From 7bd1849e46b36ceb61ebd5f79d436930359d7214 Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Thu, 26 Apr 2007 20:26:34 +0000
Subject: [PATCH] Fix bug which prevented srun from reporting to slurmctld the
job's exit code. Applies only to non-batch jobs. and the bad exit code was
only reported by "scontrol show jobs" (the exit code in accounting data was
correct).
---
src/srun/srun.c | 4 ++--
src/srun/srun_job.c | 29 +++++++++++++----------------
src/srun/srun_job.h | 3 ++-
3 files changed, 17 insertions(+), 19 deletions(-)
diff --git a/src/srun/srun.c b/src/srun/srun.c
index e2239510f7e..60bdd19c92d 100644
--- a/src/srun/srun.c
+++ b/src/srun/srun.c
@@ -421,6 +421,7 @@ int srun(int ac, char **av)
/* have to check if job was cancelled here just to make sure
state didn't change when we were waiting for the message thread */
+ exitcode = set_job_rc(job);
if (job->state == SRUN_JOB_CANCELLED) {
info("Cancelling job");
srun_job_destroy(job, NO_VAL);
@@ -428,7 +429,7 @@ int srun(int ac, char **av)
info("Terminating job");
srun_job_destroy(job, job->rc);
} else
- srun_job_destroy(job, 0);
+ srun_job_destroy(job, job->rc);
/* wait for launch thread */
if (pthread_join(job->lid, NULL) < 0)
@@ -455,7 +456,6 @@ int srun(int ac, char **av)
/*
* Let exit() clean up remaining threads.
*/
- exitcode = job_rc(job);
log_fini();
exit(exitcode);
}
diff --git a/src/srun/srun_job.c b/src/srun/srun_job.c
index b6e7fc2adde..1d28c43346f 100644
--- a/src/srun/srun_job.c
+++ b/src/srun/srun_job.c
@@ -556,32 +556,29 @@ job_force_termination(srun_job_t *job)
int
-job_rc(srun_job_t *job)
+set_job_rc(srun_job_t *job)
{
- int i;
- int rc = 0;
-
- if (job->rc >= 0) return(job->rc);
+ int i, rc = 0, task_failed = 0;
/*
- * return (1) if any tasks failed launch
+ * return code set to at least one if any tasks failed launch
*/
for (i = 0; i < opt.nprocs; i++) {
- if (job->task_state[i] == SRUN_TASK_FAILED)
- return (job->rc = 1);
- }
-
- for (i = 0; i < opt.nprocs; i++) {
+ if (job->task_state[i] == SRUN_TASK_FAILED)
+ task_failed = 1;
if (job->rc < job->tstatus[i])
job->rc = job->tstatus[i];
}
+ if (task_failed && (job->rc <= 0)) {
+ job->rc = 1;
+ return 1;
+ }
if ((rc = WEXITSTATUS(job->rc)))
- job->rc = rc;
- else if (WIFSIGNALED(job->rc))
- job->rc = 128 + WTERMSIG(job->rc);
-
- return(job->rc);
+ return rc;
+ if (WIFSIGNALED(job->rc))
+ return (128 + WTERMSIG(job->rc));
+ return job->rc;
}
diff --git a/src/srun/srun_job.h b/src/srun/srun_job.h
index 69c3d728283..d2afa64ffce 100644
--- a/src/srun/srun_job.h
+++ b/src/srun/srun_job.h
@@ -211,9 +211,10 @@ void report_task_status(srun_job_t *job);
void report_job_status(srun_job_t *job);
/*
+ * Sets job->rc to highest task exit value.
* Returns job return code (for srun exit status)
*/
-int job_rc(srun_job_t *job);
+int set_job_rc(srun_job_t *job);
void fwd_signal(srun_job_t *job, int signal, int max_threads);
int job_active_tasks_on_host(srun_job_t *job, int hostid);
--
GitLab