Attachment #5627 for ticket #4434

View | Details | Raw Unified | Return to ticket 4434
Collapse All | Expand All




		     req->job_id);
	} else {
		hostset_t step_hset = hostset_create(req->nodes);
		int rc;

		debug3("%s: call to _forkexec_slurmstepd", __func__);
		rc = _forkexec_slurmstepd(
			LAUNCH_TASKS, (void *)launch_req, cli,
			&self, step_hset, msg->protocol_version);
		debug3("%s: return from _forkexec_slurmstepd %d", __func__, rc);

		if (rc != SLURM_SUCCESS)
			_launch_job_fail(req->job_id, rc);

		if (step_hset)
			hostset_destroy(step_hset);
	}




	int nodeid;
	char *name = NULL;

	/*
	 * The extern step can get here if something goes wrong starting the
	 * step.  If this does happen we don't have to contact the srun since
	 * there isn't one, just return.
	 */
	if ((msg->job_step_id == SLURM_EXTERN_CONT) ||
	    !msg->resp_port || !msg->num_resp_port) {
		debug2("%s: The extern step has nothing to send a launch failure to",
		       __func__);
		return;
	}

#ifndef HAVE_FRONT_END
	nodeid = nodelist_find(msg->complete_nodelist, conf->node_name);
	name = xstrdup(conf->node_name);

Return to ticket 4434

Lines 2067-2078 static void _spawn_prolog_stepd(slurm_msg_t *msg) Link Here

(-)a/src/slurmd/slurmd/req.c (-2 / +7 lines)
2067	req->job_id);	2067	req->job_id);
2068	} else {	2068	} else {
2069	hostset_t step_hset = hostset_create(req->nodes);	2069	hostset_t step_hset = hostset_create(req->nodes);
		2070	int rc;
2070		2071
2071	debug3("%s: call to _forkexec_slurmstepd", __func__);	2072	debug3("%s: call to _forkexec_slurmstepd", __func__);
2072	(void) _forkexec_slurmstepd(	2073	rc = _forkexec_slurmstepd(
2073	LAUNCH_TASKS, (void *)launch_req, cli,	2074	LAUNCH_TASKS, (void *)launch_req, cli,
2074	&self, step_hset, msg->protocol_version);	2075	&self, step_hset, msg->protocol_version);
2075	debug3("%s: return from _forkexec_slurmstepd", __func__);	2076	debug3("%s: return from _forkexec_slurmstepd %d", __func__, rc);
		2077
		2078	if (rc != SLURM_SUCCESS)
		2079	_launch_job_fail(req->job_id, rc);
		2080
2076	if (step_hset)	2081	if (step_hset)
2077	hostset_destroy(step_hset);	2082	hostset_destroy(step_hset);
2078	}	2083	}

Lines 2301-2306 _send_launch_failure(launch_tasks_request_msg_t msg, slurm_addr_t cli, int rc, Link Here

(-)a/src/slurmd/slurmstepd/mgr.c (+12 lines)
2301	int nodeid;	2301	int nodeid;
2302	char *name = NULL;	2302	char *name = NULL;
2303		2303
		2304	/*
		2305	* The extern step can get here if something goes wrong starting the
		2306	* step. If this does happen we don't have to contact the srun since
		2307	* there isn't one, just return.
		2308	*/
		2309	if ((msg->job_step_id == SLURM_EXTERN_CONT) \|\|
		2310	!msg->resp_port \|\| !msg->num_resp_port) {
		2311	debug2("%s: The extern step has nothing to send a launch failure to",
		2312	__func__);
		2313	return;
		2314	}
		2315
2304	#ifndef HAVE_FRONT_END	2316	#ifndef HAVE_FRONT_END
2305	nodeid = nodelist_find(msg->complete_nodelist, conf->node_name);	2317	nodeid = nodelist_find(msg->complete_nodelist, conf->node_name);
2306	name = xstrdup(conf->node_name);	2318	name = xstrdup(conf->node_name);