Attachment #28340 for ticket #15357

View | Details | Raw Unified | Return to ticket 15357
Collapse All | Expand All

Lines 4693-4698 second argument. Link Here

(-)a/doc/man/man5/slurm.conf.5 (+9 lines)
4693	.TP	4693	.TP
4694	\fBuser_resv_delete\fR	4694	\fBuser_resv_delete\fR
4695	Allow any user able to run in a reservation to delete it.	4695	Allow any user able to run in a reservation to delete it.
		4696	.IP
		4697
		4698	.TP
		4699	\fBvalidate_nodeaddr_threads=\fR
		4700	During startup, slurmctld looks up the address for each compute node in the
		4701	system. On large systems this can cause considerable delay, this option permits
		4702	the slurmctld to concurrently handle the lookup calls and can reduce system
		4703	startup time considerably. The default value is 1. Maximum permitted value is
		4704	64.
4696	.RE	4705	.RE
4697	.IP	4706	.IP
4698		4707




	return 0;
}

static void *_set_node_addrs(void *arg)
{
	list_t *nodes = arg;
	slurm_addr_t slurm_addr;
	node_record_t *node_ptr;

	while ((node_ptr = list_pop(nodes))) {
		slurm_set_addr(&slurm_addr, node_ptr->port,
			       node_ptr->comm_name);
		if (slurm_get_port(&slurm_addr))
			continue;
		error("%s: failure on %s", __func__, node_ptr->comm_name);
		node_ptr->node_state = NODE_STATE_FUTURE;
		node_ptr->port = 0;
		xfree(node_ptr->reason);
		node_ptr->reason = xstrdup("NO NETWORK ADDRESS FOUND");
		node_ptr->reason_time = time(NULL);
		node_ptr->reason_uid = slurm_conf.slurm_user_id;
	}

	return NULL;
}

/*
 * Validate that nodes are addressable.
 */

{
#ifndef HAVE_FRONT_END
	node_record_t *node_ptr;
	slurm_addr_t slurm_addr;
	DEF_TIMERS;
	pthread_t *work_threads;
	int threads_num = 1;
	char *temp_str;
	list_t *nodes = list_create(NULL);
	xassert(verify_lock(CONF_LOCK, READ_LOCK));

	START_TIMER;

	if ((temp_str = xstrcasestr(slurm_conf.slurmctld_params,
				    "validate_nodeaddr_threads="))) {
		int tmp_val = strtol(temp_str + 26, NULL, 10);
		if ((tmp_val >= 1) && (tmp_val <= 64))
			threads_num = tmp_val;
		else
			error("SlurmctldParameters option validate_nodeaddr_threads=%d out of range, ignored",
			      tmp_val);
	}


	for (int i = 0; (node_ptr = next_node(&i)); i++) {
		if ((node_ptr->name == NULL) ||
		    (node_ptr->name[0] == '\0'))

				continue;
		if (node_ptr->port == 0)
			node_ptr->port = slurm_conf.slurmd_port;
		list_append(nodes, node_ptr);
			       node_ptr->comm_name);
		if (slurm_get_port(&slurm_addr))
			continue;
		error("%s: failure on %s", __func__, node_ptr->comm_name);
		node_ptr->node_state = NODE_STATE_FUTURE;
		node_ptr->port = 0;
		xfree(node_ptr->reason);
		node_ptr->reason = xstrdup("NO NETWORK ADDRESS FOUND");
		node_ptr->reason_time = time(NULL);
		node_ptr->reason_uid = slurm_conf.slurm_user_id;
	}

	work_threads = xcalloc(threads_num, sizeof(pthread_t));
	for (int i = 0; i < threads_num; i++)
		slurm_thread_create(&work_threads[i], _set_node_addrs, nodes);
	for (int i = 0; i < threads_num; i++)
		pthread_join(work_threads[i], NULL);
	xfree(work_threads);
	xassert(list_is_empty(nodes));
	FREE_NULL_LIST(nodes);

	END_TIMER2("_validate_slurmd_addr");
#endif
}
- 
--
src/slurmctld/read_config.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)

Lines 444-450 static int _set_share_node_bitmap(void x, void arg) Link Here

(-)a/src/slurmctld/read_config.c (-3 / +2 lines)
444		444
445	static void _set_node_addrs(void arg)	445	static void _set_node_addrs(void arg)
446	{	446	{
447	list_t *nodes = arg;	447	List nodes = arg;
448	slurm_addr_t slurm_addr;	448	slurm_addr_t slurm_addr;
449	node_record_t *node_ptr;	449	node_record_t *node_ptr;
450		450
Lines 476-482 static void _validate_slurmd_addr(void) Link Here
476	pthread_t *work_threads;	476	pthread_t *work_threads;
477	int threads_num = 1;	477	int threads_num = 1;
478	char *temp_str;	478	char *temp_str;
479	list_t *nodes = list_create(NULL);	479	List nodes = list_create(NULL);
480	xassert(verify_lock(CONF_LOCK, READ_LOCK));	480	xassert(verify_lock(CONF_LOCK, READ_LOCK));
481		481
482	START_TIMER;	482	START_TIMER;
483	-

