Attachment #430 for ticket #445

Lines 38-43 documents those changes that are of interest to users and admins. Link Here

(-)a/NEWS (+3 lines)
38	can only be a TORUS (1).	38	can only be a TORUS (1).
39	-- Change max message length from 100MB to 1GB before generating "Insane	39	-- Change max message length from 100MB to 1GB before generating "Insane
40	message length" error.	40	message length" error.
		41	-- sched/backfill - Prevent possible memory corruption due to use of
		42	bf_continue option and long running scheduling cycle (pending jobs could
		43	have been cancelled and purged).
41		44
42	* Changes in Slurm 2.6.2	45	* Changes in Slurm 2.6.2
43	========================	46	========================




	}
	while ((job_queue_rec = (job_queue_rec_t *)
				list_pop_bottom(job_queue, sort_job_queue2))) {
		job_ptr  = job_queue_rec->job_ptr;
		orig_time_limit = job_ptr->time_limit;

		if ((time(NULL) - sched_start) >= sched_timeout) {
			uint32_t save_time_limit = job_ptr->time_limit;
			job_ptr->time_limit = orig_time_limit;
			if (debug_flags & DEBUG_FLAG_BACKFILL) {
				END_TIMER;
				info("backfill: completed yielding locks "

				rc = 1;
				break;
			}
			job_ptr->time_limit = save_time_limit;
			/* Reset backfill scheduling timers, resume testing */
			sched_start = time(NULL);
			job_test_count = 0;
			START_TIMER;
		}

		job_ptr  = job_queue_rec->job_ptr;
		/* With bf_continue configured, the original job could have
		 * been cancelled and purged. Validate pointer here. */
		if ((job_ptr->magic  != JOB_MAGIC) ||
		    (job_ptr->job_id != job_queue_rec->job_id))
			continue;
		orig_time_limit = job_ptr->time_limit;
		part_ptr = job_queue_rec->part_ptr;
		job_test_count++;


Lines 140-145 static void _job_queue_append(List job_queue, struct job_record *job_ptr, Link Here

(-)a/src/slurmctld/job_scheduler.c (+1 lines)
140	job_queue_rec_t *job_queue_rec;	140	job_queue_rec_t *job_queue_rec;
141		141
142	job_queue_rec = xmalloc(sizeof(job_queue_rec_t));	142	job_queue_rec = xmalloc(sizeof(job_queue_rec_t));
		143	job_queue_rec->job_id = job_ptr->job_id;
143	job_queue_rec->job_ptr = job_ptr;	144	job_queue_rec->job_ptr = job_ptr;
144	job_queue_rec->part_ptr = part_ptr;	145	job_queue_rec->part_ptr = part_ptr;
145	job_queue_rec->priority = prio;	146	job_queue_rec->priority = prio;

Lines 45-50 Link Here

(-)a/src/slurmctld/job_scheduler.h (-1 / +1 lines)
45	#include "src/slurmctld/slurmctld.h"	45	#include "src/slurmctld/slurmctld.h"
46		46
47	typedef struct job_queue_rec {	47	typedef struct job_queue_rec {
		48	uint32_t job_id;
48	struct job_record *job_ptr;	49	struct job_record *job_ptr;
49	struct part_record *part_ptr;	50	struct part_record *part_ptr;
50	uint32_t priority;	51	uint32_t priority;
51	-

Return to ticket 445

Lines 639-650 static int _attempt_backfill(void) Link Here

(-)a/src/plugins/sched/backfill/backfill.c (-6 / +7 lines)
639	}	639	}
640	while ((job_queue_rec = (job_queue_rec_t *)	640	while ((job_queue_rec = (job_queue_rec_t *)
641	list_pop_bottom(job_queue, sort_job_queue2))) {	641	list_pop_bottom(job_queue, sort_job_queue2))) {
642	job_ptr = job_queue_rec->job_ptr;
643	orig_time_limit = job_ptr->time_limit;
644
645	if ((time(NULL) - sched_start) >= sched_timeout) {	642	if ((time(NULL) - sched_start) >= sched_timeout) {
646	uint32_t save_time_limit = job_ptr->time_limit;
647	job_ptr->time_limit = orig_time_limit;
648	if (debug_flags & DEBUG_FLAG_BACKFILL) {	643	if (debug_flags & DEBUG_FLAG_BACKFILL) {
649	END_TIMER;	644	END_TIMER;
650	info("backfill: completed yielding locks "	645	info("backfill: completed yielding locks "
Lines 660-672 static int _attempt_backfill(void) Link Here
660	rc = 1;	655	rc = 1;
661	break;	656	break;
662	}	657	}
663	job_ptr->time_limit = save_time_limit;
664	/* Reset backfill scheduling timers, resume testing */	658	/* Reset backfill scheduling timers, resume testing */
665	sched_start = time(NULL);	659	sched_start = time(NULL);
666	job_test_count = 0;	660	job_test_count = 0;
667	START_TIMER;	661	START_TIMER;
668	}	662	}
669		663
		664	job_ptr = job_queue_rec->job_ptr;
		665	/* With bf_continue configured, the original job could have
		666	* been cancelled and purged. Validate pointer here. */
		667	if ((job_ptr->magic != JOB_MAGIC) \|\|
		668	(job_ptr->job_id != job_queue_rec->job_id))
		669	continue;
		670	orig_time_limit = job_ptr->time_limit;
670	part_ptr = job_queue_rec->part_ptr;	671	part_ptr = job_queue_rec->part_ptr;
671	job_test_count++;	672	job_test_count++;
672		673