diff --git a/NEWS b/NEWS
index 5fb6752e1f48cc272cc172f5e45fa808c480e2a5..259ce99919bddc46e723d0f6ad66651982155d57 100644
--- a/NEWS
+++ b/NEWS
@@ -3,6 +3,7 @@ documents those changes that are of interest to users and admins.
* Changes in SLURM 1.2.0-pre6
=============================
+ -- Maintain actually job step run time with suspend/resume use.
* Changes in SLURM 1.2.0-pre5
=============================
diff --git a/slurm/slurm.h.in b/slurm/slurm.h.in
index d3599f1bed426573ed005e0b0926a91b41282e7f..084e6f683fa0f4c9f6db461fdd89545b530617cc 100644
--- a/slurm/slurm.h.in
+++ b/slurm/slurm.h.in
@@ -680,6 +680,7 @@ typedef struct {
uint32_t user_id; /* user the job runs as */
uint32_t num_tasks; /* number of tasks */
time_t start_time; /* step start time */
+ time_t run_time; /* net run time (factor out time suspended) */
char *partition; /* name of assigned partition */
char *nodes; /* list of nodes allocated to job_step */
char *name; /* name of job step */
diff --git a/src/common/slurm_protocol_pack.c b/src/common/slurm_protocol_pack.c
index ab468ed553ce28e375ce94f423924add7f293bb4..8f7dad2950cb761b8d28bf76e2448e69ca8f265e 100644
--- a/src/common/slurm_protocol_pack.c
+++ b/src/common/slurm_protocol_pack.c
@@ -1827,6 +1827,7 @@ _unpack_job_step_info_members(job_step_info_t * step, Buf buffer)
safe_unpack32(&step->num_tasks, buffer);
safe_unpack_time(&step->start_time, buffer);
+ safe_unpack_time(&step->run_time, buffer);
safe_unpackstr_xmalloc(&step->partition, &uint16_tmp, buffer);
safe_unpackstr_xmalloc(&step->nodes, &uint16_tmp, buffer);
safe_unpackstr_xmalloc(&step->name, &uint16_tmp, buffer);
diff --git a/src/slurmctld/job_mgr.c b/src/slurmctld/job_mgr.c
index 4624d5f799523084b9a5d0347ac6718723b5879b..f682a4c137998af181be1f4436ac024e6dee6c60 100644
--- a/src/slurmctld/job_mgr.c
+++ b/src/slurmctld/job_mgr.c
@@ -943,6 +943,7 @@ static void _dump_job_step_state(struct step_record *step_ptr, Buf buffer)
}
pack_time(step_ptr->start_time, buffer);
+ pack_time(step_ptr->pre_sus_time, buffer);
packstr(step_ptr->host, buffer);
packstr(step_ptr->name, buffer);
packstr(step_ptr->network, buffer);
@@ -960,7 +961,7 @@ static int _load_step_state(struct job_record *job_ptr, Buf buffer)
struct step_record *step_ptr = NULL;
uint16_t step_id, cyclic_alloc, name_len, port, batch_step, bit_cnt;
uint32_t exit_code;
- time_t start_time;
+ time_t start_time, pre_sus_time;
char *host = NULL;
char *name = NULL, *network = NULL, *bit_fmt = NULL;
switch_jobinfo_t switch_tmp = NULL;
@@ -977,6 +978,7 @@ static int _load_step_state(struct job_record *job_ptr, Buf buffer)
}
safe_unpack_time(&start_time, buffer);
+ safe_unpack_time(&pre_sus_time, buffer);
safe_unpackstr_xmalloc(&host, &name_len, buffer);
safe_unpackstr_xmalloc(&name, &name_len, buffer);
safe_unpackstr_xmalloc(&network, &name_len, buffer);
@@ -1015,6 +1017,7 @@ static int _load_step_state(struct job_record *job_ptr, Buf buffer)
step_ptr->batch_step = batch_step;
host = NULL; /* re-used, nothing left to free */
step_ptr->start_time = start_time;
+ step_ptr->pre_sus_time = pre_sus_time;
slurm_step_layout_destroy(step_ptr->step_layout);
step_ptr->step_layout = step_layout;
@@ -4479,6 +4482,7 @@ extern int job_suspend(suspend_msg_t *sus_ptr, uid_t uid,
difftime(now,
job_ptr->start_time);
}
+ suspend_job_step(job_ptr);
} else if (sus_ptr->op == RESUME_JOB) {
if (job_ptr->job_state != JOB_SUSPENDED) {
rc = ESLURM_DISABLED;
diff --git a/src/slurmctld/slurmctld.h b/src/slurmctld/slurmctld.h
index 0bb3399292cea65ef81c7c60b2bdf343e07971b0..17b0e3deebdfd271a0805446d4445f7ecd0e585e 100644
--- a/src/slurmctld/slurmctld.h
+++ b/src/slurmctld/slurmctld.h
@@ -391,6 +391,10 @@ struct step_record {
uint16_t cyclic_alloc; /* set for cyclic task allocation
across nodes */
time_t start_time; /* step allocation time */
+/* time_t suspend_time; * time step last suspended or resumed
+ * implicitly the same as suspend_time
+ * in the job record */
+ time_t pre_sus_time; /* time step ran prior to last suspend */
bitstr_t *step_node_bitmap; /* bitmap of nodes allocated to job
step */
uint16_t port; /* port for srun communications */
@@ -1256,6 +1260,9 @@ extern bool step_on_node(struct job_record *job_ptr,
extern int step_partial_comp(step_complete_msg_t *req, int *rem,
int *max_rc);
+/* Update time stamps for job step suspend */
+extern void suspend_job_step(struct job_record *job_ptr);
+
/*
* Synchronize the batch job in the system with their files.
