Merge remote-tracking branch 'origin/slurm-2.5'

NEWS

@@ -29,6 +29,7 @@ documents those changes that are of interest to users and admins.
 * Changes in SLURM 2.5.2
 ========================
  -- Fix advanced reservation recovery logic when upgrading from version 2.4.
+ -- BLUEGENE - fix for QOS/Association node limits.
 
 * Changes in SLURM 2.5.1
 ========================

src/slurmctld/acct_policy.c

@@ -47,6 +47,7 @@
 
 #include "src/slurmctld/slurmctld.h"
 #include "src/slurmctld/acct_policy.h"
+#include "src/common/node_select.h"
 
 #define _DEBUG 0
 
@@ -131,10 +132,23 @@ static void _adjust_limit_usage(int type, struct job_record *job_ptr)
 	uint64_t unused_cpu_run_secs = 0;
 	uint64_t used_cpu_run_secs = 0;
 	uint32_t job_memory = 0;
+	uint32_t node_cnt;
 
 	if (!(accounting_enforce & ACCOUNTING_ENFORCE_LIMITS)
 	    || !_valid_job_assoc(job_ptr))
 		return;
+#ifdef HAVE_BG
+	xassert(job_ptr->select_jobinfo);
+	select_g_select_jobinfo_get(job_ptr->select_jobinfo,
+				    SELECT_JOBDATA_NODE_CNT, &node_cnt);
+	if (node_cnt == NO_VAL) {
+		/* This should never happen */
+		node_cnt = job_ptr->node_cnt;
+		error("node_cnt not available at %s:%d\n", __FILE__, __LINE__);
+	}
+#else
+	node_cnt = job_ptr->node_cnt;
+#endif
 
 	if (type == ACCT_POLICY_JOB_FINI)
 		unused_cpu_run_secs = _get_unused_cpu_run_secs(job_ptr);
@@ -152,7 +166,7 @@ static void _adjust_limit_usage(int type, struct job_record *job_ptr)
 			       job_memory);
 		} else {
 			job_memory = (job_ptr->details->pn_min_memory)
-				* job_ptr->node_cnt;
+				* node_cnt;
 			debug2("_adjust_limit_usage: job %u: MPN: "
 			       "job_memory set to %u", job_ptr->job_id,
 			       job_memory);
@@ -202,12 +216,12 @@ static void _adjust_limit_usage(int type, struct job_record *job_ptr)
 			qos_ptr->usage->grp_used_jobs++;
 			qos_ptr->usage->grp_used_cpus += job_ptr->total_cpus;
 			qos_ptr->usage->grp_used_mem += job_memory;
-			qos_ptr->usage->grp_used_nodes += job_ptr->node_cnt;
+			qos_ptr->usage->grp_used_nodes += node_cnt;
 			qos_ptr->usage->grp_used_cpu_run_secs +=
 				used_cpu_run_secs;
 			used_limits->jobs++;
 			used_limits->cpus += job_ptr->total_cpus;
-			used_limits->nodes += job_ptr->node_cnt;
+			used_limits->nodes += node_cnt;
 			break;
 		case ACCT_POLICY_JOB_FINI:
 
@@ -232,7 +246,7 @@ static void _adjust_limit_usage(int type, struct job_record *job_ptr)
 				       "underflow for qos %s", qos_ptr->name);
 			}
 
-			qos_ptr->usage->grp_used_nodes -= job_ptr->node_cnt;
+			qos_ptr->usage->grp_used_nodes -= node_cnt;
 			if ((int32_t)qos_ptr->usage->grp_used_nodes < 0) {
 				qos_ptr->usage->grp_used_nodes = 0;
 				debug2("acct_policy_job_fini: grp_used_nodes "
@@ -268,7 +282,7 @@ static void _adjust_limit_usage(int type, struct job_record *job_ptr)
 				       qos_ptr->name, used_limits->uid);
 			}
 
-			used_limits->nodes -= job_ptr->node_cnt;
+			used_limits->nodes -= node_cnt;
 			if ((int32_t)used_limits->nodes < 0) {
 				used_limits->nodes = 0;
 				debug2("acct_policy_job_fini: "
@@ -303,7 +317,7 @@ static void _adjust_limit_usage(int type, struct job_record *job_ptr)
 			assoc_ptr->usage->used_jobs++;
 			assoc_ptr->usage->grp_used_cpus += job_ptr->total_cpus;
 			assoc_ptr->usage->grp_used_mem += job_memory;
-			assoc_ptr->usage->grp_used_nodes += job_ptr->node_cnt;
+			assoc_ptr->usage->grp_used_nodes += node_cnt;
 			assoc_ptr->usage->grp_used_cpu_run_secs +=
 				used_cpu_run_secs;
 			debug4("acct_policy_job_begin: after adding job %i, "
@@ -335,7 +349,7 @@ static void _adjust_limit_usage(int type, struct job_record *job_ptr)
 				       assoc_ptr->acct);
 			}
 
-			assoc_ptr->usage->grp_used_nodes -= job_ptr->node_cnt;
+			assoc_ptr->usage->grp_used_nodes -= node_cnt;
 			if ((int32_t)assoc_ptr->usage->grp_used_nodes < 0) {
 				assoc_ptr->usage->grp_used_nodes = 0;
 				debug2("acct_policy_job_fini: grp_used_nodes "