From cfc6ca6cb375a4ed47eda430f84608f4662ab4be Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Fri, 1 Jun 2007 17:52:14 +0000
Subject: [PATCH] Add socket re-open logic to retry
---
src/plugins/sched/wiki/msg.c | 21 ++++++++++++++++-----
1 file changed, 16 insertions(+), 5 deletions(-)
diff --git a/src/plugins/sched/wiki/msg.c b/src/plugins/sched/wiki/msg.c
index 7c92b0d28d6..01e0c5fb43a 100644
--- a/src/plugins/sched/wiki/msg.c
+++ b/src/plugins/sched/wiki/msg.c
@@ -134,17 +134,27 @@ extern void term_msg_thread(void)
\*****************************************************************************/
static void *_msg_thread(void *no_data)
{
- slurm_fd sock_fd, new_fd;
+ slurm_fd sock_fd = -1, new_fd;
slurm_addr cli_addr;
char *msg;
slurm_ctl_conf_t *conf = slurm_conf_lock();
+ int i;
sched_port = conf->schedport;
slurm_conf_unlock();
- if ((sock_fd = slurm_init_msg_engine_port(sched_port))
- == SLURM_SOCKET_ERROR) {
- fatal("wiki: slurm_init_msg_engine_port %u %m",
+
+ /* If SchedulerPort is already taken, keep trying to open it
+ * once per minute. Slurmctld will continue to function
+ * during this interval even if nothing can be scheduled. */
+ for (i=0; (!thread_shutdown); i++) {
+ if (i > 0)
+ sleep(60);
+ sock_fd = slurm_init_msg_engine_port(sched_port);
+ if (sock_fd != SLURM_SOCKET_ERROR)
+ break;
+ error("wiki: slurm_init_msg_engine_port %u %m",
sched_port);
+ error("wiki: Unable to communicate with Moab");
}
/* Process incoming RPCs until told to shutdown */
@@ -170,7 +180,8 @@ static void *_msg_thread(void *no_data)
xfree(msg);
slurm_close_accepted_conn(new_fd);
}
- (void) slurm_shutdown_msg_engine(sock_fd);
+ if (sock_fd > 0)
+ (void) slurm_shutdown_msg_engine(sock_fd);
pthread_exit((void *) 0);
return NULL;
}
--
GitLab