \
"output_dir, " \
"status, " \
"created_at, " \
"updated_at, " \
"details) " \
"VALUES (%d,'%s','%s','%s','%s','%s','%s')" % \
(job_row['id'], queue_id, outdir, 'running', \
jobtracker.nowstr(), jobtracker.nowstr(), \
'Job submitted to queue'))
queries.append("UPDATE jobs " \
"SET status='submitted', " \
"details='Job submitted to queue', " \
"updated_at='%s' " \
"WHERE id=%d" % \
(jobtracker.nowstr(), job_row['id']))
jobtracker.query(queries)
def submit(job_row):
"""
Submits a job to QueueManager, if successful will store returned queue id.
Input:
job_row: A row from the jobs table. The datafiles associated
with this job will be submitted to be processed.
Outputs:
None
"""
fns = pipeline_utils.get_fns_for_jobid(job_row["id"])
try:
outdir = get_output_dir(fns)
# Attempt to submit the job
queue_id = config.jobpooler.queue_manager.submit(fns, outdir, job_row["id"])
except (queue_managers.QueueManagerJobFatalError, datafile.DataFileError):
# Error caught during job submission.
exceptionmsgs = traceback.format_exception(*sys.exc_info())
errormsg = "Error while submitting job!\n"
errormsg += "\tJob ID: %d\n\n" % job_row["id"]
errormsg += "".join(exceptionmsgs)
jobpool_cout.outs("Error while submitting job!\n" "\tJob ID: %d\n\t%s\n" % (job_row["id"], exceptionmsgs[-1]))
queries = []
arglist = []
queries.append(
"INSERT INTO job_submits ("
"job_id, "
"status, "
"created_at, "
"updated_at, "
"details) "
"VALUES (?, ?, ?, ?, ?)"
)
arglist.append((job_row["id"], "submission_failed", jobtracker.nowstr(), jobtracker.nowstr(), errormsg))
queries.append(
"UPDATE jobs " "SET status='failed', " "details='Error while submitting job', " "updated_at=? " "WHERE id=?"
)
arglist.append((jobtracker.nowstr(), job_row["id"]))
jobtracker.execute(queries, arglist)
except queue_managers.QueueManagerNonFatalError:
# Do nothing. Don't submit the job. Don't mark the job as 'submitted'.
# Don't mark the job as 'failed'. The job submission will be retried.
pass
except queue_managers.QueueManagerFatalError:
# A fatal error occurred. Re-raise!
raise
else:
# No error occurred
msg = "Submitted job to process:\n"
msg += "\tJob ID: %d, Queue ID: %s\n" % (job_row["id"], queue_id)
msg += "\tData file(s):\n"
for fn in fns:
msg += "\t%s\n" % fn
jobpool_cout.outs(msg)
queries = []
queries.append(
"INSERT INTO job_submits ("
"job_id, "
"queue_id, "
"output_dir, "
"status, "
"created_at, "
"updated_at, "
"details) "
"VALUES (%d,'%s','%s','%s','%s','%s','%s')"
% (
job_row["id"],
queue_id,
outdir,
"running",
jobtracker.nowstr(),
jobtracker.nowstr(),
"Job submitted to queue",
)
)
queries.append(
"UPDATE jobs "
"SET status='submitted', "
"details='Job submitted to queue', "
"updated_at='%s' "
"WHERE id=%d" % (jobtracker.nowstr(), job_row["id"])
)
jobtracker.query(queries)
def submit(job_row):
"""
Submits a job to QueueManager, if successful will store returned queue id.
Input:
job_row: A row from the jobs table. The datafiles associated
with this job will be submitted to be processed.
Outputs:
None
"""
fns = pipeline_utils.get_fns_for_jobid(job_row["id"])
script = os.path.join(config.basic.pipelinedir, "bin", "%s_search.py" % config.basic.survey)
# Specify requested resources for job submission
if job_row["task"] == "rfifind":
res = [4 * 60 * 60, 1024, 25]
elif "search" in job_row["task"]:
res = [165240, 1024, 28] # 45.9 hrs
elif job_row["task"] == "sifting": # Sifting should be quick
res = [30 * 60, 256, 5]
elif "folding" in job_row["task"]:
res = [96 * 60 * 60, 3000, 28]
# elif job_row['task']=='tidyup':
# res = [30*60, 256, 5]
options = job_row["task"]
try:
SPAN512_job.presubmission_check(fns)
outdir = SPAN512_job.get_output_dir(fns)
# Attempt to submit the job
queue_id = config.jobpooler.queue_manager.submit(
fns, outdir, job_row["id"], resources=res, script=script, opts=options
)
except (FailedPreCheckError):
