From 4d43c7f89255c33c40da3bf9c1fd6413c1d3549e Mon Sep 17 00:00:00 2001
From: Morris Jette <jette@schedmd.com>
Date: Tue, 15 Jul 2014 15:50:37 -0700
Subject: [PATCH] Refactor job suspend/resume logic
For improved support of job arrays
---
src/slurmctld/job_mgr.c | 136 +++++++++++++++++++++-------------------
1 file changed, 71 insertions(+), 65 deletions(-)
diff --git a/src/slurmctld/job_mgr.c b/src/slurmctld/job_mgr.c
index 136052e8371..0436f978026 100644
--- a/src/slurmctld/job_mgr.c
+++ b/src/slurmctld/job_mgr.c
@@ -11109,95 +11109,58 @@ static int _job_resume_test(struct job_record *job_ptr)
}
/*
- * job_suspend - perform some suspend/resume operation
- * IN sus_ptr - suspend/resume request message
- * IN uid - user id of the user issuing the RPC
- * IN conn_fd - file descriptor on which to send reply,
- * -1 if none
+ * _job_suspend - perform some suspend/resume operation
+ * job_ptr - job to operate upon
+ * op IN - operation: suspend/resume
* indf_susp IN - set if job is being suspended indefinitely by user or admin
* and we should clear it's priority, otherwise suspended
* temporarily for gang scheduling
- * IN protocol_version - slurm protocol version of client
* RET 0 on success, otherwise ESLURM error code
*/
-extern int job_suspend(suspend_msg_t *sus_ptr, uid_t uid,
- slurm_fd_t conn_fd, bool indf_susp,
- uint16_t protocol_version)
+static int _job_suspend(struct job_record *job_ptr, uint16_t op, bool indf_susp)
{
int rc = SLURM_SUCCESS;
time_t now = time(NULL);
- struct job_record *job_ptr = NULL;
- slurm_msg_t resp_msg;
- return_code_msg_t rc_msg;
-
-#ifdef HAVE_BG
- rc = ESLURM_NOT_SUPPORTED;
- goto reply;
-#endif
-
- /* find the job */
- job_ptr = find_job_record (sus_ptr->job_id);
- if (job_ptr == NULL) {
- rc = ESLURM_INVALID_JOB_ID;
- goto reply;
- }
- /* validate the request */
- if ((uid != 0) && (uid != getuid())) {
- rc = ESLURM_ACCESS_DENIED;
- goto reply;
- }
- if (IS_JOB_PENDING(job_ptr)) {
- rc = ESLURM_JOB_PENDING;
- goto reply;
- }
- if (IS_JOB_FINISHED(job_ptr)) {
- rc = ESLURM_ALREADY_DONE;
- goto reply;
- }
- if ((sus_ptr->op == SUSPEND_JOB) &&
- (_job_suspend_switch_test(job_ptr) != SLURM_SUCCESS)) {
- rc = ESLURM_NOT_SUPPORTED;
- goto reply;
- }
- if ((sus_ptr->op == RESUME_JOB) && (rc = _job_resume_test(job_ptr)))
- goto reply;
+ if (IS_JOB_PENDING(job_ptr))
+ return ESLURM_JOB_PENDING;
+ if (IS_JOB_FINISHED(job_ptr))
+ return ESLURM_ALREADY_DONE;
+ if ((op == SUSPEND_JOB) &&
+ (_job_suspend_switch_test(job_ptr) != SLURM_SUCCESS))
+ return ESLURM_NOT_SUPPORTED;
+ if ((op == RESUME_JOB) && (rc = _job_resume_test(job_ptr)))
+ return rc;
/* Notify salloc/srun of suspend/resume */
- srun_job_suspend(job_ptr, sus_ptr->op);
+ srun_job_suspend(job_ptr, op);
/* perform the operation */
- if (sus_ptr->op == SUSPEND_JOB) {
- if (!IS_JOB_RUNNING(job_ptr)) {
- rc = ESLURM_JOB_NOT_RUNNING;
- goto reply;
