Attachment #4557 for ticket #3763




	struct timeval tv_now = {0, 0};
	long delta_t;

	if (purge_quit) {
		info("job %u not purged, job purging all done for this iteration", job_ptr->job_id);
		return 0;
	}
	gettimeofday(&tv_now, NULL);
	delta_t  = (tv_now.tv_sec - purge_start_time.tv_sec) * 1000000;
	delta_t += tv_now.tv_usec;
	delta_t -= purge_start_time.tv_usec;
	if (delta_t > 1000000) {
		info("job %u not purged, purging timeout reached", job_ptr->job_id);
		purge_quit = true;
		return 0;
	}

			job_ptr->time_last_active = now;
			re_kill_job(job_ptr);
		}
		info("job %u not purged, job still completing", job_ptr->job_id);
		return 0;       /* Job still completing */
	}

	if (job_ptr->epilog_running) {
		info("job %u not purged, EpilogSlurmctld still running", job_ptr->job_id);
		return 0;       /* EpilogSlurmctld still running */
	}

	if (slurmctld_conf.min_job_age == 0) {
		info("job %u not purged, no job record purging", job_ptr->job_id);
		return 0;	/* No job record purging */
	}

	min_age  = now - slurmctld_conf.min_job_age;
	if (fed_mgr_is_origin_job(job_ptr) &&
	    (job_ptr->end_time > min_age)) {
		info("job %u not purged, too new to purge", job_ptr->job_id);
		return 0;	/* Too new to purge */
	}

	if (!(IS_JOB_COMPLETED(job_ptr))) {
		info("job %u not purged, job still active", job_ptr->job_id);
		return 0;	/* Job still active */
	}

	if (job_ptr->step_list && list_count(job_ptr->step_list)) {
		debug("Job %u still has %d active steps",

		if (job_ptr->array_recs->tot_run_tasks ||
		    !test_job_array_completed(job_ptr->array_job_id)) {
			/* Some tasks from this job array still active */
			info("job %u not purged, some tasks from this job array still active",
			     job_ptr->job_id);
			return 0;
		}
	}

	select_g_select_jobinfo_get(job_ptr->select_jobinfo,
				    SELECT_JOBDATA_CLEANING,
				    &cleaning);
	if (cleaning) {
		info("job %u not purged, job hasn't finished yet", job_ptr->job_id);
		return 0;      /* Job hasn't finished yet */
	}

	if (bb_g_job_test_stage_out(job_ptr) != 1) {
		info("job %u not purged, stage out in progress", job_ptr->job_id);
		return 0;      /* Stage out in progress */
	}

	/* If we don't have a db_index by now and we are running with
	 * the slurmdbd, lets put it on the list to be handled later

	if (with_slurmdbd && !job_ptr->db_index)
		jobacct_storage_g_job_start(acct_db_conn, job_ptr);

	info("purging job %u", job_ptr->job_id);
	return 1;		/* Purge the job */
}


Return to ticket 3763

Lines 8455-8467 static int _list_find_job_old(void job_entry, void key) Link Here

