diff --git a/NEWS b/NEWS
index 1339748d6d1f7398605709553474c0fd8a348794..912665216c971956e6dd5391af7ea0229d43decc 100644
--- a/NEWS
+++ b/NEWS
@@ -128,6 +128,12 @@ documents those changes that are of interest to users and admins.
to modify a job's name.
-- Add configuration check for sys/syslog.h and include it as needed.
-- Add --propagate option to sbatch for control over limit propagation.
+ -- Added Gold interface to the jobacct plugin. To configure in the config
+ file specify...
+ JobAcctType=jobacct/gold
+ JobAcctLogFile=CLUSTER_NAME:GOLD_AUTH_KEY_FILE:GOLDD_HOST:GOLDD_PORT7112
+ -- In slurmctld job record, set begin_time to time when all of a job's
+ dependencies are met.
* Changes in SLURM 1.2.20
=========================
diff --git a/src/plugins/select/bluegene/plugin/sfree.c b/src/plugins/select/bluegene/plugin/sfree.c
index 1b8ceecbe4fa27fa73743548dfe18c1ec38bbc77..6dcd10c50d03ed72872e6560c677c050cf66168f 100644
--- a/src/plugins/select/bluegene/plugin/sfree.c
+++ b/src/plugins/select/bluegene/plugin/sfree.c
@@ -113,7 +113,10 @@ int main(int argc, char *argv[])
bridge_init();
if (!have_db2) {
- printf("must be on BG SN to resolve.\n");
+ error("Required libraries can not be found "
+ "to access the Bluegene system.\nPlease "
+ "set your LD_LIBRARY_PATH correctly to "
+ "point to them.");
exit(0);
}
diff --git a/src/slurmctld/controller.c b/src/slurmctld/controller.c
index 8ce3c6a3122f9d45b29f0f0b3d09e783e2066472..16b8d2a4bb3c311d6a02cae00bfe1f394550ab88 100644
--- a/src/slurmctld/controller.c
+++ b/src/slurmctld/controller.c
@@ -918,6 +918,7 @@ static void *_slurmctld_background(void *no_data)
last_sched_time = now;
if (schedule())
last_checkpoint_time = 0; /* force state save */
+ set_job_elig_time();
}
if (difftime(now, last_trigger) > TRIGGER_INTERVAL) {
diff --git a/src/slurmctld/job_mgr.c b/src/slurmctld/job_mgr.c
index 5d3a1a9bba8a19ff6a5a53df8384026f1be5289e..f7e8677ca93609db31cae8a75418bfdd8fa9ab84 100644
--- a/src/slurmctld/job_mgr.c
+++ b/src/slurmctld/job_mgr.c
@@ -1378,13 +1378,14 @@ extern int job_allocate(job_desc_msg_t * job_specs, int immediate,
error_code = _job_create(job_specs, allocate, will_run,
&job_ptr, submit_uid);
*job_pptr = job_ptr;
+ time_t now = time(NULL);
if (error_code) {
if (immediate && job_ptr) {
job_ptr->job_state = JOB_FAILED;
job_ptr->exit_code = 1;
job_ptr->state_reason = FAIL_BAD_CONSTRAINTS;
- job_ptr->start_time = job_ptr->end_time = time(NULL);
+ job_ptr->start_time = job_ptr->end_time = now;
jobacct_storage_g_job_start(job_ptr);
job_completion_logger(job_ptr);
}
@@ -1417,7 +1418,7 @@ extern int job_allocate(job_desc_msg_t * job_specs, int immediate,
job_ptr->job_state = JOB_FAILED;
job_ptr->exit_code = 1;
job_ptr->state_reason = FAIL_BAD_CONSTRAINTS;
- job_ptr->start_time = job_ptr->end_time = time(NULL);
+ job_ptr->start_time = job_ptr->end_time = now;
jobacct_storage_g_job_start(job_ptr);
job_completion_logger(job_ptr);
if (!independent)
@@ -1435,18 +1436,23 @@ extern int job_allocate(job_desc_msg_t * job_specs, int immediate,
error_code = select_nodes(job_ptr, no_alloc, NULL);
jobacct_storage_g_job_start(job_ptr);
if (!test_only) {
- last_job_update = time(NULL);
+ last_job_update = now;
slurm_sched_schedule(); /* work for external scheduler */
}
- if ((error_code == ESLURM_NODES_BUSY)
- || (error_code == ESLURM_JOB_HELD)
- || (error_code == ESLURM_REQUESTED_PART_CONFIG_UNAVAILABLE)) {
+ if (independent &&
+ (job_ptr->details && (job_ptr->details->begin_time == 0)) &&
+ ((error_code == SLURM_SUCCESS) || (error_code == ESLURM_NODES_BUSY)))
+ job_ptr->details->begin_time = now;
+
+ if ((error_code == ESLURM_NODES_BUSY) ||
+ (error_code == ESLURM_JOB_HELD) ||
+ (error_code == ESLURM_REQUESTED_PART_CONFIG_UNAVAILABLE)) {
/* Not fatal error, but job can't be scheduled right now */
if (immediate) {
job_ptr->job_state = JOB_FAILED;
job_ptr->exit_code = 1;
job_ptr->state_reason = FAIL_BAD_CONSTRAINTS;
