svn merge -r9906:9908 https://eris.llnl.gov/svn/slurm/branches/slurm-1.1

NEWS

@@ -87,6 +87,8 @@ documents those changes that are of interest to users and admins.
  - In sched/wiki2, add support for EHost and EHostBackup configuration 
    parameters in wiki.conf file
  - In sched/wiki2, fix memory management bug for JOBWILLRUN command.
+ - In sched/wiki2, consider job Busy while in Completing state for 
+   KillWait+10 seconds (used to be 30 seconds).
 
 * Changes in SLURM 1.1.17
 =========================

src/plugins/sched/wiki2/event.c

@@ -68,12 +68,6 @@ extern int	event_notify(char *msg)
 	pthread_mutex_lock(&event_mutex);
 	if (event_addr_set == 0) {
 		/* Identify address for socket connection */
-		if (e_host[0] == '\0') {
-			slurm_ctl_conf_t *conf = slurm_conf_lock();
-			strncpy(e_host, conf->control_addr, 
-				sizeof(e_host));
-			slurm_conf_unlock();
-		}
 		slurm_set_addr(&moab_event_addr, e_port, e_host);
 		event_addr_set = 1;
 		if (e_host_bu[0] != '\0') {

src/plugins/sched/wiki2/get_jobs.c

@@ -319,15 +319,13 @@ static char *	_get_job_state(struct job_record *job_ptr)
 		return "Running";
 
 	if (state & JOB_COMPLETING) {
-		/* Give 60 seconds to clear out, then 
-		 * then consider job done. Let Moab 
-		 * deal with inconsistency between 
-		 * job state (DONE) and node state
-		 * (some IDLE and others still 
-		 * BUSY). */
+		/* Give configured KillWait+10 for job
+		 * to clear out, then then consider job 
+		 * done. Moab will allocate jobs to 
+		 * nodes that are already Idle. */ 
 		int age = (int) difftime(time(NULL), 
 			job_ptr->end_time);
-		if (age < 60)
+		if (age < (kill_wait+10))
 			return "Running";
 	}
 

src/plugins/sched/wiki2/msg.c

@@ -53,6 +53,7 @@ char     e_host_bu[E_HOST_SIZE] = "";
 uint16_t e_port = 0;
 uint16_t job_aggregation_time = 10;	/* Default value is 10 seconds */
 int      init_prio_mode = PRIO_HOLD;
+uint16_t kill_wait;
 uint16_t use_host_exp = 0;
 
 static char *	_get_wiki_conf_path(void);
@@ -205,6 +206,13 @@ static void _parse_wiki_config(void)
 	s_p_hashtbl_t *tbl;
 	char *key = NULL, *priority_mode = NULL, *wiki_conf;
 	struct stat buf;
+	slurm_ctl_conf_t *conf;
+
+	/* Set default values */
+	conf = slurm_conf_lock();
+	strncpy(e_host, conf->control_addr, sizeof(e_host));
+	kill_wait = conf->kill_wait;
+	slurm_conf_unlock();
 
 	wiki_conf = _get_wiki_conf_path();
 	if ((wiki_conf == NULL) || (stat(wiki_conf, &buf) == -1)) {
@@ -227,7 +235,8 @@ static void _parse_wiki_config(void)
 	if ( s_p_get_string(&key, "EHost", tbl)) {
 		strncpy(e_host, key, sizeof(e_host));
 		xfree(key);
-	}
+	} else
+		debug("wiki: Using ControlAddr for EHost value");
 	if ( s_p_get_string(&key, "EHostBackup", tbl)) {
 		strncpy(e_host_bu, key, sizeof(e_host_bu));
 		xfree(key);
@@ -247,6 +256,15 @@ static void _parse_wiki_config(void)
 	s_p_hashtbl_destroy(tbl);
 	xfree(wiki_conf);
 
+#if 0
+	info("AuthKey            = %s", auth_key);
+	info("EHost              = %s", e_host);
+	info("EHostBackup        = %s", e_host_bu);
+	info("EPort              = %u", e_port);
+	info("JobAggregationTime = %u sec", job_aggregation_time);
+	info("JobPriority        = %s", init_prio_mode ? "run" : "hold");
+	info("KillWait           = %u sec", kill_wait);      
+#endif
 	return;
 }
 

src/plugins/sched/wiki2/msg.h

@@ -91,6 +91,7 @@ extern char	e_host[E_HOST_SIZE];
 extern char	e_host_bu[E_HOST_SIZE];
 extern uint16_t	e_port;
 extern uint16_t	job_aggregation_time;
+extern uint16_t kill_wait;
 extern uint16_t use_host_exp;
 
 extern int	event_notify(char *msg);