From fe8b04fb2fad3f0311a6acd264aa41d5e6b612a5 Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Thu, 18 Jan 2007 17:40:18 +0000
Subject: [PATCH] Make frequency with which we purge jobs (and resend terminate
job RPC) a parameter, effectively raise the time from 1 to 5 mins.
---
src/slurmctld/controller.c | 12 +++++++++---
src/slurmctld/slurmctld.h | 4 ++++
2 files changed, 13 insertions(+), 3 deletions(-)
diff --git a/src/slurmctld/controller.c b/src/slurmctld/controller.c
index 4fd41c1552a..885b0a58eb3 100644
--- a/src/slurmctld/controller.c
+++ b/src/slurmctld/controller.c
@@ -729,6 +729,7 @@ static void *_slurmctld_background(void *no_data)
static time_t last_group_time;
static time_t last_ping_node_time;
static time_t last_ping_srun_time;
+ static time_t last_purge_job_time;
static time_t last_timelimit_time;
static time_t last_assert_primary_time;
time_t now;
@@ -755,6 +756,7 @@ static void *_slurmctld_background(void *no_data)
/* Let the dust settle before doing work */
now = time(NULL);
last_sched_time = last_checkpoint_time = last_group_time = now;
+ last_purge_job_time = now;
last_timelimit_time = last_assert_primary_time = now;
if (slurmctld_conf.slurmd_timeout) {
/* We ping nodes that haven't responded in SlurmdTimeout/2,
@@ -830,12 +832,16 @@ static void *_slurmctld_background(void *no_data)
unlock_slurmctld(part_write_lock);
}
- if (difftime(now, last_sched_time) >= PERIODIC_SCHEDULE) {
- last_sched_time = now;
+ if (difftime(now, last_purge_job_time) >= PURGE_JOB_INTERVAL) {
+ last_purge_job_time = now;
debug2("Performing purge of old job records");
lock_slurmctld(job_write_lock);
- purge_old_job(); /* remove defunct job recs */
+ purge_old_job();
unlock_slurmctld(job_write_lock);
+ }
+
+ if (difftime(now, last_sched_time) >= PERIODIC_SCHEDULE) {
+ last_sched_time = now;
if (schedule())
last_checkpoint_time = 0; /* force state save */
}
diff --git a/src/slurmctld/slurmctld.h b/src/slurmctld/slurmctld.h
index f38f32addb5..d0ae1c968c6 100644
--- a/src/slurmctld/slurmctld.h
+++ b/src/slurmctld/slurmctld.h
@@ -108,6 +108,10 @@
/* Check for jobs reaching their time limit every PERIODIC_TIMEOUT seconds */
#define PERIODIC_TIMEOUT 60
+/* Attempt to purge defunct job records and resend job kill requests
+ * every PURGE_JOB_INTERVAL seconds */
+#define PURGE_JOB_INTERVAL 300
+
/* Pathname of group file record for checking update times */
#define GROUP_FILE "/etc/group"
--
GitLab