- job_ptr->start_time = job_ptr->end_time = time(NULL);
+ job_ptr->start_time = job_ptr->end_time = now;
job_completion_logger(job_ptr);
} else { /* job remains queued */
if (error_code == ESLURM_NODES_BUSY) {
@@ -1460,7 +1466,7 @@ extern int job_allocate(job_desc_msg_t * job_specs, int immediate,
job_ptr->job_state = JOB_FAILED;
job_ptr->exit_code = 1;
job_ptr->state_reason = FAIL_BAD_CONSTRAINTS;
- job_ptr->start_time = job_ptr->end_time = time(NULL);
+ job_ptr->start_time = job_ptr->end_time = now;
job_completion_logger(job_ptr);
return error_code;
}
@@ -1468,7 +1474,7 @@ extern int job_allocate(job_desc_msg_t * job_specs, int immediate,
if (will_run) { /* job would run, flag job destruction */
job_ptr->job_state = JOB_FAILED;
job_ptr->exit_code = 1;
- job_ptr->start_time = job_ptr->end_time = time(NULL);
+ job_ptr->start_time = job_ptr->end_time = now;
}
return SLURM_SUCCESS;
}
diff --git a/src/slurmctld/job_scheduler.c b/src/slurmctld/job_scheduler.c
index e4eb5a25b23f5eb2e294d4c45c9e05483e3633f3..74f096ce540d36891dd02b6d1f2526fafc08ba91 100644
--- a/src/slurmctld/job_scheduler.c
+++ b/src/slurmctld/job_scheduler.c
@@ -144,6 +144,46 @@ extern bool job_is_completing(void)
return completing;
}
+/*
+ * set_job_elig_time - set the eligible time for pending jobs once their
+ * dependencies are lifted (in job->details->begin_time)
+ */
+extern void set_job_elig_time(void)
+{
+ struct job_record *job_ptr = NULL;
+ struct part_record *part_ptr = NULL;
+ ListIterator job_iterator;
+ slurmctld_lock_t job_write_lock =
+ { READ_LOCK, WRITE_LOCK, WRITE_LOCK, READ_LOCK };
+ time_t now = time(NULL);
+
+ lock_slurmctld(job_write_lock);
+ job_iterator = list_iterator_create(job_list);
+ while ((job_ptr = (struct job_record *) list_next(job_iterator))) {
+ part_ptr = job_ptr->part_ptr;
+ if (job_ptr->job_state != JOB_PENDING)
+ continue;
+ if (part_ptr == NULL)
+ continue;
+ if ((job_ptr->details == NULL) || job_ptr->details->begin_time)
+ continue;
+ if (part_ptr->state_up == 0)
+ continue;
+ if ((job_ptr->time_limit != NO_VAL) &&
+ (job_ptr->time_limit > part_ptr->max_time))
+ continue;
+ if ((job_ptr->details->max_nodes != 0) &&
+ ((job_ptr->details->max_nodes < part_ptr->min_nodes) ||
+ (job_ptr->details->min_nodes > part_ptr->max_nodes)))
+ continue;
+ if (!job_independent(job_ptr))
+ continue;
+ job_ptr->details->begin_time = now;
+ }
+ list_iterator_destroy(job_iterator);
+ unlock_slurmctld(job_write_lock);
+}
+
/*
* schedule - attempt to schedule all pending jobs
* pending jobs for each partition will be scheduled in priority
diff --git a/src/slurmctld/node_scheduler.c b/src/slurmctld/node_scheduler.c
index 3c1231b92735f98c8a7d974fed8e45a26be4d8fd..3a420fb94bfc7863da0acc792787c2348fba96b0 100644
--- a/src/slurmctld/node_scheduler.c
+++ b/src/slurmctld/node_scheduler.c
@@ -1148,6 +1148,7 @@ extern int select_nodes(struct job_record *job_ptr, bool test_only,
uint32_t min_nodes, max_nodes, req_nodes;
int super_user = false;
enum job_state_reason fail_reason;
+ time_t now = time(NULL);
xassert(job_ptr);
xassert(job_ptr->magic == JOB_MAGIC);
@@ -1181,7 +1182,7 @@ extern int select_nodes(struct job_record *job_ptr, bool test_only,
fail_reason = WAIT_PART_NODE_LIMIT;
if (fail_reason != WAIT_NO_REASON) {
job_ptr->state_reason = fail_reason;
- last_job_update = time(NULL);
+ last_job_update = now;
if (job_ptr->priority == 0) /* user/admin hold */
return ESLURM_JOB_HELD;
job_ptr->priority = 1; /* sys hold, move to end of queue */
@@ -1248,7 +1249,7 @@ extern int select_nodes(struct job_record *job_ptr, bool test_only,
job_ptr->job_id);
if (job_ptr->priority != 0) /* Move to end of queue */
job_ptr->priority = 1;
- last_job_update = time(NULL);
+ last_job_update = now;
} else if (error_code == ESLURM_NODES_BUSY)
slurm_sched_job_is_pending();
goto cleanup;
@@ -1284,7 +1285,7 @@ extern int select_nodes(struct job_record *job_ptr, bool test_only,
error("select_g_update_nodeinfo(%u): %m", job_ptr->job_id);
