From 9cd903cb394fece00e25ddeafec33150d9e04d76 Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Tue, 30 Nov 2004 22:56:45 +0000
Subject: [PATCH] Define new function for plugin to terminate jobs on
initiation failure.
---
src/slurmctld/job_mgr.c | 35 +++++++++++++++++++++++++++++++++++
src/slurmctld/proc_req.c | 27 +++++++++++++++++++++++++--
src/slurmctld/proc_req.h | 22 +++++++++++++++++++++-
src/slurmctld/slurmctld.h | 17 ++++++++---------
4 files changed, 89 insertions(+), 12 deletions(-)
diff --git a/src/slurmctld/job_mgr.c b/src/slurmctld/job_mgr.c
index 37758ea0541..b3a84914ad8 100644
--- a/src/slurmctld/job_mgr.c
+++ b/src/slurmctld/job_mgr.c
@@ -1268,6 +1268,41 @@ extern int job_allocate(job_desc_msg_t * job_specs, int immediate, int will_run,
return SLURM_SUCCESS;
}
+/*
+ * job_fail - terminate a job due to initiation failure
+ * IN job_id - id of the job to be killed
+ * RET 0 on success, otherwise ESLURM error code
+ */
+extern int job_fail(uint32_t job_id)
+{
+ struct job_record *job_ptr;
+ time_t now = time(NULL);
+
+ job_ptr = find_job_record(job_id);
+ if (job_ptr == NULL) {
+ error("job_fail: invalid job id %u", job_id);
+ return ESLURM_INVALID_JOB_ID;
+ }
+
+ if (IS_JOB_FINISHED(job_ptr))
+ return ESLURM_ALREADY_DONE;
+ if (job_ptr->job_state == JOB_RUNNING) {
+ /* No need to signal steps, deallocate kills them */
+ job_ptr->time_last_active = now;
+ job_ptr->end_time = now;
+ last_job_update = now;
+ job_ptr->job_state = JOB_FAILED | JOB_COMPLETING;
+ deallocate_nodes(job_ptr, false);
+ job_completion_logger(job_ptr);
+ return SLURM_SUCCESS;
+ }
+ /* All other states */
+ verbose("job_fail: job %u can't be killed from state=%s",
+ job_id, job_state_string(job_ptr->job_state));
+ return ESLURM_TRANSITION_STATE_NO_UPDATE;
+
+}
+
/*
* job_signal - signal the specified job
* IN job_id - id of the job to be signaled
diff --git a/src/slurmctld/proc_req.c b/src/slurmctld/proc_req.c
index 63970823b5d..ee31935b74d 100644
--- a/src/slurmctld/proc_req.c
+++ b/src/slurmctld/proc_req.c
@@ -1507,11 +1507,12 @@ static void _slurm_rpc_update_job(slurm_msg_t * msg)
/*
* slurm_drain_nodes - process a request to drain a list of nodes,
- * no-op for nodes already drained or draining
+ * no-op for nodes already drained or draining
* node_list IN - list of nodes to drain
* reason IN - reason to drain the nodes
* RET SLURM_SUCCESS or error code
- * NOTE: This is utilzed by plugins and not via RPC
+ * NOTE: This is utilzed by plugins and not via RPC and it sets its
+ * own locks.
*/
extern int slurm_drain_nodes(char *node_list, char *reason)
{
@@ -1527,6 +1528,28 @@ extern int slurm_drain_nodes(char *node_list, char *reason)
return error_code;
}
+/*
+ * slurm_fail_job - terminate a job due to a launch failure
+ * no-op for jobs already terminated
+ * job_id IN - slurm job id
+ * RET SLURM_SUCCESS or error code
+ * NOTE: This is utilzed by plugins and not via RPC and it sets its
+ * own locks.
