From e4c4a8bbaf89f88147eb4e7b23c1307e4657af7b Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Fri, 3 Sep 2004 21:08:02 +0000
Subject: [PATCH] Send SIGTERM to batch script before SIGKILL for mpirun to
cleanup on BGL.
---
NEWS | 2 ++
src/slurmd/req.c | 29 ++++++++++++++++-------------
2 files changed, 18 insertions(+), 13 deletions(-)
diff --git a/NEWS b/NEWS
index bcb203892e0..ba878f00aae 100644
--- a/NEWS
+++ b/NEWS
@@ -3,6 +3,8 @@ documents those changes that are of interest to users and admins.
* Changes in SLURM 0.4.0-pre3
=============================
+ -- Send SIGTERM to batch script before SIGKILL for mpirun cleanup on
+ Blue Gene/L
* Changes in SLURM 0.4.0-pre2
=============================
diff --git a/src/slurmd/req.c b/src/slurmd/req.c
index e95321a8120..b84d107d0e6 100644
--- a/src/slurmd/req.c
+++ b/src/slurmd/req.c
@@ -61,7 +61,7 @@
static bool _slurm_authorized_user(uid_t uid);
static bool _job_still_running(uint32_t job_id);
-static int _kill_all_active_steps(uint32_t jobid, int sig);
+static int _kill_all_active_steps(uint32_t jobid, int sig, bool batch);
static int _launch_tasks(launch_tasks_request_msg_t *, slurm_addr *);
static void _rpc_launch_tasks(slurm_msg_t *, slurm_addr *);
static void _rpc_spawn_task(slurm_msg_t *, slurm_addr *);
@@ -675,7 +675,7 @@ _rpc_timelimit(slurm_msg_t *msg, slurm_addr *cli_addr)
}
/*
- * Indicate to slurmctld that we've recieved the message
+ * Indicate to slurmctld that we've received the message
*/
slurm_send_rc_msg(msg, SLURM_SUCCESS);
slurm_close_accepted_conn(msg->conn_fd);
@@ -687,7 +687,7 @@ _rpc_timelimit(slurm_msg_t *msg, slurm_addr *cli_addr)
*/
_kill_running_session_mgrs(req->job_id, SIGXCPU, "SIGXCPU");
- nsteps = _kill_all_active_steps(req->job_id, SIGTERM);
+ nsteps = _kill_all_active_steps(req->job_id, SIGTERM, false);
verbose( "Job %u: timeout: sent SIGTERM to %d active steps",
req->job_id, nsteps );
@@ -846,9 +846,15 @@ _rpc_reattach_tasks(slurm_msg_t *msg, slurm_addr *cli)
}
-
+/*
+ * _kill_all_active_steps - signals all steps of a job
+ * jobid IN - id of job to signal
+ * sig IN - signal to send
+ * batch IN - if true signal batch script, otherwise skip it
+ * RET count of signaled job steps (plus batch script, if applicable)
+ */
static int
-_kill_all_active_steps(uint32_t jobid, int sig)
+_kill_all_active_steps(uint32_t jobid, int sig, bool batch)
{
List steps = shm_get_steps();
ListIterator i = list_iterator_create(steps);
@@ -864,10 +870,7 @@ _kill_all_active_steps(uint32_t jobid, int sig)
continue;
}
- /* XXX?
- * We don't send anything but SIGKILL to batch jobs
- */
- if ((s->stepid == NO_VAL) && (sig != SIGKILL))
+ if ((s->stepid == NO_VAL) && (!batch))
continue;
step_cnt++;
@@ -972,15 +975,15 @@ _rpc_kill_job(slurm_msg_t *msg, slurm_addr *cli)
debug("credential for job %u revoked", req->job_id);
}
- nsteps = _kill_all_active_steps(req->job_id, SIGTERM);
+ nsteps = _kill_all_active_steps(req->job_id, SIGTERM, true);
/*
- * If there are currently no active job steps, and no
+ * If there are currently no active job steps and no
* configured epilog to run, bypass asynchronous reply and
* notify slurmctld that we have already completed this
* request.
*/
- if ((nsteps == 0) && !conf->epilog && (msg->conn_fd >= 0)) {
+ if ((nsteps == 0) && !conf->epilog) {
if (msg->conn_fd >= 0)
slurm_send_rc_msg(msg,
ESLURMD_KILL_JOB_ALREADY_COMPLETE);
@@ -1005,7 +1008,7 @@ _rpc_kill_job(slurm_msg_t *msg, slurm_addr *cli)
* Check for corpses
*/
if ( !_pause_for_job_completion (req->job_id, 5)
- && _kill_all_active_steps(req->job_id, SIGKILL) ) {
+ && _kill_all_active_steps(req->job_id, SIGKILL, true) ) {
/*
* Block until all user processes are complete.
*/
--
GitLab