From 3ec7382ae14f29ce9d1a8fcf1b9faa6d4ad9f50c Mon Sep 17 00:00:00 2001
From: Alejandro Sanchez <alex@schedmd.com>
Date: Mon, 27 Nov 2017 11:33:16 -0700
Subject: [PATCH] Fix hetjob component incorrect Priority state reason at
allocation time.
_top_priority() is used to determine at job allocation time (called from
job_allocate()) if any other job has a higher priority than the specified
job being allocated. If so, it sets the job state_reason to WAIT_PRIORITY.
This patch skips the priority comparison if the two jobs being compared
are pack job components from the same heterogeneous job.
Bug 4400.
---
src/slurmctld/job_mgr.c | 9 ++++++---
1 file changed, 6 insertions(+), 3 deletions(-)
diff --git a/src/slurmctld/job_mgr.c b/src/slurmctld/job_mgr.c
index 61bb085f9d5..e1ee9871f8d 100644
--- a/src/slurmctld/job_mgr.c
+++ b/src/slurmctld/job_mgr.c
@@ -245,7 +245,7 @@ static void _signal_job(struct job_record *job_ptr, int signal, uint16_t flags);
static void _suspend_job(struct job_record *job_ptr, uint16_t op,
bool indf_susp);
static int _suspend_job_nodes(struct job_record *job_ptr, bool indf_susp);
-static bool _top_priority(struct job_record *job_ptr);
+static bool _top_priority(struct job_record *job_ptr, uint32_t pack_job_offset);
static int _valid_job_part(job_desc_msg_t * job_desc,
uid_t submit_uid, bitstr_t *req_bitmap,
struct part_record **part_pptr,
@@ -4794,7 +4794,7 @@ extern int job_allocate(job_desc_msg_t * job_specs, int immediate,
too_fragmented = true;
if (independent && (!too_fragmented))
- top_prio = _top_priority(job_ptr);
+ top_prio = _top_priority(job_ptr, job_specs->pack_job_offset);
else
top_prio = true; /* don't bother testing,
* it is not runable anyway */
@@ -11045,7 +11045,7 @@ extern void sync_job_priorities(void)
* IN job_ptr - pointer to selected job
* RET true if selected job has highest priority
*/
-static bool _top_priority(struct job_record *job_ptr)
+static bool _top_priority(struct job_record *job_ptr, uint32_t pack_job_offset)
{
struct job_details *detail_ptr = job_ptr->details;
time_t now = time(NULL);
@@ -11083,6 +11083,9 @@ static bool _top_priority(struct job_record *job_ptr)
list_next(job_iterator))) {
if (job_ptr2 == job_ptr)
continue;
+ if ((pack_job_offset != NO_VAL) && (job_ptr->job_id ==
+ (job_ptr2->job_id + pack_job_offset)))
+ continue;
if (!IS_JOB_PENDING(job_ptr2))
continue;
if (IS_JOB_COMPLETING(job_ptr2)) {
--
GitLab