From 72f7c1fda7e298b9dd160f4aea65aef950dc5a64 Mon Sep 17 00:00:00 2001
From: Danny Auble <da@schedmd.com>
Date: Tue, 10 May 2016 14:56:29 -0700
Subject: [PATCH] If running cached information and the database loses all TRES
information make sure we handle it correctly when the database comes back up.
---
src/slurmctld/controller.c | 12 ++++++++++++
1 file changed, 12 insertions(+)
diff --git a/src/slurmctld/controller.c b/src/slurmctld/controller.c
index 73b9da4bb94..211ddaaa2b4 100644
--- a/src/slurmctld/controller.c
+++ b/src/slurmctld/controller.c
@@ -2676,6 +2676,18 @@ static void *_assoc_cache_mgr(void *no_data)
assoc_mgr_refresh_lists(acct_db_conn, 0);
if (running_cache)
unlock_slurmctld(job_write_lock);
+ else if (g_tres_count != slurmctld_tres_cnt) {
+ /* This has to be done outside of the job write lock.
+ * This should only happen in very rare situations
+ * where we have state, but the database some how has
+ * changed out from under us. */
+ unlock_slurmctld(job_write_lock);
+ info("TRES in database does not match cache "
+ "(%u != %u). Updating...",
+ g_tres_count, slurmctld_tres_cnt);
+ _init_tres();
+ lock_slurmctld(job_write_lock);
+ }
slurm_mutex_unlock(&assoc_cache_mutex);
}
--
GitLab