From 364fedac77dfe8798a902e57f574b7adc5f19026 Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Fri, 1 Jun 2007 18:08:21 +0000
Subject: [PATCH] svn merge -r11598:11609
https://eris.llnl.gov/svn/slurm/branches/slurm-1.1
---
NEWS | 2 ++
src/plugins/sched/wiki2/msg.c | 53 ++++++++++++++++++-----------------
2 files changed, 29 insertions(+), 26 deletions(-)
diff --git a/NEWS b/NEWS
index 9d3c5985aed..b5c61d3504e 100644
--- a/NEWS
+++ b/NEWS
@@ -5,6 +5,7 @@ documents those changes that are of interest to users and admins.
========================
-- Add new sinfo field to sort by "%E" sorts by the time associated with a
node's state (from Prashanth Tamraparni, HP).
+ -- In sched/wiki: fix logic for restarting backup slurmctld.
* Changes in SLURM 1.2.8
========================
@@ -318,6 +319,7 @@ documents those changes that are of interest to users and admins.
issue in srun to fail job has failed instead of waiting around for threads
that will never end.
- Added fork handlers in the slurmstepd
+ - In sched/wiki2: fix logic for restarting backup slurmctld.
* Changes in SLURM 1.1.36
=========================
diff --git a/src/plugins/sched/wiki2/msg.c b/src/plugins/sched/wiki2/msg.c
index 843634da366..b07ab5c22eb 100644
--- a/src/plugins/sched/wiki2/msg.c
+++ b/src/plugins/sched/wiki2/msg.c
@@ -105,28 +105,28 @@ extern void term_msg_thread(void)
pthread_mutex_lock(&thread_flag_mutex);
if (thread_running) {
int fd;
- slurm_addr addr;
+ slurm_addr addr;
thread_shutdown = true;
- /* Open and close a connection to the wiki listening port.
- * Allows slurm_accept_msg_conn() to return in
- * _msg_thread() so that it can check the thread_shutdown
- * flag.
- */
- slurm_set_addr(&addr, sched_port, "localhost");
- fd = slurm_open_stream(&addr);
- if (fd != -1) {
- /* we don't care if the open failed */
- slurm_close_stream(fd);
- }
-
- debug2("waiting for sched/wiki2 thread to exit");
- pthread_join(msg_thread_id, NULL);
- msg_thread_id = 0;
- thread_shutdown = false;
- thread_running = false;
- debug2("join of sched/wiki2 thread was successful");
+ /* Open and close a connection to the wiki listening port.
+ * Allows slurm_accept_msg_conn() to return in
+ * _msg_thread() so that it can check the thread_shutdown
+ * flag.
+ */
+ slurm_set_addr(&addr, sched_port, "localhost");
+ fd = slurm_open_stream(&addr);
+ if (fd != -1) {
+ /* we don't care if the open failed */
+ slurm_close_stream(fd);
+ }
+
+ debug2("waiting for sched/wiki thread to exit");
+ pthread_join(msg_thread_id, NULL);
+ msg_thread_id = 0;
+ thread_shutdown = false;
+ thread_running = false;
+ debug2("join of sched/wiki thread was successful");
}
pthread_mutex_unlock(&thread_flag_mutex);
}
@@ -136,7 +136,7 @@ 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();
@@ -148,7 +148,7 @@ static void *_msg_thread(void *no_data)
/* 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; ; i++) {
+ for (i=0; (!thread_shutdown); i++) {
if (i > 0)
sleep(60);
sock_fd = slurm_init_msg_engine_port(sched_port);
@@ -167,10 +167,10 @@ static void *_msg_thread(void *no_data)
error("wiki: slurm_accept_msg_conn %m");
continue;
}
- if (thread_shutdown) {
- close(new_fd);
- break;
- }
+ if (thread_shutdown) {
+ close(new_fd);
+ break;
+ }
/* It would be nice to create a pthread for each new
* RPC, but that leaks memory on some systems when
* done from a plugin.
@@ -182,7 +182,8 @@ 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