+ */
+extern int slurm_fail_job(uint32_t job_id)
+{
+ int error_code;
+ /* Locks: Write job and node */
+ slurmctld_lock_t job_write_lock = {
+ NO_LOCK, WRITE_LOCK, WRITE_LOCK, NO_LOCK };
+
+ lock_slurmctld(job_write_lock);
+ error_code = job_fail(job_id);
+ unlock_slurmctld(job_write_lock);
+
+ return error_code;
+}
+
/* _slurm_rpc_update_node - process RPC to update the configuration of a
* node (e.g. UP/DOWN) */
static void _slurm_rpc_update_node(slurm_msg_t * msg)
diff --git a/src/slurmctld/proc_req.h b/src/slurmctld/proc_req.h
index 324f7a5efaf..5be6ad8709f 100644
--- a/src/slurmctld/proc_req.h
+++ b/src/slurmctld/proc_req.h
@@ -3,7 +3,7 @@
*****************************************************************************
* Copyright (C) 2002 The Regents of the University of California.
* Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
- * Written by Moe Jette <jette@llnl.gov>, Kevin Tew <tew1@llnl.gov>, et. al.
+ * Written by Morris Jette <jette1@llnl.gov> and Kevin Tew <tew1@llnl.gov>
* UCRL-CODE-2002-040.
*
* This file is part of SLURM, a resource management program.
@@ -62,5 +62,25 @@ extern inline void diff_tv_str(struct timeval *tv1,struct timeval *tv2,
*/
void slurmctld_req (slurm_msg_t * msg);
+/*
+ * slurm_drain_nodes - process a request to drain a list of nodes,
+ * no-op for nodes already drained or draining
+ * node_list IN - list of nodes to drain
+ * reason IN - reason to drain the nodes
+ * RET SLURM_SUCCESS or error code
+ * NOTE: This is utilzed by plugins and not via RPC and it sets its
+ * own locks.
+ */
+extern int slurm_drain_nodes(char *node_list, char *reason);
+
+/*
+ * slurm_fail_job - terminate a job due to a launch failure
+ * no-op for jobs already terminated
+ * job_id IN - slurm job id
+ * RET SLURM_SUCCESS or error code
+ * NOTE: This is utilzed by plugins and not via RPC and it sets its
+ * own locks.
+ */
+extern int slurm_fail_job(uint32_t job_id);
#endif /* !_HAVE_PROC_REQ_H */
diff --git a/src/slurmctld/slurmctld.h b/src/slurmctld/slurmctld.h
index 8a84bdb0bbf..9a23b1082c8 100644
--- a/src/slurmctld/slurmctld.h
+++ b/src/slurmctld/slurmctld.h
@@ -3,7 +3,7 @@
*****************************************************************************
* Copyright (C) 2002 The Regents of the University of California.
* Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
- * Written by Morris Jette <jette@llnl.gov> et. al.
+ * Written by Morris Jette <jette1@llnl.gov> et. al.
* UCRL-CODE-2002-040.
*
* This file is part of SLURM, a resource management program.
@@ -620,6 +620,13 @@ extern void job_fini (void);
*/
extern bool job_is_completing(void);
+/*
+ * job_fail - terminate a job due to initiation failure
+ * IN job_id - id of the job to be killed
+ * RET 0 on success, otherwise ESLURM error code
+ */
+extern int job_fail(uint32_t job_id);
+
/*
* job_signal - signal the specified job
* IN job_id - id of the job to be signaled
@@ -1029,14 +1036,6 @@ extern void signal_step_tasks(struct step_record *step_ptr, uint16_t signal);
*/
extern int slurmctld_shutdown(void);
-/*
- * slurm_drain_nodes - process a request to drain a list of nodes
- * node_list IN - list of nodes to drain
- * reason IN - reason to drain the nodes
- * RET SLURM_SUCCESS or error code
- */
-extern int slurm_drain_nodes(char *node_list, char *reason);
-
/*
* step_create - creates a step_record in step_specs->job_id, sets up the
* accoding to the step_specs.
--
GitLab