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() */
char *get_next_retryable_jobid(
int *iter)
{
job_exiting_retry_info *jeri;
job *pjob;
time_t time_now = time(NULL);
char log_buf[LOCAL_LOG_BUF_SIZE];
pthread_mutex_lock(exiting_jobs_info->hm_mutex);
mutex_mgr exit_mgr(exiting_jobs_info->hm_mutex, true);
while ((jeri = (job_exiting_retry_info *)next_from_hash_map(exiting_jobs_info, iter, true)) != NULL)
{
if (time_now - jeri->last_attempt > EXITING_RETRY_TIME)
{
if (jeri->attempts >= MAX_EXITING_RETRY_ATTEMPTS)
{
std::string jid(jeri->jobid);
remove_from_hash_map(exiting_jobs_info, jeri->jobid, true);
free(jeri);
exit_mgr.unlock(); //Don't hold on to a mutex when trying to lock another.
if ((pjob = svr_find_job((char *)jid.c_str(), TRUE)) != NULL)
{
snprintf(log_buf, sizeof(log_buf), "Job %s has had its exiting re-tried %d times, purging.",
jeri->jobid, MAX_EXITING_RETRY_ATTEMPTS);
log_event(PBSEVENT_JOB, PBS_EVENTCLASS_JOB, __func__, log_buf);
force_purge_work(pjob);
}
exit_mgr.lock();
}
else
{
jeri->attempts++;
jeri->last_attempt = time_now;
char *jobid = strdup(jeri->jobid);
return(jobid);
}
}
}
return(NULL);
} /* END get_next_retryable_jobid() */
job_array *next_array(
int *iter)
{
job_array *pa = NULL;
pthread_mutex_lock(allarrays.allarrays_mutex);
pa = (job_array *)next_from_hash_map(allarrays.hm, iter);
if (pa != NULL)
lock_ai_mutex(pa, __func__, NULL, LOGLEVEL);
pthread_mutex_unlock(allarrays.allarrays_mutex);
return(pa);
} /* END next_array() */
