Hook for HTCondor job_router that removes its job from a route

htc_exit_hook.py

#!/usr/bin/python
from __future__ import print_function, with_statement
import sys
import ast
import subprocess
import time

route_test_performed, want_route_test = False, False

# read input so router does not stall
# we only need ClusterId and ProcId, but the others are useful for
# debugging and should be defined
for line in sys.stdin:
    if not line.strip():
        continue
    key, value = (elem.strip() for elem in line.split('=', 1))
    if key.lower() == "routedfromjobid":
        real_job_id = ast.literal_eval(value)  # "78249.0"
    elif key.lower() == "routename":
        route_name = ast.literal_eval(value)  # "MyRoute"
    elif key.lower() == "routetestperformed":
        route_test_performed = True
    elif key.lower() == "clusterid":
        cluster_id = ast.literal_eval(value)  # 78251
    elif key.lower() == "procid":
        proc_id = ast.literal_eval(value)  # 3
    elif key.lower() == "wantroutetest":
        want_route_test = value

if not want_route_test:
    sys.exit(0)

try:
    assert cluster_id and proc_id and real_job_id and route_name
except NameError as err:
    print(err, file=sys.stderr)
    sys.exit(0)

job_id = '%d.%d' % (cluster_id, proc_id)
assert job_id != real_job_id, 'Somebody set us up the inplace transformed job! Abandon hooks!'

if route_test_performed:
    with open('/tmp/htc_exit_hook.log', 'a') as log_file:
        log_file.write('[%s] job %s => %s survived previous route test\n' % (time.asctime(), job_id, real_job_id))

with open('/tmp/htc_exit_hook.log', 'a') as log_file:
    log_file.write('[%s] job %s => %s about to be removed\n' % (time.asctime(), job_id, real_job_id))

# mark us as routed already
print('RouteRestPerformed = True')

# remove the ROUTED job
condor_rm = subprocess.Popen(('condor_rm', str(job_id)), stdout=subprocess.PIPE, stderr=subprocess.PIPE)

stderr, stdout = condor_rm.communicate()

if condor_rm.poll() == 0:
    with open('/tmp/htc_exit_hook.log', 'a') as log_file:
        log_file.write('[%s] job %s => %s marked for removal\n' % (time.asctime(), job_id, real_job_id))
else:
    with open('/tmp/htc_exit_hook.log', 'a') as log_file:
        log_file.write('[%s] job %s => %s removal failed: %s\n' % (time.asctime(), job_id, real_job_id, condor_rm.poll()))
    print('rmexit: ', condor_rm.poll(), file=sys.stderr)
    print('stdout:\n', stdout, file=sys.stderr)
    print('stderr:\n', stderr, file=sys.stderr)