diff --git a/src/salloc/salloc.c b/src/salloc/salloc.c
index 9222ab27b2d68ffc6cfc641dada8ee219950e4c1..ff800f5b81b683ac8a569d432cefefc1453eab5b 100644
--- a/src/salloc/salloc.c
+++ b/src/salloc/salloc.c
@@ -357,7 +357,7 @@ int main(int argc, char *argv[])
}
#else
if (!_wait_nodes_ready(alloc)) {
- if(!allocation_interrupted)
+ if (!allocation_interrupted)
error("Something is wrong with the "
"boot of the nodes.");
goto relinquish;
@@ -1076,6 +1076,8 @@ static int _wait_nodes_ready(resource_allocation_response_msg_t *alloc)
pending_job_id = alloc->job_id;
+ if (alloc->alias_list && !strcmp(alloc->alias_list, "TBD"))
+ opt.wait_all_nodes = 1; /* Wait for boot & addresses */
if (opt.wait_all_nodes == (uint16_t) NO_VAL)
opt.wait_all_nodes = DEFAULT_WAIT_ALL_NODES;
@@ -1110,8 +1112,18 @@ static int _wait_nodes_ready(resource_allocation_response_msg_t *alloc)
break;
}
if (is_ready) {
+ resource_allocation_response_msg_t *resp;
+ char *tmp_str;
if (i > 0)
- info ("Nodes %s are ready for job", alloc->node_list);
+ info("Nodes %s are ready for job", alloc->node_list);
+ if (alloc->alias_list && !strcmp(alloc->alias_list, "TBD") &&
+ (slurm_allocation_lookup_lite(pending_job_id, &resp)
+ == SLURM_SUCCESS)) {
+ tmp_str = alloc->alias_list;
+ alloc->alias_list = resp->alias_list;
+ resp->alias_list = tmp_str;
+ slurm_free_resource_allocation_response_msg(resp);
+ }
} else if (!allocation_interrupted)
error("Nodes %s are still not ready", alloc->node_list);
else /* allocation_interrupted or slurmctld not responing */
diff --git a/src/slurmctld/agent.c b/src/slurmctld/agent.c
index 48f517f520df7c592a4849a6c21701ab114fc40f..e13c64ef3d54b54d37f14c52c1f3404312b48042 100644
--- a/src/slurmctld/agent.c
+++ b/src/slurmctld/agent.c
@@ -1522,7 +1522,7 @@ static int _batch_launch_defer(queued_request_t *queued_req_ptr)
batch_job_launch_msg_t *launch_msg_ptr;
time_t now = time(NULL);
struct job_record *job_ptr;
- int delay_time, nodes_ready = 0;
+ int delay_time, nodes_ready = 0, tmp;
agent_arg_ptr = queued_req_ptr->agent_arg_ptr;
if (agent_arg_ptr->msg_type != REQUEST_BATCH_JOB_LAUNCH)
@@ -1544,7 +1544,19 @@ static int _batch_launch_defer(queued_request_t *queued_req_ptr)
}
if (job_ptr->wait_all_nodes) {
- (void) job_node_ready(launch_msg_ptr->job_id, &nodes_ready);
+ (void) job_node_ready(launch_msg_ptr->job_id, &tmp);
+ if (tmp == (READY_JOB_STATE | READY_NODE_STATE)) {
+ nodes_ready = 1;
+ if (launch_msg_ptr->alias_list &&
+ !strcmp(launch_msg_ptr->alias_list, "TBD")) {
+ struct job_record *job_ptr;
+ job_ptr = find_job_record(launch_msg_ptr->
+ job_id);
+ xfree(launch_msg_ptr->alias_list);
+ launch_msg_ptr->alias_list = xstrdup(job_ptr->
+ alias_list);
+ }
+ }
} else {
#ifdef HAVE_FRONT_END
nodes_ready = 1;
diff --git a/src/slurmctld/job_mgr.c b/src/slurmctld/job_mgr.c
index 4f080efa12c75bb5dfb324fb206e27632daa1a13..51dc7ba718078c62cc08f60533d042ba086fdf4d 100644
--- a/src/slurmctld/job_mgr.c
+++ b/src/slurmctld/job_mgr.c
@@ -8530,6 +8530,12 @@ job_alloc_info(uint32_t uid, uint32_t job_id, struct job_record **job_pptr)
if (IS_JOB_FINISHED(job_ptr))
return ESLURM_ALREADY_DONE;
+ if (job_ptr->alias_list && !strcmp(job_ptr->alias_list, "TBD") &&
+ job_ptr->node_bitmap &&
+ (bit_overlap(power_node_bitmap, job_ptr->node_bitmap) == 0)) {
+ set_job_alias_list(job_ptr);
+ }
+
*job_pptr = job_ptr;
return SLURM_SUCCESS;
}
@@ -8992,18 +8998,23 @@ extern int job_node_ready(uint32_t job_id, int *ready)
/* Always call select_g_job_ready() so that select/bluegene can
* test and update block state information. */
rc = select_g_job_ready(job_ptr);
-
if (rc == READY_JOB_FATAL)
return ESLURM_INVALID_PARTITION_NAME;
if (rc == READY_JOB_ERROR)
return EAGAIN;
-
if (rc)
