From 97aedaf890f603275b560180f4e3c60f2f787c85 Mon Sep 17 00:00:00 2001
From: Morris Jette <jette@schedmd.com>
Date: Tue, 28 Apr 2015 08:07:38 -0700
Subject: [PATCH] minor updates to sched_min_interval logic
Minor revisiions to the logic and documentation of commit
266246025044bdf4ed7c470f320b6c610f24d70e
---
doc/man/man5/slurm.conf.5 | 4 +++-
src/slurmctld/job_scheduler.c | 28 ++++++++++++++++------------
2 files changed, 19 insertions(+), 13 deletions(-)
diff --git a/doc/man/man5/slurm.conf.5 b/doc/man/man5/slurm.conf.5
index 2db1dfca395..8021179996d 100644
--- a/doc/man/man5/slurm.conf.5
+++ b/doc/man/man5/slurm.conf.5
@@ -2278,8 +2278,10 @@ The scheduler runs in a limited fashion every time that any event happens which
could enable a job to start (e.g. job submit, job terminate, etc.).
If these events happen at a high frequency, the scheduler can run very
frequently and consume significant resources if not throttled by this option.
+This option specifies the minimum time between the start of a scheduling
+cycle, without considering how long the scheduling logic runs.
A value of zero will disable throttling of the scheduling logic interval.
-The default value is 10,000 microseconds.
+The default value is zero microseconds (throttling is disabled).
.RE
.TP
diff --git a/src/slurmctld/job_scheduler.c b/src/slurmctld/job_scheduler.c
index b816d6b4fcf..c33e163e533 100644
--- a/src/slurmctld/job_scheduler.c
+++ b/src/slurmctld/job_scheduler.c
@@ -770,7 +770,7 @@ static bool _all_partition_priorities_same(void)
*/
extern int schedule(uint32_t job_limit)
{
- static int sched_job_limit = 1;
+ static int sched_job_limit = -1;
int job_count = 0;
struct timeval now;
long delta_t;
@@ -781,20 +781,23 @@ extern int schedule(uint32_t job_limit)
} else {
delta_t = (now.tv_sec - sched_last.tv_sec) * 1000000;
delta_t += now.tv_usec - sched_last.tv_usec;
-
}
slurm_mutex_lock(&sched_mutex);
- if (job_limit == 0)
- sched_job_limit = 0; /* unlimited */
- else if (job_limit > sched_job_limit)
- sched_job_limit = job_limit;
+ if (sched_job_limit == 0)
+ ; /* leave unlimited */
+ else if (job_limit == 0)
+ sched_job_limit = 0; /* set unlimited */
+ else if (sched_job_limit == -1)
+ sched_job_limit = job_limit; /* set initial value */
+ else
+ sched_job_limit += job_limit; /* test more jobs */
if (delta_t >= sched_min_interval) {
sched_last.tv_sec = now.tv_sec;
sched_last.tv_usec = now.tv_usec;
job_limit = sched_job_limit;
- sched_job_limit = 1;
+ sched_job_limit = -1;
slurm_mutex_unlock(&sched_mutex);
job_count = _schedule(job_limit);
@@ -813,7 +816,7 @@ extern int schedule(uint32_t job_limit)
error("pthread_create error %m");
} else
sched_pend_thread = 1;
- slurm_attr_destroy(&attr_agent);
+ slurm_attr_destroy(&attr_agent);
slurm_mutex_unlock(&sched_mutex);
} else {
/* Nothing to do, agent already pending */
@@ -851,8 +854,8 @@ static void *_sched_agent(void *args)
slurm_mutex_unlock(&sched_mutex);
if (job_cnt) {
/* jobs were started, save state */
- schedule_node_save(); /* Has own locking */
- schedule_job_save(); /* Has own locking */
+ schedule_node_save(); /* Has own locking */
+ schedule_job_save(); /* Has own locking */
}
return NULL;
@@ -1051,9 +1054,10 @@ static int _schedule(uint32_t job_limit)
sched_update = slurmctld_conf.last_update;
info("SchedulerParameters=default_queue_depth=%d,"
"max_rpc_cnt=%d,max_sched_time=%d,partition_job_depth=%d,"
- "sched_max_job_start=%d",
+ "sched_max_job_start=%d,sched_min_interval=%d",
def_job_limit, defer_rpc_cnt, sched_timeout,
- max_jobs_per_part, sched_max_job_start);
+ max_jobs_per_part, sched_max_job_start,
+ sched_min_interval);
}
if ((defer_rpc_cnt > 0) &&
--
GitLab