/*
* check_exiting_jobs()
*
* loops over the recorded exiting job information and retries
* jobs that have been stale long enough.
*/
int check_exiting_jobs()
{
exiting_jobs_info_iterator *iter = NULL;
char *jobid;
job *pjob;
while ((jobid = get_next_retryable_jobid(&iter)) != NULL)
{
if ((pjob = svr_find_job(jobid, TRUE)) == NULL)
{
remove_from_exiting_list_by_jobid(jobid);
free(jobid);
}
else
{
mutex_mgr pjob_mutex = mutex_mgr(pjob->ji_mutex, true);
if (pjob->ji_qs.ji_state == JOB_STATE_COMPLETE)
{
remove_from_exiting_list_by_jobid(jobid);
free(jobid);
}
else
{
pjob_mutex.unlock();
/* jobid is freed in on_job_exit() */
retry_job_exit(jobid);
}
}
} /* END loop over exiting job information */
return(PBSE_NONE);
} /* END check_exiting_jobs() */
int check_exiting_jobs()
{
int iter = -1;
job_exiting_retry_info *jeri;
job *pjob;
time_t time_now = time(NULL);
while ((jeri = (job_exiting_retry_info *)next_from_hash_map(exiting_jobs_info, &iter)) != NULL)
{
if (time_now - jeri->last_attempt > EXITING_RETRY_TIME)
{
if ((pjob = svr_find_job(jeri->jobid, TRUE)) == NULL)
{
remove_entry_from_exiting_list(jeri);
}
else
{
if (pjob->ji_qs.ji_state == JOB_STATE_COMPLETE)
{
remove_entry_from_exiting_list(jeri);
unlock_ji_mutex(pjob, __func__, NULL, LOGLEVEL);
}
else
{
unlock_ji_mutex(pjob, __func__, NULL, LOGLEVEL);
retry_job_exit(jeri);
}
}
}
}
return(PBSE_NONE);
} /* END check_exiting_jobs() */
END_TEST
START_TEST(retry_job_exit_test)
{
fail_unless(retry_job_exit(strdup("1.napali")) == 0, "Didn't retry job");
}