(-)a/src/slurmctld/job_mgr.c (-7 / +26 lines)
8455	struct timeval tv_now = {0, 0};	8455	struct timeval tv_now = {0, 0};
8456	long delta_t;	8456	long delta_t;
8457		8457
8458	if (purge_quit)	8458	if (purge_quit) {
		8459	info("job %u not purged, job purging all done for this iteration", job_ptr->job_id);
8459	return 0;	8460	return 0;
		8461	}
8460	gettimeofday(&tv_now, NULL);	8462	gettimeofday(&tv_now, NULL);
8461	delta_t = (tv_now.tv_sec - purge_start_time.tv_sec) * 1000000;	8463	delta_t = (tv_now.tv_sec - purge_start_time.tv_sec) * 1000000;
8462	delta_t += tv_now.tv_usec;	8464	delta_t += tv_now.tv_usec;
8463	delta_t -= purge_start_time.tv_usec;	8465	delta_t -= purge_start_time.tv_usec;
8464	if (delta_t > 1000000) {	8466	if (delta_t > 1000000) {
		8467	info("job %u not purged, purging timeout reached", job_ptr->job_id);
8465	purge_quit = true;	8468	purge_quit = true;
8466	return 0;	8469	return 0;
8467	}	8470	}
Lines 8473-8494 static int _list_find_job_old(void job_entry, void key) Link Here
8473	job_ptr->time_last_active = now;	8476	job_ptr->time_last_active = now;
8474	re_kill_job(job_ptr);	8477	re_kill_job(job_ptr);
8475	}	8478	}
		8479	info("job %u not purged, job still completing", job_ptr->job_id);
8476	return 0; /* Job still completing */	8480	return 0; /* Job still completing */
8477	}	8481	}
8478		8482
8479	if (job_ptr->epilog_running)	8483	if (job_ptr->epilog_running) {
		8484	info("job %u not purged, EpilogSlurmctld still running", job_ptr->job_id);
8480	return 0; /* EpilogSlurmctld still running */	8485	return 0; /* EpilogSlurmctld still running */
		8486	}
8481		8487
8482	if (slurmctld_conf.min_job_age == 0)	8488	if (slurmctld_conf.min_job_age == 0) {
		8489	info("job %u not purged, no job record purging", job_ptr->job_id);
8483	return 0; /* No job record purging */	8490	return 0; /* No job record purging */
		8491	}
8484		8492
8485	min_age = now - slurmctld_conf.min_job_age;	8493	min_age = now - slurmctld_conf.min_job_age;
8486	if (fed_mgr_is_origin_job(job_ptr) &&	8494	if (fed_mgr_is_origin_job(job_ptr) &&
8487	(job_ptr->end_time > min_age))	8495	(job_ptr->end_time > min_age)) {
		8496	info("job %u not purged, too new to purge", job_ptr->job_id);
8488	return 0; /* Too new to purge */	8497	return 0; /* Too new to purge */
		8498	}
8489		8499
8490	if (!(IS_JOB_COMPLETED(job_ptr)))	8500	if (!(IS_JOB_COMPLETED(job_ptr))) {
		8501	info("job %u not purged, job still active", job_ptr->job_id);
8491	return 0; /* Job still active */	8502	return 0; /* Job still active */
		8503	}
8492		8504
8493	if (job_ptr->step_list && list_count(job_ptr->step_list)) {	8505	if (job_ptr->step_list && list_count(job_ptr->step_list)) {
8494	debug("Job %u still has %d active steps",	8506	debug("Job %u still has %d active steps",
Lines 8500-8505 static int _list_find_job_old(void job_entry, void key) Link Here
8500	if (job_ptr->array_recs->tot_run_tasks \|\|	8512	if (job_ptr->array_recs->tot_run_tasks \|\|
8501	!test_job_array_completed(job_ptr->array_job_id)) {	8513	!test_job_array_completed(job_ptr->array_job_id)) {
8502	/* Some tasks from this job array still active */	8514	/* Some tasks from this job array still active */
		8515	info("job %u not purged, some tasks from this job array still active",
		8516	job_ptr->job_id);
8503	return 0;	8517	return 0;
8504	}	8518	}
8505	}	8519	}
Lines 8507-8517 static int _list_find_job_old(void job_entry, void key) Link Here
8507	select_g_select_jobinfo_get(job_ptr->select_jobinfo,	8521	select_g_select_jobinfo_get(job_ptr->select_jobinfo,
8508	SELECT_JOBDATA_CLEANING,	8522	SELECT_JOBDATA_CLEANING,
8509	&cleaning);	8523	&cleaning);
8510	if (cleaning)	8524	if (cleaning) {
		8525	info("job %u not purged, job hasn't finished yet", job_ptr->job_id);
8511	return 0; /* Job hasn't finished yet */	8526	return 0; /* Job hasn't finished yet */
		8527	}
8512		8528
8513	if (bb_g_job_test_stage_out(job_ptr) != 1)	8529	if (bb_g_job_test_stage_out(job_ptr) != 1) {
		8530	info("job %u not purged, stage out in progress", job_ptr->job_id);
8514	return 0; /* Stage out in progress */	8531	return 0; /* Stage out in progress */
		8532	}
8515		8533
8516	/* If we don't have a db_index by now and we are running with	8534	/* If we don't have a db_index by now and we are running with
8517	* the slurmdbd, lets put it on the list to be handled later	8535	* the slurmdbd, lets put it on the list to be handled later
Lines 8521-8526 static int _list_find_job_old(void job_entry, void key) Link Here
8521	if (with_slurmdbd && !job_ptr->db_index)	8539	if (with_slurmdbd && !job_ptr->db_index)
8522	jobacct_storage_g_job_start(acct_db_conn, job_ptr);	8540	jobacct_storage_g_job_start(acct_db_conn, job_ptr);
8523		8541
		8542	info("purging job %u", job_ptr->job_id);
8524	return 1; /* Purge the job */	8543	return 1; /* Purge the job */
8525	}	8544	}
8526		8545