Attachment #220 for ticket #187

View | Details | Raw Unified | Return to ticket 187
Collapse All | Expand All




{
	static int launch_type_poe = -1;
	int error_code = SLURM_SUCCESS, i, qos_error;
	enum job_state_reason fail_reason;
	struct part_record *part_ptr = NULL;
	List part_ptr_list = NULL;
	bitstr_t *req_bitmap = NULL, *exc_bitmap = NULL;

	}
	job_ptr->best_switch = true;

	/* Insure that requested partition is valid right now,
	 * otherwise leave job queued and provide warning code */
	fail_reason = job_limits_check(&job_ptr);
	if (fail_reason != WAIT_NO_REASON) {
		if (fail_reason == WAIT_QOS_THRES)
			error_code = ESLURM_QOS_THRES;
		else
			error_code = ESLURM_REQUESTED_PART_CONFIG_UNAVAILABLE;
		job_ptr->state_reason = fail_reason;
		xfree(job_ptr->state_desc);
	}

	FREE_NULL_LIST(license_list);
	FREE_NULL_BITMAP(req_bitmap);
	FREE_NULL_BITMAP(exc_bitmap);

Lines 1455-1468 extern int select_nodes(struct job_record *job_ptr, bool test_only, Link Here

(-)a/src/slurmctld/node_scheduler.c (-8 / +13 lines)
1455	return ESLURM_JOB_HELD;	1455	return ESLURM_JOB_HELD;
1456	}	1456	}
1457		1457
1458	/* Confirm that partition is up and has compatible nodes limits */	1458	/* Confirm that partition is up and has compatible nodes limits.
		1459	* If not, we still continue to determine if the job can ever run.
		1460	* For example, the geometry might be incompatible with BlueGene. */
1459	fail_reason = job_limits_check(&job_ptr);	1461	fail_reason = job_limits_check(&job_ptr);
1460	if (fail_reason != WAIT_NO_REASON) {	1462	if (fail_reason != WAIT_NO_REASON)
1461	last_job_update = now;	1463	test_only = true;
1462	xfree(job_ptr->state_desc);
1463	job_ptr->state_reason = fail_reason;
1464	return ESLURM_REQUESTED_PART_CONFIG_UNAVAILABLE;
1465	}
1466		1464
1467	/* build sets of usable nodes based upon their configuration */	1465	/* build sets of usable nodes based upon their configuration */
1468	error_code = _build_node_list(job_ptr, &node_set_ptr, &node_set_size);	1466	error_code = _build_node_list(job_ptr, &node_set_ptr, &node_set_size);
Lines 1551-1557 extern int select_nodes(struct job_record *job_ptr, bool test_only, Link Here
1551	}	1549	}
1552	}	1550	}
1553	}	1551	}
1554	if (error_code) {	1552	if (error_code \|\| fail_reason) {
		1553	/* Fatal errors for job here */
1555	if (error_code == ESLURM_REQUESTED_PART_CONFIG_UNAVAILABLE) {	1554	if (error_code == ESLURM_REQUESTED_PART_CONFIG_UNAVAILABLE) {
1556	/* Too many nodes requested */	1555	/* Too many nodes requested */
1557	debug3("JobId=%u not runnable with present config",	1556	debug3("JobId=%u not runnable with present config",
Lines 1559-1564 extern int select_nodes(struct job_record *job_ptr, bool test_only, Link Here
1559	job_ptr->state_reason = WAIT_PART_NODE_LIMIT;	1558	job_ptr->state_reason = WAIT_PART_NODE_LIMIT;
1560	xfree(job_ptr->state_desc);	1559	xfree(job_ptr->state_desc);
1561	last_job_update = now;	1560	last_job_update = now;
		1561	} else if (fail_reason) {
		1562	job_ptr->state_reason = fail_reason;
		1563	xfree(job_ptr->state_desc);
		1564	last_job_update = now;
		1565
		1566	/* Non-fatal errors for job below */
1562	} else if (error_code == ESLURM_NODE_NOT_AVAIL) {	1567	} else if (error_code == ESLURM_NODE_NOT_AVAIL) {
1563	/* Required nodes are down or drained */	1568	/* Required nodes are down or drained */
1564	debug3("JobId=%u required nodes not avail",	1569	debug3("JobId=%u required nodes not avail",

Return to ticket 187

Lines 3709-3715 static int _job_create(job_desc_msg_t * job_desc, int allocate, int will_run, Link Here

(-)a/src/slurmctld/job_mgr.c (-13 lines)
3709	{	3709	{
3710	static int launch_type_poe = -1;	3710	static int launch_type_poe = -1;
3711	int error_code = SLURM_SUCCESS, i, qos_error;	3711	int error_code = SLURM_SUCCESS, i, qos_error;
3712	enum job_state_reason fail_reason;
3713	struct part_record *part_ptr = NULL;	3712	struct part_record *part_ptr = NULL;
3714	List part_ptr_list = NULL;	3713	List part_ptr_list = NULL;
3715	bitstr_t req_bitmap = NULL, exc_bitmap = NULL;	3714	bitstr_t req_bitmap = NULL, exc_bitmap = NULL;
Lines 4100-4117 static int _job_create(job_desc_msg_t * job_desc, int allocate, int will_run, Link Here
4100	}	4099	}
4101	job_ptr->best_switch = true;	4100	job_ptr->best_switch = true;
4102		4101
4103	/* Insure that requested partition is valid right now,
4104	* otherwise leave job queued and provide warning code */
4105	fail_reason = job_limits_check(&job_ptr);
4106	if (fail_reason != WAIT_NO_REASON) {
4107	if (fail_reason == WAIT_QOS_THRES)
4108	error_code = ESLURM_QOS_THRES;
4109	else
4110	error_code = ESLURM_REQUESTED_PART_CONFIG_UNAVAILABLE;
4111	job_ptr->state_reason = fail_reason;
4112	xfree(job_ptr->state_desc);
4113	}
4114
4115	FREE_NULL_LIST(license_list);	4102	FREE_NULL_LIST(license_list);
4116	FREE_NULL_BITMAP(req_bitmap);	4103	FREE_NULL_BITMAP(req_bitmap);
4117	FREE_NULL_BITMAP(exc_bitmap);	4104	FREE_NULL_BITMAP(exc_bitmap);