- }
+ if (op == SUSPEND_JOB) {
+ if (!IS_JOB_RUNNING(job_ptr))
+ return ESLURM_JOB_NOT_RUNNING;
rc = _suspend_job_nodes(job_ptr, indf_susp);
if (rc != SLURM_SUCCESS)
- goto reply;
- _suspend_job(job_ptr, sus_ptr->op, indf_susp);
+ return rc;
+ _suspend_job(job_ptr, op, indf_susp);
job_ptr->job_state = JOB_SUSPENDED;
if (indf_susp)
job_ptr->priority = 0;
if (job_ptr->suspend_time) {
job_ptr->pre_sus_time +=
- difftime(now,
- job_ptr->suspend_time);
+ difftime(now, job_ptr->suspend_time);
} else {
job_ptr->pre_sus_time +=
- difftime(now,
- job_ptr->start_time);
+ difftime(now, job_ptr->start_time);
}
suspend_job_step(job_ptr);
- } else if (sus_ptr->op == RESUME_JOB) {
- if (!IS_JOB_SUSPENDED(job_ptr)) {
- rc = ESLURM_JOB_NOT_SUSPENDED;
- goto reply;
- }
+ } else if (op == RESUME_JOB) {
+ if (!IS_JOB_SUSPENDED(job_ptr))
+ return ESLURM_JOB_NOT_SUSPENDED;
rc = _resume_job_nodes(job_ptr, indf_susp);
if (rc != SLURM_SUCCESS)
- goto reply;
- _suspend_job(job_ptr, sus_ptr->op, indf_susp);
+ return rc;
+ _suspend_job(job_ptr, op, indf_susp);
if (job_ptr->priority == 0)
set_job_prio(job_ptr);
job_ptr->job_state = JOB_RUNNING;
@@ -11218,8 +11181,7 @@ extern int job_suspend(suspend_msg_t *sus_ptr, uid_t uid,
(!job_ptr->preempt_time)) {
debug3("Job %u resumed, updating end_time",
job_ptr->job_id);
- job_ptr->end_time = now +
- (job_ptr->time_limit * 60)
+ job_ptr->end_time = now + (job_ptr->time_limit * 60)
- job_ptr->pre_sus_time;
}
resume_job_step(job_ptr);
@@ -11229,6 +11191,50 @@ extern int job_suspend(suspend_msg_t *sus_ptr, uid_t uid,
job_ptr->suspend_time = now;
jobacct_storage_g_job_suspend(acct_db_conn, job_ptr);
+ return rc;
+}
+
+/*
+ * job_suspend - perform some suspend/resume operation
+ * IN sus_ptr - suspend/resume request message
+ * IN uid - user id of the user issuing the RPC
+ * IN conn_fd - file descriptor on which to send reply,
+ * -1 if none
+ * indf_susp IN - set if job is being suspended indefinitely by user or admin
+ * and we should clear it's priority, otherwise suspended
+ * temporarily for gang scheduling
+ * IN protocol_version - slurm protocol version of client
+ * RET 0 on success, otherwise ESLURM error code
+ */
+extern int job_suspend(suspend_msg_t *sus_ptr, uid_t uid,
+ slurm_fd_t conn_fd, bool indf_susp,
+ uint16_t protocol_version)
+{
+ int rc = SLURM_SUCCESS;
+ struct job_record *job_ptr = NULL;
+ slurm_msg_t resp_msg;
+ return_code_msg_t rc_msg;
+
+#ifdef HAVE_BG
+ rc = ESLURM_NOT_SUPPORTED;
+ goto reply;
+#endif
+
+ /* find the job */
+ job_ptr = find_job_record (sus_ptr->job_id);
+ if (job_ptr == NULL) {
+ rc = ESLURM_INVALID_JOB_ID;
+ goto reply;
+ }
+
+ /* validate the request */
+ if ((uid != 0) && (uid != getuid())) {
+ rc = ESLURM_ACCESS_DENIED;
+ goto reply;
+ }
+
+ rc = _job_suspend(job_ptr, sus_ptr->op, indf_susp);
+
reply:
if (conn_fd >= 0) {
slurm_msg_t_init(&resp_msg);
--
GitLab