From 7734d9c0a057420c73e5ee8097c3c2b04ee9e0ea Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Tue, 11 Aug 2009 20:54:30 +0000
Subject: [PATCH] avoid resuming a job in gang that was not suspended by gang
(e.g. admin action). add some more logging
---
src/slurmctld/gang.c | 14 ++++++++++----
1 file changed, 10 insertions(+), 4 deletions(-)
diff --git a/src/slurmctld/gang.c b/src/slurmctld/gang.c
index 58c3e411cb4..ea4d2ad8b39 100644
--- a/src/slurmctld/gang.c
+++ b/src/slurmctld/gang.c
@@ -1430,6 +1430,8 @@ extern int gs_reconfig(void)
for (i = 0; i < p_ptr->num_jobs; i++) {
if (p_ptr->job_list[i]->sig_state ==
GS_SUSPEND) {
+ info("resuming job in missing part %s",
+ p_ptr->part_name);
_resume_job(p_ptr->job_list[i]->
job_id);
p_ptr->job_list[i]->sig_state =
@@ -1447,8 +1449,8 @@ extern int gs_reconfig(void)
* necessary). NOTE: there could be jobs that only overlap
* on nodes that are no longer in the partition, but we're
* not going to worry about those cases.
- */
- /* add the jobs from p_ptr into new_ptr in their current order
+ *
+ * add the jobs from p_ptr into new_ptr in their current order
* to preserve the state of timeslicing.
*/
for (i = 0; i < p_ptr->num_jobs; i++) {
@@ -1457,9 +1459,13 @@ extern int gs_reconfig(void)
/* job no longer exists in SLURM, so drop it */
continue;
}
- /* resume any job that is suspended */
- if (IS_JOB_SUSPENDED(job_ptr))
+ /* resume any job that is suspended by us */
+ if (IS_JOB_SUSPENDED(job_ptr) &&
+ (job_ptr->priority != 0)) {
+ debug3("resuming job %u apparently suspended "
+ " by gang", job_ptr->job_id);
_resume_job(job_ptr->job_id);
+ }
/* transfer the job as long as it is still active */
if (IS_JOB_SUSPENDED(job_ptr) ||
--
GitLab