# Error caught during presubmission check.
exceptionmsgs = traceback.format_exception(*sys.exc_info())
errormsg = "Job ID: %d " % job_row["id"]
errormsg += "failed presubmission check!\n\n"
errormsg += "".join(exceptionmsgs)
jobpool_cout.outs("Job ID: %d failed presubmission check!\n\t%s\n" % (job_row["id"], exceptionmsgs[-1]))
if config.email.send_on_terminal_failures:
# Send error email
msg = "Presubmission check failed!\n"
msg += "Job ID: %d\n\n" % (job_row["id"])
msg += errormsg
msg += "\n*** Job has been terminally failed. ***\n"
msg += "*** Job will NOT be re-submitted! ***\n"
if config.basic.delete_rawdata:
jobpool_cout.outs("Job #%d will NOT be retried. " "Data files will be deleted." % job_row["id"])
msg += "*** Raw data files will be deleted. ***\n"
else:
jobpool_cout.outs("Job #%d will NOT be retried. " % job_row["id"])
notification = mailer.ErrorMailer(msg, subject="Job failed presubmission check - Terminal")
notification.send()
if config.basic.delete_rawdata:
pipeline_utils.clean_up(job_row["id"])
queries = []
arglist = []
queries.append(
"INSERT INTO job_submits ("
"job_id, "
"status, "
"created_at, "
"updated_at, "
"details) "
"VALUES (%d, %s, '%s', '%s', %s)"
% (job_row["id"], "precheck_failed", jobtracker.nowstr(), jobtracker.nowstr(), errormsg)
)
queries.append(
"UPDATE jobs "
"SET status='terminal_failure', "
"details='Failed presubmission check', "
"updated_at='%s'"
"WHERE id=%d" % (jobtracker.nowstr(), job_row["id"])
)
jobtracker.query(queries)
except (queue_managers.QueueManagerJobFatalError, datafile.DataFileError):
# Error caught during job submission.
exceptionmsgs = traceback.format_exception(*sys.exc_info())
errormsg = "Error while submitting job!\n"
errormsg += "\tJob ID: %d\n\n" % job_row["id"]
errormsg += "".join(exceptionmsgs)
jobpool_cout.outs("Error while submitting job!\n" "\tJob ID: %d\n\t%s\n" % (job_row["id"], exceptionmsgs[-1]))
queries = []
arglist = []
queries.append(
"INSERT INTO job_submits ("
"job_id, "
"status, "
"created_at, "
"updated_at, "
"details) "
"VALUES (%d, %s, '%s', '%s', %s)"
% (job_row["id"], "submission_failed", jobtracker.nowstr(), jobtracker.nowstr(), errormsg)
)
queries.append(
"UPDATE jobs "
"SET status='failed', "
"details='Error while submitting job', "
"updated_at='%s' "
"WHERE id=%d" % (jobtracker.nowstr(), job_row["id"])
)
jobtracker.execute(queries)
except queue_managers.QueueManagerNonFatalError:
# Do nothing. Don't submit the job. Don't mark the job as 'submitted'.
# Don't mark the job as 'failed'. The job submission will be retried.
pass
except queue_managers.QueueManagerFatalError:
# A fatal error occurred. Re-raise!
raise
else:
# No error occurred
msg = "Submitted job to process:\n"
msg += "\tJob ID: %d, Queue ID: %s\n" % (job_row["id"], queue_id)
msg += "\tData file(s):\n"
for fn in fns:
msg += "\t%s\n" % fn
jobpool_cout.outs(msg)
queries = []
queries.append(
"INSERT INTO job_submits ("
"job_id, "
"queue_id, "
"output_dir, "
"status, "
"created_at, "
"updated_at, "
"details) "
"VALUES (%d,'%s','%s','%s','%s','%s','%s')"
% (
job_row["id"],
queue_id,
outdir,
"running",
jobtracker.nowstr(),
jobtracker.nowstr(),
"Job submitted to queue",
)
)
queries.append(
"UPDATE jobs "
"SET status='submitted', "
"details='Job submitted to queue', "
"updated_at='%s' "
"WHERE id=%d" % (jobtracker.nowstr(), job_row["id"])
)
jobtracker.query(queries)