Return to ticket 15357

Lines 442-447 static int _set_share_node_bitmap(void x, void arg) Link Here

(-)a/src/slurmctld/read_config.c (-15 / +49 lines)
442	return 0;	442	return 0;
443	}	443	}
444		444
		445	static void _set_node_addrs(void arg)
		446	{
		447	list_t *nodes = arg;
		448	slurm_addr_t slurm_addr;
		449	node_record_t *node_ptr;
		450
		451	while ((node_ptr = list_pop(nodes))) {
		452	slurm_set_addr(&slurm_addr, node_ptr->port,
		453	node_ptr->comm_name);
		454	if (slurm_get_port(&slurm_addr))
		455	continue;
		456	error("%s: failure on %s", __func__, node_ptr->comm_name);
		457	node_ptr->node_state = NODE_STATE_FUTURE;
		458	node_ptr->port = 0;
		459	xfree(node_ptr->reason);
		460	node_ptr->reason = xstrdup("NO NETWORK ADDRESS FOUND");
		461	node_ptr->reason_time = time(NULL);
		462	node_ptr->reason_uid = slurm_conf.slurm_user_id;
		463	}
		464
		465	return NULL;
		466	}
		467
445	/*	468	/*
446	* Validate that nodes are addressable.	469	* Validate that nodes are addressable.
447	*/	470	*/
Lines 449-460 static void _validate_slurmd_addr(void) Link Here
449	{	472	{
450	#ifndef HAVE_FRONT_END	473	#ifndef HAVE_FRONT_END
451	node_record_t *node_ptr;	474	node_record_t *node_ptr;
452	slurm_addr_t slurm_addr;
453	DEF_TIMERS;	475	DEF_TIMERS;
454		476	pthread_t *work_threads;
		477	int threads_num = 1;
		478	char *temp_str;
		479	list_t *nodes = list_create(NULL);
455	xassert(verify_lock(CONF_LOCK, READ_LOCK));	480	xassert(verify_lock(CONF_LOCK, READ_LOCK));
456		481
457	START_TIMER;	482	START_TIMER;
		483
		484	if ((temp_str = xstrcasestr(slurm_conf.slurmctld_params,
		485	"validate_nodeaddr_threads="))) {
		486	int tmp_val = strtol(temp_str + 26, NULL, 10);
		487	if ((tmp_val >= 1) && (tmp_val <= 64))
		488	threads_num = tmp_val;
		489	else
		490	error("SlurmctldParameters option validate_nodeaddr_threads=%d out of range, ignored",
		491	tmp_val);
		492	}
		493
		494
458	for (int i = 0; (node_ptr = next_node(&i)); i++) {	495	for (int i = 0; (node_ptr = next_node(&i)); i++) {
459	if ((node_ptr->name == NULL) \|\|	496	if ((node_ptr->name == NULL) \|\|
460	(node_ptr->name[0] == '\0'))	497	(node_ptr->name[0] == '\0'))
Lines 467-485 static void _validate_slurmd_addr(void) Link Here
467	continue;	504	continue;
468	if (node_ptr->port == 0)	505	if (node_ptr->port == 0)
469	node_ptr->port = slurm_conf.slurmd_port;	506	node_ptr->port = slurm_conf.slurmd_port;
470	slurm_set_addr(&slurm_addr, node_ptr->port,	507	list_append(nodes, node_ptr);
471	node_ptr->comm_name);
472	if (slurm_get_port(&slurm_addr))
473	continue;
474	error("%s: failure on %s", __func__, node_ptr->comm_name);
475	node_ptr->node_state = NODE_STATE_FUTURE;
476	node_ptr->port = 0;
477	xfree(node_ptr->reason);
478	node_ptr->reason = xstrdup("NO NETWORK ADDRESS FOUND");
479	node_ptr->reason_time = time(NULL);
480	node_ptr->reason_uid = slurm_conf.slurm_user_id;
481	}	508	}
482		509
		510	work_threads = xcalloc(threads_num, sizeof(pthread_t));
		511	for (int i = 0; i < threads_num; i++)
		512	slurm_thread_create(&work_threads[i], _set_node_addrs, nodes);
		513	for (int i = 0; i < threads_num; i++)
		514	pthread_join(work_threads[i], NULL);
		515	xfree(work_threads);
		516	xassert(list_is_empty(nodes));
		517	FREE_NULL_LIST(nodes);
		518
483	END_TIMER2("_validate_slurmd_addr");	519	END_TIMER2("_validate_slurmd_addr");
484	#endif	520	#endif
485	}	521	}
486	-
487	--
488	src/slurmctld/read_config.c \| 4 ++--	522	src/slurmctld/read_config.c \| 4 ++--
489	1 file changed, 2 insertions(+), 2 deletions(-)	523	1 file changed, 2 insertions(+), 2 deletions(-)