From a803b44b51570a50bf72fdf0098d162c5e66c6f3 Mon Sep 17 00:00:00 2001
From: Morris Jette <jette@schedmd.com>
Date: Tue, 15 Apr 2014 11:56:15 -0700
Subject: [PATCH] Job array scheduling optimizations
If one task of a job array can not be scheduled in the main
scheduling loop, skip the subsequent tasks of that same
job array.
---
src/slurmctld/job_scheduler.c | 27 ++++++++++++++++++++-------
1 file changed, 20 insertions(+), 7 deletions(-)
diff --git a/src/slurmctld/job_scheduler.c b/src/slurmctld/job_scheduler.c
index 595ddd7705e..4bebe57da04 100644
--- a/src/slurmctld/job_scheduler.c
+++ b/src/slurmctld/job_scheduler.c
@@ -747,6 +747,7 @@ extern int schedule(uint32_t job_limit)
static int max_jobs_per_part = 0;
static int defer_rpc_cnt = 0;
time_t now, sched_start;
+ uint32_t reject_array_job_id = 0;
DEF_TIMERS;
if (sched_update != slurmctld_conf.last_update) {
@@ -869,14 +870,14 @@ extern int schedule(uint32_t job_limit)
unlock_slurmctld(job_write_lock);
debug("sched: schedule() returning, no front end nodes are "
- "available");
+ "available");
return 0;
}
/* Avoid resource fragmentation if important */
if ((!wiki_sched) && job_is_completing()) {
unlock_slurmctld(job_write_lock);
debug("sched: schedule() returning, some job is still "
- "completing");
+ "completing");
return 0;
}
@@ -988,6 +989,12 @@ next_part: part_ptr = (struct part_record *)
break;
}
+ if (job_ptr->array_task_id != NO_VAL) {
+ if (reject_array_job_id == job_ptr->array_job_id)
+ continue; /* already rejected array element */
+ /* assume reject whole array for now, clear if OK */
+ reject_array_job_id = job_ptr->array_job_id;
+ }
if (max_jobs_per_part) {
bool skip_job = false;
for (j = 0; j < part_cnt; j++) {
@@ -998,11 +1005,14 @@ next_part: part_ptr = (struct part_record *)
skip_job = true;
break;
}
- if (skip_job)
+ if (skip_job) {
+ debug("sched: reacked partition %s job limit",
+ job_ptr->part_ptr->name);
continue;
+ }
}
if (job_depth++ > job_limit) {
- debug3("sched: already tested %u jobs, breaking out",
+ debug("sched: already tested %u jobs, breaking out",
job_depth);
break;
}
@@ -1043,7 +1053,7 @@ next_part: part_ptr = (struct part_record *)
xfree(job_ptr->state_desc);
last_job_update = now;
}
- debug3("sched: JobId=%u. State=PENDING. "
+ debug("sched: JobId=%u. State=PENDING. "
"Reason=%s(Priority), Priority=%u, "
"Partition=%s.",
job_ptr->job_id,
@@ -1070,6 +1080,8 @@ next_part: part_ptr = (struct part_record *)
job_ptr->assoc_id = assoc_rec.id;
last_job_update = now;
} else {
+ debug("sched: JobId=%u has invalid association",
+ job_ptr->job_id);
continue;
}
}
@@ -1080,8 +1092,8 @@ next_part: part_ptr = (struct part_record *)
!bit_test(assoc_ptr->usage->valid_qos,
job_ptr->qos_id) &&
!job_ptr->limit_set_qos) {
- info("sched: JobId=%u has invalid QOS",
- job_ptr->job_id);
+ debug("sched: JobId=%u has invalid QOS",
+ job_ptr->job_id);
xfree(job_ptr->state_desc);
job_ptr->state_reason = FAIL_QOS;
last_job_update = now;
@@ -1261,6 +1273,7 @@ next_part: part_ptr = (struct part_record *)
launch_job(job_ptr);
rebuild_job_part_list(job_ptr);
job_cnt++;
+ reject_array_job_id = 0;
} else if ((error_code ==
ESLURM_REQUESTED_NODE_CONFIG_UNAVAILABLE) &&
job_ptr->part_ptr_list) {
--
GitLab