From 0908fc1e3c615d2f8a319dac8f9d520409660504 Mon Sep 17 00:00:00 2001
From: Danny Auble <da@llnl.gov>
Date: Thu, 10 Dec 2009 22:29:52 +0000
Subject: [PATCH] patch from Matthew to avoid a NULL task being handled
incorrectly
---
src/common/jobacct_common.c | 2 +-
src/common/plugstack.c | 14 --------------
src/slurmd/slurmstepd/mgr.c | 25 +++++++++----------------
src/slurmd/slurmstepd/slurmstepd_job.c | 1 +
src/slurmd/slurmstepd/slurmstepd_job.h | 16 ++++++++++++++++
5 files changed, 27 insertions(+), 31 deletions(-)
diff --git a/src/common/jobacct_common.c b/src/common/jobacct_common.c
index cb352144d71..c27a7ab1390 100644
--- a/src/common/jobacct_common.c
+++ b/src/common/jobacct_common.c
@@ -1115,7 +1115,7 @@ extern struct jobacctinfo *jobacct_common_remove_task(pid_t pid, List task_list)
debug2("removing task %u pid %d from jobacct",
jobacct->max_vsize_id.taskid, jobacct->pid);
} else {
- error("pid(%d) not being watched in jobacct!", pid);
+ debug2("pid(%d) not being watched in jobacct!", pid);
}
error:
slurm_mutex_unlock(&jobacct_lock);
diff --git a/src/common/plugstack.c b/src/common/plugstack.c
index 88ecfed246d..d8a56bcb725 100644
--- a/src/common/plugstack.c
+++ b/src/common/plugstack.c
@@ -1384,20 +1384,6 @@ int spank_get_remote_options(job_options_t opts)
return (0);
}
-/*
- * Return a task info structure corresponding to pid.
- */
-static slurmd_task_info_t * job_task_info_by_pid (slurmd_job_t *job, pid_t pid)
-{
- slurmd_task_info_t *task = NULL;
- int i;
- for (i = 0; i < job->ntasks; i++) {
- if (job->task[i]->pid == pid)
- task = job->task[i];
- }
- return (task);
-}
-
static int tasks_execd (spank_t spank)
{
return ( (spank->phase == STEP_TASK_POST_FORK)
diff --git a/src/slurmd/slurmstepd/mgr.c b/src/slurmd/slurmstepd/mgr.c
index c3aa6eb88e0..92f05490b69 100644
--- a/src/slurmd/slurmstepd/mgr.c
+++ b/src/slurmd/slurmstepd/mgr.c
@@ -1332,7 +1332,6 @@ static int
_wait_for_any_task(slurmd_job_t *job, bool waitflag)
{
slurmd_task_info_t *t = NULL;
- int i;
int status;
pid_t pid;
int completed = 0;
@@ -1367,22 +1366,15 @@ _wait_for_any_task(slurmd_job_t *job, bool waitflag)
}
/*********************************************/
- /* See if the pid matches that of one of the tasks */
- for (i = 0; i < job->ntasks; i++) {
- if (job->task[i]->pid == pid) {
- t = job->task[i];
- completed++;
- break;
- }
- }
- if (t != NULL) {
- _log_task_exit(job->task[i]->gtid, pid, status);
+ if ((t = job_task_info_by_pid(job, pid))) {
+ completed++;
+ _log_task_exit(t->gtid, pid, status);
t->exited = true;
t->estatus = status;
job->envtp->env = job->env;
- job->envtp->procid = job->task[i]->gtid;
- job->envtp->localid = job->task[i]->id;
+ job->envtp->procid = t->gtid;
+ job->envtp->localid = t->id;
job->envtp->distribution = -1;
job->envtp->batch_flag = job->batch;
@@ -1403,10 +1395,11 @@ _wait_for_any_task(slurmd_job_t *job, bool waitflag)
job, -1, job->env);
xfree(my_epilog);
}
- job->envtp->procid = i;
+ job->envtp->procid = t->id;
- if (spank_task_exit (job, i) < 0)
- error ("Unable to spank task %d at exit", i);
+ if (spank_task_exit (job, t->id) < 0)
+ error ("Unable to spank task %d at exit",
+ t->id);
post_term(job);
}
diff --git a/src/slurmd/slurmstepd/slurmstepd_job.c b/src/slurmd/slurmstepd/slurmstepd_job.c
index 2f17aeb0dfb..92c67e6f619 100644
--- a/src/slurmd/slurmstepd/slurmstepd_job.c
+++ b/src/slurmd/slurmstepd/slurmstepd_job.c
@@ -638,3 +638,4 @@ _task_info_destroy(slurmd_task_info_t *t, uint16_t multi_prog)
} /* otherwise, t->argv is a pointer to job->argv */
xfree(t);
}
+
diff --git a/src/slurmd/slurmstepd/slurmstepd_job.h b/src/slurmd/slurmstepd/slurmstepd_job.h
index 0a58865faaa..0cafdade1a5 100644
--- a/src/slurmd/slurmstepd/slurmstepd_job.h
+++ b/src/slurmd/slurmstepd/slurmstepd_job.h
@@ -220,4 +220,20 @@ void srun_info_destroy(struct srun_info *srun);
slurmd_task_info_t * task_info_create(int taskid, int gtaskid,
char *ifname, char *ofname, char *efname);
+/*
+ * Return a task info structure corresponding to pid.
+ * We inline it here so that it can be included from src/common/plugstack.c
+ * without undefined symbol warnings.
+ */
+static inline slurmd_task_info_t *
+job_task_info_by_pid (slurmd_job_t *job, pid_t pid)
+{
+ int i;
+ for (i = 0; i < job->ntasks; i++) {
+ if (job->task[i]->pid == pid)
+ return (job->task[i]);
+ }
+ return (NULL);
+}
+
#endif /* !_SLURMSTEPD_JOB_H */
--
GitLab