/* not critical ... by now */
}
- job_ptr->start_time = job_ptr->time_last_active = time(NULL);
+ job_ptr->start_time = job_ptr->time_last_active = now;
if (job_ptr->time_limit == NO_VAL)
job_ptr->time_limit = part_ptr->max_time;
if (job_ptr->time_limit == INFINITE)
@@ -1719,6 +1720,7 @@ extern void build_node_details(struct job_record *job_ptr)
int error_code = SLURM_SUCCESS;
int node_inx = 0, cpu_inx = -1;
int cr_count = 0;
+ uint32_t total_procs = 0;
if ((job_ptr->node_bitmap == NULL) || (job_ptr->nodes == NULL)) {
/* No nodes allocated, we're done... */
@@ -1768,6 +1770,7 @@ extern void build_node_details(struct job_record *job_ptr)
job_ptr->cpus_per_node[cpu_inx] =
job_ptr->num_procs;
+ total_procs += job_ptr->num_procs;
job_ptr->cpu_count_reps[cpu_inx] = 1;
job_ptr->alloc_lps[0] = job_ptr->num_procs;
job_ptr->used_lps[0] = 0;
@@ -1803,7 +1806,8 @@ extern void build_node_details(struct job_record *job_ptr)
job_ptr->cpu_count_reps[cpu_inx] = 1;
} else
job_ptr->cpu_count_reps[cpu_inx]++;
-
+ total_procs += usable_lps;
+
} else {
error("Invalid node %s in JobId=%u",
this_node_name, job_ptr->job_id);
@@ -1819,6 +1823,8 @@ extern void build_node_details(struct job_record *job_ptr)
job_ptr->job_id, job_ptr->node_cnt, node_inx);
}
job_ptr->num_cpu_groups = cpu_inx + 1;
+ if (job_ptr->details)
+ job_ptr->details->total_procs = total_procs;
}
/*
diff --git a/src/slurmctld/slurmctld.h b/src/slurmctld/slurmctld.h
index 7521fe356f0062a5aff921b4343fd77ee9e96d55..e60860582910b6c458643e1f757cef9a6862129b 100644
--- a/src/slurmctld/slurmctld.h
+++ b/src/slurmctld/slurmctld.h
@@ -320,7 +320,9 @@ struct job_details {
uint32_t total_procs; /* number of allocated processors,
for accounting */
time_t submit_time; /* time of submission */
- time_t begin_time; /* start after this time */
+ time_t begin_time; /* start at this time (srun --being),
+ * resets to time first eligible
+ * (all dependencies satisfied) */
char *work_dir; /* pathname of working directory */
char **argv; /* arguments for a batch job script */
uint16_t argc; /* count of argv elements */
@@ -350,7 +352,7 @@ struct job_record {
char *nodes_completing; /* nodes still in completing state
* for this job, used to insure
* epilog is not re-run for job */
- uint32_t num_procs; /* count of required/allocated processors */
+ uint32_t num_procs; /* count of required processors */
uint32_t time_limit; /* time_limit minutes or INFINITE,
* NO_VAL implies partition max_time */
time_t start_time; /* time execution begins,
@@ -1249,6 +1251,12 @@ extern void save_all_state(void);
*/
extern int schedule (void);
+/*
+ * set_job_elig_time - set the eligible time for pending jobs once their
+ * dependencies are lifted (in job->details->begin_time)
+ */
+extern void set_job_elig_time(void);
+
/*
* set_node_down - make the specified node's state DOWN if possible
* (not in a DRAIN state), kill jobs as needed
diff --git a/src/smap/smap.c b/src/smap/smap.c
index e66bd702f85c6f5d233fbfd19afb76d7ed3fe522..be0c3473a5b316e0b05eb9d946e2844631209949 100644
--- a/src/smap/smap.c
+++ b/src/smap/smap.c
@@ -64,7 +64,7 @@ int main_xcord = 1;
int main_ycord = 1;
WINDOW *grid_win = NULL;
WINDOW *text_win = NULL;
-
+
/************
* Functions *
************/
@@ -111,7 +111,10 @@ int main(int argc, char *argv[])
#ifdef HAVE_BG_FILES
if (!have_db2) {
- printf("must be on BG SN to resolve.\n");
+ printf("Required libraries can not be found "
+ "to access the Bluegene system.\nPlease "
+ "set your LD_LIBRARY_PATH correctly to "
+ "point to them.\n");
goto part_fini;
}
@@ -149,7 +152,7 @@ int main(int argc, char *argv[])
}
part_fini:
#else
- printf("must be on BG SN to resolve.\n");
+ printf("Must be on BG System to resolve.\n");
#endif
ba_fini();
exit(0);
@@ -245,7 +248,7 @@ part_fini:
break;
#else
default:
- error("must be on a BG SYSTEM to run this command");
+ error("Must be on a BG SYSTEM to run this command");
endwin();
ba_fini();
exit(0);