rc = READY_NODE_STATE;
if (IS_JOB_RUNNING(job_ptr) || IS_JOB_SUSPENDED(job_ptr))
rc |= READY_JOB_STATE;
+ if ((rc == (READY_NODE_STATE | READY_JOB_STATE)) &&
+ job_ptr->alias_list && !strcmp(job_ptr->alias_list, "TBD") &&
+ job_ptr->node_bitmap &&
+ (bit_overlap(power_node_bitmap, job_ptr->node_bitmap) == 0)) {
+ set_job_alias_list(job_ptr);
+ }
+
*ready = rc;
return SLURM_SUCCESS;
}
diff --git a/src/slurmctld/node_scheduler.c b/src/slurmctld/node_scheduler.c
index 464ff0375d12b2926054bc7d7b26e16f9ab2d687..a9517f2f1d4220988b25c0d8ffd2785b2ad55dfe 100644
--- a/src/slurmctld/node_scheduler.c
+++ b/src/slurmctld/node_scheduler.c
@@ -114,6 +114,7 @@ static int _pick_best_nodes(struct node_set *node_set_ptr,
uint32_t req_nodes, bool test_only,
List preemptee_candidates,
List *preemptee_job_list);
+static void _set_alias_list(struct job_record *job_ptr);
static bool _valid_feature_counts(struct job_details *detail_ptr,
bitstr_t *node_bitmap, bool *has_xor);
static bitstr_t *_valid_features(struct job_details *detail_ptr,
@@ -130,6 +131,7 @@ extern void allocate_nodes(struct job_record *job_ptr)
{
int i;
struct node_record *node_ptr;
+ bool has_cloud = false, has_cloud_power_save = false;
#ifdef HAVE_FRONT_END
job_ptr->front_end_ptr = assign_front_end();
@@ -138,12 +140,52 @@ extern void allocate_nodes(struct job_record *job_ptr)
job_ptr->batch_host = xstrdup(job_ptr->front_end_ptr->name);
#endif
+ for (i = 0, node_ptr = node_record_table_ptr; i < node_record_count;
+ i++, node_ptr++) {
+ if (!bit_test(job_ptr->node_bitmap, i))
+ continue;
+
+ if (IS_NODE_CLOUD(node_ptr)) {
+ has_cloud = true;
+ if (IS_NODE_POWER_SAVE(node_ptr))
+ has_cloud_power_save = true;
+ }
+ make_node_alloc(&node_record_table_ptr[i], job_ptr);
+ if (job_ptr->batch_host == NULL)
+ job_ptr->batch_host = xstrdup(node_ptr->name);
+ }
+ last_node_update = time(NULL);
+ license_job_get(job_ptr);
+
+ if (has_cloud) {
+ if (has_cloud_power_save) {
+ job_ptr->alias_list = xstrdup("TBD");
+ job_ptr->wait_all_nodes = 1;
+ } else
+ set_job_alias_list(job_ptr);
+ }
+
+ return;
+}
+
+/* Set a job's alias_list string */
+extern void set_job_alias_list(struct job_record *job_ptr)
+{
+ int i;
+ struct node_record *node_ptr;
+
xfree(job_ptr->alias_list);
for (i = 0, node_ptr = node_record_table_ptr; i < node_record_count;
i++, node_ptr++) {
if (!bit_test(job_ptr->node_bitmap, i))
continue;
+
if (IS_NODE_CLOUD(node_ptr)) {
+ if (IS_NODE_POWER_SAVE(node_ptr)) {
+ xfree(job_ptr->alias_list);
+ job_ptr->alias_list = xstrdup("TBD");
+ break;
+ }
if (job_ptr->alias_list)
xstrcat(job_ptr->alias_list, ",");
xstrcat(job_ptr->alias_list, node_ptr->name);
@@ -152,17 +194,9 @@ extern void allocate_nodes(struct job_record *job_ptr)
xstrcat(job_ptr->alias_list, ":");
xstrcat(job_ptr->alias_list, node_ptr->node_hostname);
}
- make_node_alloc(&node_record_table_ptr[i], job_ptr);
- if (job_ptr->batch_host)
- continue;
- job_ptr->batch_host = xstrdup(node_ptr->name);
}
- last_node_update = time(NULL);
- license_job_get(job_ptr);
- return;
}
-
/*
* deallocate_nodes - for a given job, deallocate its nodes and make
* their state NODE_STATE_COMPLETING also release the job's licenses
diff --git a/src/slurmctld/slurmctld.h b/src/slurmctld/slurmctld.h
index e1d707db75b6426cbe819db7490a5e4fc03bfba6..839c3f6b826126da7765650780fc4469350e889d 100644
--- a/src/slurmctld/slurmctld.h
+++ b/src/slurmctld/slurmctld.h
@@ -1538,6 +1538,9 @@ extern int send_jobs_to_accounting();
*/
extern int send_nodes_to_accounting(time_t event_time);
+/* Set a job's alias_list string */
+extern void set_job_alias_list(struct job_record *job_ptr);
+
/*
* set_job_prio - set a default job priority
* IN job_ptr - pointer to the job_record