Merge pull request #5838 from rhc54/topic/ev

Correctly notify upon process failure

opal/mca/pmix/pmix4x/pmix4x_server_south.c

@@ -516,6 +516,7 @@ int pmix4x_server_notify_event(int status,
     size_t sz, n;
     pmix_status_t rc;
     pmix4x_opcaddy_t *op;
+    pmix_data_range_t range = PMIX_RANGE_SESSION;
 
     OPAL_PMIX_ACQUIRE_THREAD(&opal_pmix_base.lock);
     if (0 >= opal_pmix_base.initialized) {
@@ -535,6 +536,9 @@ int pmix4x_server_notify_event(int status,
                 pinfo[n].value.data.status = pmix4x_convert_opalrc(kv->data.integer);
             } else {
                 pmix4x_value_load(&pinfo[n].value, kv);
+                if (0 == strcmp(kv->key, OPAL_PMIX_EVENT_CUSTOM_RANGE)) {
+                    range = PMIX_RANGE_CUSTOM;
+                }
             }
             ++n;
         }
@@ -561,7 +565,7 @@ int pmix4x_server_notify_event(int status,
     rc = pmix4x_convert_opalrc(status);
     /* the range must be nonlocal so the server will pass
      * the event down to its local clients */
-    rc = PMIx_Notify_event(rc, &op->p, PMIX_RANGE_SESSION,
+    rc = PMIx_Notify_event(rc, &op->p, range,
                            pinfo, sz, opcbfunc, op);
     if (PMIX_SUCCESS != rc) {
         OBJ_RELEASE(op);

orte/mca/state/base/state_base_fns.c

@@ -551,37 +551,18 @@ static void _send_notification(int status,
         return;
     }
 
-    if (OPAL_ERR_PROC_ABORTED == status) {
-        /* we will pass three opal_value_t's */
-        rc = 3;
-        if (ORTE_SUCCESS != (rc = opal_dss.pack(buf, &rc, 1, OPAL_INT))) {
-            ORTE_ERROR_LOG(rc);
-            OBJ_RELEASE(buf);
-            return;
-        }
-        /* pass along the affected proc(s) */
-        OBJ_CONSTRUCT(&kv, opal_value_t);
-        kv.key = strdup(OPAL_PMIX_EVENT_AFFECTED_PROC);
-        kv.type = OPAL_NAME;
-        kv.data.name.jobid = proc->jobid;
-        kv.data.name.vpid = proc->vpid;
-        kvptr = &kv;
-        if (ORTE_SUCCESS != (rc = opal_dss.pack(buf, &kvptr, 1, OPAL_VALUE))) {
-            ORTE_ERROR_LOG(rc);
-            OBJ_DESTRUCT(&kv);
-            OBJ_RELEASE(buf);
-            return;
-        }
-        OBJ_DESTRUCT(&kv);
+    if (ORTE_VPID_WILDCARD == target->vpid) {
+        /* we will only pass the affected proc */
+        rc = 1;
     } else {
-        /* we are going to pass two opal_value_t's */
+        /* we have to pass the target */
         rc = 2;
+    }
     if (ORTE_SUCCESS != (rc = opal_dss.pack(buf, &rc, 1, OPAL_INT))) {
         ORTE_ERROR_LOG(rc);
         OBJ_RELEASE(buf);
         return;
     }
-    }
 
     /* pass along the affected proc(s) */
     OBJ_CONSTRUCT(&kv, opal_value_t);
@@ -598,7 +579,21 @@ static void _send_notification(int status,
     }
     OBJ_DESTRUCT(&kv);
 
-    /* pass along the proc(s) to be notified */
+    if (ORTE_VPID_WILDCARD == target->vpid) {
+        /* xcast it to everyone */
+        OBJ_CONSTRUCT(&sig, orte_grpcomm_signature_t);
+        sig.signature = (orte_process_name_t*)malloc(sizeof(orte_process_name_t));
+        sig.signature[0].jobid = ORTE_PROC_MY_NAME->jobid;
+        sig.signature[0].vpid = ORTE_VPID_WILDCARD;
+        sig.sz = 1;
+
+        if (ORTE_SUCCESS != (rc = orte_grpcomm.xcast(&sig, ORTE_RML_TAG_NOTIFICATION, buf))) {
+            ORTE_ERROR_LOG(rc);
+        }
+        OBJ_DESTRUCT(&sig);
+        OBJ_RELEASE(buf);
+    } else {
+        /* pass along the proc to be notified */
         OBJ_CONSTRUCT(&kv, opal_value_t);
         kv.key = strdup(OPAL_PMIX_EVENT_CUSTOM_RANGE);
         kv.type = OPAL_NAME;
@@ -612,21 +607,6 @@ static void _send_notification(int status,
             return;
         }
         OBJ_DESTRUCT(&kv);
-
-    /* if the targets are a wildcard, then xcast it to everyone */
-    if (ORTE_VPID_WILDCARD == target->vpid) {
-        OBJ_CONSTRUCT(&sig, orte_grpcomm_signature_t);
-        sig.signature = (orte_process_name_t*)malloc(sizeof(orte_process_name_t));
-        sig.signature[0].jobid = ORTE_PROC_MY_NAME->jobid;
-        sig.signature[0].vpid = ORTE_VPID_WILDCARD;
-        sig.sz = 1;
-
-        if (ORTE_SUCCESS != (rc = orte_grpcomm.xcast(&sig, ORTE_RML_TAG_NOTIFICATION, buf))) {
-            ORTE_ERROR_LOG(rc);
-        }
-        OBJ_DESTRUCT(&sig);
-        OBJ_RELEASE(buf);
-    } else {
         /* get the daemon hosting the proc to be notified */
         daemon.jobid = ORTE_PROC_MY_NAME->jobid;
         daemon.vpid = orte_get_proc_daemon_vpid(target);