From 231af0d51c0c30cc70c1d03a12a0754f2641f64c Mon Sep 17 00:00:00 2001
From: Danny Auble <da@schedmd.com>
Date: Mon, 12 Jun 2017 16:33:30 -0600
Subject: [PATCH] Make the fini array in slurmd based on the number of cpus on
the node instead of a hard coded value.
This probably needs to be something higher to accommodate potential systems
with with oversubscribe given.
See bug 3833
---
src/slurmd/slurmd/req.c | 19 ++++++-------------
src/slurmd/slurmd/slurmd.c | 10 +++++++++-
src/slurmd/slurmd/slurmd.h | 3 +++
3 files changed, 18 insertions(+), 14 deletions(-)
diff --git a/src/slurmd/slurmd/req.c b/src/slurmd/slurmd/req.c
index e9959e1458a..f61fda3fd28 100644
--- a/src/slurmd/slurmd/req.c
+++ b/src/slurmd/slurmd/req.c
@@ -253,13 +253,6 @@ static pthread_mutex_t job_limits_mutex = PTHREAD_MUTEX_INITIALIZER;
static List job_limits_list = NULL;
static bool job_limits_loaded = false;
-/*
- * To be fixed in 17.11 to match the count of cpus on a node instead of a hard
- * code.
- */
-#define FINI_JOB_CNT 256
-static pthread_mutex_t fini_mutex = PTHREAD_MUTEX_INITIALIZER;
-static uint32_t fini_job_id[FINI_JOB_CNT];
static int next_fini_job_inx = 0;
/* NUM_PARALLEL_SUSP_JOBS controls the number of jobs that can be suspended or
@@ -1835,24 +1828,24 @@ static bool _is_batch_job_finished(uint32_t job_id)
bool found_job = false;
int i;
- slurm_mutex_lock(&fini_mutex);
- for (i = 0; i < FINI_JOB_CNT; i++) {
+ slurm_mutex_lock(&fini_job_mutex);
+ for (i = 0; i < fini_job_cnt; i++) {
if (fini_job_id[i] == job_id) {
found_job = true;
break;
}
}
- slurm_mutex_unlock(&fini_mutex);
+ slurm_mutex_unlock(&fini_job_mutex);
return found_job;
}
static void _note_batch_job_finished(uint32_t job_id)
{
- slurm_mutex_lock(&fini_mutex);
+ slurm_mutex_lock(&fini_job_mutex);
fini_job_id[next_fini_job_inx] = job_id;
- if (++next_fini_job_inx >= FINI_JOB_CNT)
+ if (++next_fini_job_inx >= fini_job_cnt)
next_fini_job_inx = 0;
- slurm_mutex_unlock(&fini_mutex);
+ slurm_mutex_unlock(&fini_job_mutex);
}
/* Send notification to slurmctld we are finished running the prolog.
diff --git a/src/slurmd/slurmd/slurmd.c b/src/slurmd/slurmd/slurmd.c
index eda00ad1428..059b4cc88db 100644
--- a/src/slurmd/slurmd/slurmd.c
+++ b/src/slurmd/slurmd/slurmd.c
@@ -125,6 +125,9 @@
/* global, copied to STDERR_FILENO in tasks before the exec */
int devnull = -1;
slurmd_conf_t * conf = NULL;
+int fini_job_cnt = 0;
+uint32_t *fini_job_id = NULL;
+pthread_mutex_t fini_job_mutex = PTHREAD_MUTEX_INITIALIZER;
/*
* count of active threads
@@ -1506,7 +1509,8 @@ _slurmd_init(void)
*/
_read_config();
- cpu_cnt = MAX(conf->conf_cpus, conf->block_map_size);
+ fini_job_cnt = cpu_cnt = MAX(conf->conf_cpus, conf->block_map_size);
+ fini_job_id = xmalloc(sizeof(uint32_t) * fini_job_cnt);
if ((gres_plugin_init() != SLURM_SUCCESS) ||
(gres_plugin_node_config_load(cpu_cnt, conf->node_name, NULL)
@@ -1727,6 +1731,10 @@ _slurmd_fini(void)
acct_gather_conf_destroy();
fini_system_cgroup();
route_fini();
+ slurm_mutex_lock(&fini_job_mutex);
+ xfree(fini_job_id);
+ fini_job_cnt = 0;
+ slurm_mutex_unlock(&fini_job_mutex);
return SLURM_SUCCESS;
}
diff --git a/src/slurmd/slurmd/slurmd.h b/src/slurmd/slurmd/slurmd.h
index b5e6339cadd..f38a8f9f059 100644
--- a/src/slurmd/slurmd/slurmd.h
+++ b/src/slurmd/slurmd/slurmd.h
@@ -170,6 +170,9 @@ typedef struct slurmd_config {
} slurmd_conf_t;
extern slurmd_conf_t * conf;
+extern int fini_job_cnt;
+extern uint32_t *fini_job_id;
+extern pthread_mutex_t fini_job_mutex;
/* Send node registration message with status to controller
* IN status - same values slurm error codes (for node shutdown)
--
GitLab