Ignore comm failures when shutting down orteds

This commit was SVN r23201.

orte/mca/errmgr/hnp/errmgr_hnp.c

@@ -369,8 +369,8 @@ static int update_state(orte_jobid_t job,
         case ORTE_PROC_STATE_COMM_FAILED:
             /* is this to a daemon? */
             if (ORTE_PROC_MY_NAME->jobid == proc->jobid) {
-                /* if we have ordered jobs to terminate, ignore this */
-                if (orte_job_term_ordered) {
+                /* if we have ordered orteds to terminate, ignore this */
+                if (orte_orteds_term_ordered) {
                     break;
                 }
                 /* if this is my own connection, ignore it */

orte/mca/plm/base/plm_base_orted_cmds.c

@@ -103,12 +103,8 @@ int orte_plm_base_orted_exit(orte_daemon_cmd_flag_t command)
 
     OBJ_CONSTRUCT(&cmd, opal_buffer_t);
     
-    /* since the orteds are being ordered to exit, and we are
-     * technically a member of that "job", first update our
-     * information to indicate we are "terminated". This will
-     * ensure that we can exit correctly when all daemons report
-     * in as "terminated"
-     */
+    /* flag that orteds are being terminated */
+    orte_orteds_term_ordered = true;
     
     /* get the job object for the daemons */
     if (NULL == (daemons = orte_get_job_data_object(ORTE_PROC_MY_NAME->jobid))) {

orte/runtime/orte_globals.c

@@ -97,6 +97,7 @@ int orte_exit_status = 0;
 bool orte_abnormal_term_ordered = false;
 bool orte_routing_is_enabled = false;
 bool orte_job_term_ordered = false;
+bool orte_orteds_term_ordered = false;
 
 int orte_startup_timeout;
 

orte/runtime/orte_globals.h

@@ -592,6 +592,7 @@ ORTE_DECLSPEC extern int orte_exit_status;
 ORTE_DECLSPEC extern bool orte_abnormal_term_ordered;
 ORTE_DECLSPEC extern bool orte_routing_is_enabled;
 ORTE_DECLSPEC extern bool orte_job_term_ordered;
+ORTE_DECLSPEC extern bool orte_orteds_term_ordered;
 
 ORTE_DECLSPEC extern int orte_startup_timeout;