Enforce QOS MaxCPUsMin limit when job has no time limit

bug 436

NEWS

@@ -18,6 +18,8 @@ documents those changes that are of interest to users and admins.
  -- switch/nrt - Don't allocate network resources unless job step has 2+ nodes.
  -- select/cons_res - Avoid extraneous "oversubscribe" error messages.
  -- Reorder get config logic to avoid deadlock.
+ -- Enforce QOS MaxCPUsMin limit when job submission contains no user-specified
+    time limit.
 
 * Changes in Slurm 2.5.7
 ========================

src/slurmctld/acct_policy.c

@@ -441,9 +441,8 @@ extern bool acct_policy_validate(job_desc_msg_t *job_desc,
 	bool rc = true;
 	uint32_t qos_max_cpus_limit = INFINITE;
 	uint32_t qos_max_nodes_limit = INFINITE;
+	uint32_t qos_time_limit = INFINITE;
 	uint32_t job_memory = 0;
-	uint64_t cpu_time_limit;
-	uint64_t job_cpu_time_limit;
 	bool admin_set_memory_limit = false;
 	assoc_mgr_lock_t locks = { READ_LOCK, NO_LOCK,
 				   READ_LOCK, NO_LOCK, NO_LOCK };
@@ -647,23 +646,14 @@ extern bool acct_policy_validate(job_desc_msg_t *job_desc,
 		 * if you can end up in PENDING QOSJobLimit, you need
 		 * to validate it if DenyOnLimit is set
 		 */
-		if (strict_checking && (qos_ptr->max_cpu_mins_pj != INFINITE)
-		    && (job_desc->time_limit != NO_VAL)
-		    && (job_desc->min_cpus != NO_VAL)) {
-			cpu_time_limit = qos_ptr->max_cpu_mins_pj;
-			job_cpu_time_limit = (uint64_t)job_desc->time_limit
-				* (uint64_t)job_desc->min_cpus;
-			if (job_cpu_time_limit > cpu_time_limit) {
-				if (reason)
-					*reason = WAIT_QOS_JOB_LIMIT;
-				debug2("job submit for user %s(%u): "
-				       "cpu time limit %"PRIu64" exceeds "
-				       "qos max per-job %"PRIu64"",
-				       user_name, job_desc->user_id,
-				       job_cpu_time_limit, cpu_time_limit);
-				rc = false;
-				goto end_it;
-			}
+		if (((job_desc->min_cpus  != NO_VAL) ||
+		     (job_desc->min_nodes != NO_VAL)) &&
+		    (qos_ptr->max_cpu_mins_pj != INFINITE)) {
+			uint32_t cpu_cnt = job_desc->min_nodes;
+			if ((job_desc->min_nodes == NO_VAL) ||
+			    (job_desc->min_cpus > job_desc->min_nodes))
+				cpu_cnt = job_desc->min_cpus;
+			qos_time_limit = qos_ptr->max_cpu_mins_pj / cpu_cnt;
 		}
 
 		if ((acct_policy_limit_set->max_cpus == ADMIN_SET_LIMIT)
@@ -770,33 +760,36 @@ extern bool acct_policy_validate(job_desc_msg_t *job_desc,
 		    || (qos_ptr->max_wall_pj == INFINITE)
 		    || (update_call && (job_desc->time_limit == NO_VAL))) {
 			/* no need to check/set */
-		} else {
-			time_limit = qos_ptr->max_wall_pj;
+		} else if (qos_time_limit > qos_ptr->max_wall_pj) {
+			qos_time_limit = qos_ptr->max_wall_pj;
+		}
+
+		if (qos_time_limit != INFINITE) {
 			if (job_desc->time_limit == NO_VAL) {
 				if (part_ptr->max_time == INFINITE)
-					job_desc->time_limit = time_limit;
-				else
+					job_desc->time_limit = qos_time_limit;
+				else {
 					job_desc->time_limit =
-						MIN(time_limit,
+						MIN(qos_time_limit,
 						    part_ptr->max_time);
+				}
 				acct_policy_limit_set->time = 1;
 			} else if (acct_policy_limit_set->time &&
-				   job_desc->time_limit > time_limit) {
-				job_desc->time_limit = time_limit;
+				   job_desc->time_limit > qos_time_limit) {
+				job_desc->time_limit = qos_time_limit;
 			} else if (strict_checking
-				   && job_desc->time_limit > time_limit) {
+				   && job_desc->time_limit > qos_time_limit) {
 				if (reason)
 					*reason = WAIT_QOS_JOB_LIMIT;
 				debug2("job submit for user %s(%u): "
 				       "time limit %u exceeds qos max %u",
 				       user_name,
 				       job_desc->user_id,
-				       job_desc->time_limit, time_limit);
+				       job_desc->time_limit, qos_time_limit);
 				rc = false;
 				goto end_it;
 			}
 		}
-
 	}
 
 	while (assoc_ptr) {