From 250a02121d2080af92f508226be3169ffd3bd742 Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Thu, 5 Dec 2002 19:34:10 +0000
Subject: [PATCH] Add flag to job when initiated with
allocated_resources_and_run_job_step(). If set, the job step termination is
treated like job termination (i.e. the nodes are released and the job state
set to complete).
---
src/slurmctld/controller.c | 8 +++++---
src/slurmctld/slurmctld.h | 11 +++++++++--
src/slurmctld/step_mgr.c | 13 +++++++++++--
3 files changed, 25 insertions(+), 7 deletions(-)
diff --git a/src/slurmctld/controller.c b/src/slurmctld/controller.c
index 6526f471d4d..8b78a11487f 100644
--- a/src/slurmctld/controller.c
+++ b/src/slurmctld/controller.c
@@ -1026,6 +1026,7 @@ static void _slurm_rpc_job_step_complete(slurm_msg_t * msg)
debug("Processing RPC: REQUEST_COMPLETE_JOB_STEP");
uid = slurm_auth_uid(msg->cred);
lock_slurmctld(job_write_lock);
+
/* do RPC call */
/* First set node down as needed on fatal error */
if (complete_job_step_msg->slurm_rc != SLURM_SUCCESS) {
@@ -1077,7 +1078,8 @@ static void _slurm_rpc_job_step_complete(slurm_msg_t * msg)
error_code =
job_step_complete(complete_job_step_msg->job_id,
complete_job_step_msg->job_step_id,
- uid);
+ uid, job_requeue,
+ complete_job_step_msg->job_rc);
unlock_slurmctld(job_write_lock);
/* return result */
@@ -1502,7 +1504,7 @@ static void _slurm_rpc_allocate_and_run(slurm_msg_t * msg)
req_step_msg.node_count = INFINITE;
req_step_msg.cpu_count = job_desc_msg->num_procs;
req_step_msg.num_tasks = job_desc_msg->num_tasks;
- error_code = step_create(&req_step_msg, &step_rec);
+ error_code = step_create(&req_step_msg, &step_rec, true);
/* note: no need to free step_rec, pointer to global job step record */
if (error_code) {
unlock_slurmctld(job_write_lock);
@@ -1841,7 +1843,7 @@ static void _slurm_rpc_job_step_create(slurm_msg_t * msg)
if (error_code == 0) {
/* issue the RPC */
lock_slurmctld(job_write_lock);
- error_code = step_create(req_step_msg, &step_rec);
+ error_code = step_create(req_step_msg, &step_rec, false);
}
/* return result */
diff --git a/src/slurmctld/slurmctld.h b/src/slurmctld/slurmctld.h
index fcaf15b2bfc..92be49c958e 100644
--- a/src/slurmctld/slurmctld.h
+++ b/src/slurmctld/slurmctld.h
@@ -224,6 +224,8 @@ struct job_record {
enum job_states job_state; /* state of the job */
uint16_t kill_on_node_fail; /* 1 if job should be killed on
node failure */
+ uint16_t kill_on_step_done; /* 1 if job should be killed when
+ the job step completes */
char *nodes; /* list of nodes allocated to job */
bitstr_t *node_bitmap; /* bitmap of nodes allocated to job */
uint32_t time_limit; /* time_limit minutes or INFINITE */
@@ -597,12 +599,14 @@ extern int job_complete (uint32_t job_id, uid_t uid, bool requeue,
* IN job_id - id of the job to be completed
* IN step_id - id of the job step to be completed
* IN uid - user id of user issuing the RPC
+ * IN requeue - job should be run again if possible
+ * IN job_return_code - job's return code, if set then set state to JOB_FAILED
* RET 0 on success, otherwise ESLURM error code
* global: job_list - pointer global job list
* last_job_update - time of last job table update
*/
extern int job_step_complete (uint32_t job_id, uint32_t job_step_id,
- uid_t uid);
+ uid_t uid, bool requeue, uint32_t job_return_code);
/*
* job_time_limit - terminate jobs which have exceeded their time limit
@@ -853,11 +857,14 @@ extern void set_slurmd_addr (void);
* accoding to the step_specs.
* IN step_specs - job step specifications
* OUT new_step_record - pointer to the new step_record (NULL on error)
+ * IN kill_job_when_step_done - if set kill the job on step completion
* RET - 0 or error code
* NOTE: don't free the returned step_record because that is managed through
* the job.
*/
-extern int step_create ( step_specs *step_specs, struct step_record** );
+extern int step_create ( step_specs *step_specs,
+ struct step_record** new_step_record,
+ bool kill_job_when_step_done );
/*
* update_job - update a job's parameters per the supplied specifications
diff --git a/src/slurmctld/step_mgr.c b/src/slurmctld/step_mgr.c
index 09d4f35f9c4..7460c4dd93a 100644
--- a/src/slurmctld/step_mgr.c
+++ b/src/slurmctld/step_mgr.c
@@ -249,11 +249,14 @@ int job_step_cancel(uint32_t job_id, uint32_t step_id, uid_t uid)
* IN job_id - id of the job to be completed
* IN step_id - id of the job step to be completed
* IN uid - user id of user issuing the RPC
+ * IN requeue - job should be run again if possible
+ * IN job_return_code - job's return code, if set then set state to JOB_FAILED
* RET 0 on success, otherwise ESLURM error code
* global: job_list - pointer global job list
* last_job_update - time of last job table update
*/
-int job_step_complete(uint32_t job_id, uint32_t step_id, uid_t uid)
+int job_step_complete(uint32_t job_id, uint32_t step_id, uid_t uid,
+ bool requeue, uint32_t job_return_code)
{
struct job_record *job_ptr;
int error_code;
@@ -264,6 +267,9 @@ int job_step_complete(uint32_t job_id, uint32_t step_id, uid_t uid)
return ESLURM_INVALID_JOB_ID;
}
+ if (job_ptr->kill_on_step_done)
+ return job_complete(job_id, uid, requeue, job_return_code);
+
if ((job_ptr->job_state == JOB_FAILED) ||
(job_ptr->job_state == JOB_COMPLETE) ||
(job_ptr->job_state == JOB_TIMEOUT))
@@ -413,12 +419,14 @@ cleanup:
* according to the step_specs.
* IN step_specs - job step specifications
* OUT new_step_record - pointer to the new step_record (NULL on error)
+ * IN kill_job_when_step_done - if set kill the job on step completion
* RET - 0 or error code
* NOTE: don't free the returned step_record because that is managed through
* the job.
*/
int
-step_create ( step_specs *step_specs, struct step_record** new_step_record )
+step_create ( step_specs *step_specs, struct step_record** new_step_record,
+ bool kill_job_when_step_done )
{
struct step_record *step_ptr;
struct job_record *job_ptr;
@@ -450,6 +458,7 @@ step_create ( step_specs *step_specs, struct step_record** new_step_record )
#endif
job_ptr->time_last_active = time(NULL);
+ job_ptr->kill_on_step_done = kill_job_when_step_done;
nodeset = _pick_step_nodes (job_ptr, step_specs);
if (nodeset == NULL)
return ESLURM_REQUESTED_NODE_CONFIG_UNAVAILABLE ;
--
GitLab