From d61a5159e884b448eab652adfc1eccd14ac4fca2 Mon Sep 17 00:00:00 2001
From: Mark Grondona <mgrondona@llnl.gov>
Date: Thu, 10 Jul 2003 21:38:12 +0000
Subject: [PATCH] o Handle forwarding of SIGUSR1,2 o Block SIGALRM by default
o Only allow one thread to enter fwd_signal() at a time.
---
src/srun/allocate.c | 19 +++++--------------
src/srun/launch.c | 6 ++++++
src/srun/signals.c | 11 ++++++++++-
src/srun/srun.c | 8 +++++---
4 files changed, 26 insertions(+), 18 deletions(-)
diff --git a/src/srun/allocate.c b/src/srun/allocate.c
index 4079d200b1d..81081aac615 100644
--- a/src/srun/allocate.c
+++ b/src/srun/allocate.c
@@ -68,22 +68,15 @@ allocate_nodes(void)
resource_allocation_response_msg_t *resp = NULL;
job_desc_msg_t *j = job_desc_msg_create();
- /*
- * Save old signal mask for this thread
- */
- if ((rc = pthread_sigmask(SIG_BLOCK, NULL, &oset)) != 0) {
- error("pthread_sigmask: %s", slurm_strerror(rc));
- return NULL;
- }
+ oquitf = xsignal(SIGQUIT, _intr_handler);
+ ointf = xsignal(SIGINT, _intr_handler);
+ otermf = xsignal(SIGTERM, _intr_handler);
+ xsignal_save_mask(&oset);
xsignal_unblock(SIGQUIT);
xsignal_unblock(SIGINT);
xsignal_unblock(SIGTERM);
- oquitf = xsignal(SIGQUIT, _intr_handler);
- ointf = xsignal(SIGINT, _intr_handler);
- otermf = xsignal(SIGTERM, _intr_handler);
-
while ((rc = slurm_allocate_resources(j, &resp) < 0) && _retry()) {
if (destroy_job)
goto done;
@@ -96,9 +89,7 @@ allocate_nodes(void)
}
done:
- if ((rc = pthread_sigmask(SIG_BLOCK, &oset, NULL)) != 0)
- error("Unable to restore signal mask: %s", slurm_strerror(rc));
-
+ xsignal_restore_mask(&oset);
xsignal(SIGINT, ointf);
xsignal(SIGTERM, otermf);
xsignal(SIGQUIT, oquitf);
diff --git a/src/srun/launch.c b/src/srun/launch.c
index baa035d313e..1bb6ca21687 100644
--- a/src/srun/launch.c
+++ b/src/srun/launch.c
@@ -240,6 +240,7 @@ static int _check_pending_threads(thd_t *thd, int count)
for (i = 0; i < count; i++) {
if ((thd[i].state == DSH_ACTIVE)
&& ((now - thd[i].tstart) >= 2) )
+ verbose("sending SIGALRM to thread %d", thd[i].thread);
pthread_kill(thd[i].thread, SIGALRM);
}
@@ -305,8 +306,12 @@ static void _p_launch(slurm_msg_t *req, job_t *job)
thd_t *thd;
int rc = 0;
SigFunc *oldh;
+ sigset_t set;
oldh = xsignal(SIGALRM, (SigFunc *) _alrm_handler);
+ xsignal_save_mask(&set);
+ xsignal_unblock(SIGALRM);
+
/*
* Set job timeout to maximum launch time + current time
*/
@@ -341,6 +346,7 @@ static void _p_launch(slurm_msg_t *req, job_t *job)
_wait_on_active(thd, job);
pthread_mutex_unlock(&active_mutex);
+ xsignal_restore_mask(&set);
xsignal(SIGALRM, oldh);
xfree(thd);
diff --git a/src/srun/signals.c b/src/srun/signals.c
index 7dfc0951e78..cd2aca04849 100644
--- a/src/srun/signals.c
+++ b/src/srun/signals.c
@@ -102,6 +102,9 @@ sig_setup_sigmask(void)
sigaddset(&sigset, SIGTSTP);
sigaddset(&sigset, SIGSTOP);
sigaddset(&sigset, SIGCONT);
+ sigaddset(&sigset, SIGALRM);
+ sigaddset(&sigset, SIGUSR1);
+ sigaddset(&sigset, SIGUSR2);
if ((err = pthread_sigmask(SIG_BLOCK, &sigset, NULL)) != 0) {
error("pthread_sigmask: %s", slurm_strerror(err));
@@ -137,6 +140,9 @@ fwd_signal(job_t *job, int signo)
int i;
slurm_msg_t *req;
kill_tasks_msg_t msg;
+ static pthread_mutex_t sig_mutex = PTHREAD_MUTEX_INITIALIZER;
+
+ slurm_mutex_lock(&sig_mutex);
if (signo == SIGKILL || signo == SIGINT || signo == SIGTERM) {
slurm_mutex_lock(&job->state_mutex);
@@ -172,6 +178,7 @@ fwd_signal(job_t *job, int signo)
debug2("All tasks have been signalled");
xfree(req);
+ slurm_mutex_unlock(&sig_mutex);
}
@@ -219,7 +226,8 @@ _sig_thr_setup(sigset_t *set)
sigaddset(set, SIGQUIT);
sigaddset(set, SIGTSTP);
sigaddset(set, SIGSTOP);
- sigaddset(set, SIGCONT);
+ sigaddset(set, SIGUSR1);
+ sigaddset(set, SIGUSR2);
}
/* simple signal handling thread */
@@ -256,6 +264,7 @@ _sig_thr(void *arg)
job_force_termination(job);
break;
default:
+ fwd_signal(job, signo);
break;
}
}
diff --git a/src/srun/srun.c b/src/srun/srun.c
index 4036c0572cd..32c8f3b9a61 100644
--- a/src/srun/srun.c
+++ b/src/srun/srun.c
@@ -226,15 +226,17 @@ srun(int ac, char **av)
/* Tell slurmctld that job is done */
job_destroy(job, 0);
+ unblock_all_signals();
+
/* kill msg server thread */
- /*pthread_kill(job->jtid, SIGHUP);*/
+ pthread_kill(job->jtid, SIGTERM);
/* kill signal thread */
- /*pthread_kill(job->sigid, SIGHUP);*/
+ pthread_kill(job->sigid, SIGKILL);
log_fini();
- return job_rc(job);
+ exit(job_rc(job));
}
--
GitLab