* All pending batch jobs must have script and environment files
diff --git a/src/slurmctld/step_mgr.c b/src/slurmctld/step_mgr.c
index 9275a1580a83fc5de10da22c1bf8ad15139e7db6..e6fc8ecddb4de9e1a834a0958064f82e5bb7e525 100644
--- a/src/slurmctld/step_mgr.c
+++ b/src/slurmctld/step_mgr.c
@@ -845,29 +845,39 @@ extern slurm_step_layout_t *step_layout_create(struct step_record *step_ptr,
* IN step - pointer to a job step record
* IN/OUT buffer - location to store data, pointers automatically advanced
*/
-static void _pack_ctld_job_step_info(struct step_record *step, Buf buffer)
+static void _pack_ctld_job_step_info(struct step_record *step_ptr, Buf buffer)
{
int task_cnt;
char *node_list = NULL;
+ time_t begin_time, run_time;
- if(step->step_layout) {
- task_cnt = step->step_layout->task_cnt;
- node_list = step->step_layout->node_list;
+ if (step_ptr->step_layout) {
+ task_cnt = step_ptr->step_layout->task_cnt;
+ node_list = step_ptr->step_layout->node_list;
} else {
- task_cnt = step->job_ptr->num_procs;
- node_list = step->job_ptr->nodes;
+ task_cnt = step_ptr->job_ptr->num_procs;
+ node_list = step_ptr->job_ptr->nodes;
}
- pack32((uint32_t)step->job_ptr->job_id, buffer);
- pack16((uint16_t)step->step_id, buffer);
- pack32((uint32_t)step->job_ptr->user_id, buffer);
+ pack32((uint32_t)step_ptr->job_ptr->job_id, buffer);
+ pack16((uint16_t)step_ptr->step_id, buffer);
+ pack32((uint32_t)step_ptr->job_ptr->user_id, buffer);
pack32((uint32_t)task_cnt, buffer);
- pack_time(step->start_time, buffer);
- packstr(step->job_ptr->partition, buffer);
+ pack_time(step_ptr->start_time, buffer);
+ if (step_ptr->job_ptr->job_state == JOB_SUSPENDED) {
+ run_time = step_ptr->pre_sus_time;
+ } else {
+ begin_time = MAX(step_ptr->start_time,
+ step_ptr->job_ptr->suspend_time);
+ run_time = step_ptr->pre_sus_time +
+ difftime(time(NULL), begin_time);
+ }
+ pack_time(run_time, buffer);
+ packstr(step_ptr->job_ptr->partition, buffer);
packstr(node_list, buffer);
- packstr(step->name, buffer);
- packstr(step->network, buffer);
- pack_bit_fmt(step->step_node_bitmap, buffer);
+ packstr(step_ptr->name, buffer);
+ packstr(step_ptr->network, buffer);
+ pack_bit_fmt(step_ptr->step_node_bitmap, buffer);
}
@@ -1332,3 +1342,35 @@ extern int step_epilog_complete(struct job_record *job_ptr,
return rc;
}
+static void
+_suspend_job_step(struct job_record *job_ptr,
+ struct step_record *step_ptr, time_t now)
+{
+ if ((job_ptr->suspend_time)
+ && (job_ptr->suspend_time > step_ptr->start_time)) {
+ step_ptr->pre_sus_time +=
+ difftime(now,
+ job_ptr->suspend_time);
+ } else {
+ step_ptr->pre_sus_time +=
+ difftime(now,
+ step_ptr->start_time);
+ }
+
+}
+
+/* Update time stamps for job step suspend */
+extern void
+suspend_job_step(struct job_record *job_ptr)
+{
+ time_t now = time(NULL);
+ ListIterator step_iterator;
+ struct step_record *step_ptr;
+
+ step_iterator = list_iterator_create (job_ptr->step_list);
+ while ((step_ptr = (struct step_record *) list_next (step_iterator))) {
+ _suspend_job_step(job_ptr, step_ptr, now);
+ }
+ list_iterator_destroy (step_iterator);
+}
+
diff --git a/src/squeue/print.c b/src/squeue/print.c
index 76cedac0352c443914a1ffe2d2548e72ba3352d7..3ed0857d8a86e7e1192156ff05c1f9c0eec97420 100644
--- a/src/squeue/print.c
+++ b/src/squeue/print.c
@@ -1158,7 +1158,7 @@ int _print_step_time_used(job_step_info_t * step, int width, bool right,
if (step == NULL) /* Print the Header instead */
_print_str("TIME", width, right, true);
else {
- long delta_t = difftime(time(NULL), step->start_time);
+ long delta_t = step->run_time;
_print_secs(delta_t, width, right, false);
}
if (suffix)