From 04fbf26a2254c8505a7111a66867acdef984ca6c Mon Sep 17 00:00:00 2001
From: Morris Jette <jette@schedmd.com>
Date: Wed, 16 Jan 2013 11:53:18 -0800
Subject: [PATCH] Fix for scheduling batch jobs in multiple partitions
Without this change a high priority batch job may not start at submit
time. In addtion, a pending job with mutltiple partitions be cancelled
when the scheduler runs if any of it's partitions can not be used by
the job.
---
src/slurmctld/job_scheduler.c | 6 ++++++
src/slurmctld/proc_req.c | 5 ++++-
2 files changed, 10 insertions(+), 1 deletion(-)
diff --git a/src/slurmctld/job_scheduler.c b/src/slurmctld/job_scheduler.c
index 03fd0d7cc15..cf740c0422a 100644
--- a/src/slurmctld/job_scheduler.c
+++ b/src/slurmctld/job_scheduler.c
@@ -1016,6 +1016,12 @@ next_part: part_ptr = (struct part_record *)
launch_job(job_ptr);
rebuild_job_part_list(job_ptr);
job_cnt++;
+ } else if ((error_code ==
+ ESLURM_REQUESTED_NODE_CONFIG_UNAVAILABLE) &&
+ job_ptr->part_ptr_list) {
+ debug("JobId=%u non-runnable in partition %s: %s",
+ job_ptr->job_id, job_ptr->part_ptr->name,
+ slurm_strerror(error_code));
} else if ((error_code !=
ESLURM_REQUESTED_PART_CONFIG_UNAVAILABLE) &&
(error_code != ESLURM_NODE_NOT_AVAIL) &&
diff --git a/src/slurmctld/proc_req.c b/src/slurmctld/proc_req.c
index 1d3b711177a..410d88299ca 100644
--- a/src/slurmctld/proc_req.c
+++ b/src/slurmctld/proc_req.c
@@ -2695,6 +2695,9 @@ static void _slurm_rpc_submit_batch_job(slurm_msg_t * msg)
slurm_strerror(error_code));
slurm_send_rc_msg(msg, error_code);
} else {
+ int sched_count = 1; /* Job count to attempt to schedule */
+ if (job_ptr->part_ptr_list)
+ sched_count = list_count(job_ptr->part_ptr_list);
info("_slurm_rpc_submit_batch_job JobId=%u %s",
job_ptr->job_id, TIME_STR);
/* send job_ID */
@@ -2709,7 +2712,7 @@ static void _slurm_rpc_submit_batch_job(slurm_msg_t * msg)
* We also run schedule() even if this job could not start,
* say due to a higher priority job, since the locks are
* released above and we might start some other job here. */
- schedule(1); /* has own locks */
+ schedule(sched_count); /* has own locks */
schedule_job_save(); /* has own locks */
schedule_node_save(); /* has own locks */
}
--
GitLab