From eaa33e31731a695260c7fd16d4aa2237c5d5d4dc Mon Sep 17 00:00:00 2001
From: Mark Grondona <mgrondona@llnl.gov>
Date: Wed, 11 Jun 2003 17:55:44 +0000
Subject: [PATCH] o Need to wait kill_wait + 2 seconds for slurmd reply to
kill job request instead of just kill_wait seconds because slurmd sleeps
for kill_wait seconds, so therefore slurmctld would never recv a reply.
---
src/slurmctld/agent.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/src/slurmctld/agent.c b/src/slurmctld/agent.c
index 23b2c02a111..d3f44f4c9a9 100644
--- a/src/slurmctld/agent.c
+++ b/src/slurmctld/agent.c
@@ -496,7 +496,8 @@ static void *_thread_per_node_rpc(void *args)
msg.data = task_ptr->msg_args_ptr;
if (task_ptr->msg_type == REQUEST_KILL_TIMELIMIT)
- timeout = slurmctld_conf.kill_wait;
+ timeout = slurmctld_conf.kill_wait + 2; /* 2 extra seconds
+ for slurmd reply */
if (slurm_send_recv_rc_msg(&msg, &rc, timeout) < 0) {
error("agent: %s: %m", thread_ptr->node_name);
--
GitLab