Update the FTB notifier wrt events decided by the CIFTS working group

This commit was SVN r24001.

orte/mca/notifier/ftb/help-ftb-event-schema.txt

@@ -15,23 +15,15 @@ start
 
 ftb.mpi.openmpi
 
-MPI_INIT INFO
+FTB_MPI_PROCS_DEAD ERROR
-MPI_FINALIZE INFO
+FTB_MPI_PROCS_UNREACHABLE ERROR
-MPI_NODE_DEAD ERROR
+FTB_MPI_PROCS_COMM_ERROR WARN
-MPI_NODE_RESTORED INFO
+FTB_MPI_PROCS_MIGRATED INFO
-MPI_RANK_DEAD ERROR
+FTB_MPI_PROCS_MIGRATE_FAIL ERROR
-MPI_RANK_RESTORED INFO
+FTB_MPI_PROCS_CKPTED INFO
-MPI_NODE_MIGRATE_DONE INFO
+FTB_MPI_PROCS_CKPT_FAIL ERROR
-MPI_JOB_ABORT_CMD ERROR
+FTB_MPI_PROCS_RESTARTED INFO
-MPI_JOB_RESUME_CMD INFO
+FTB_MPI_PROCS_RESTART_FAIL ERROR
-MPI_JOB_ABORTED ERROR
+FTB_MPI_PROCS_ABORTED ERROR
-MPI_JOB_RESUMED INFO
-MPI_MSG_CORRUPT ERROR
-MPI_IFACE_DEAD ERROR
-MPI_IFACE_RESTORED ERROR
-MPI_UNKNOWN_ERROR ERROR
-MPI_OUT_OF_RESOURCE ERROR
-MPI_NODE_UNREACHABLE ERROR
-MPI_COMM_FAILURE ERROR
 
 end

orte/mca/notifier/ftb/notifier_ftb.h

@@ -53,8 +53,8 @@ typedef enum {
     FTB_EVENT_RESPONSE = 2
 } ftb_event_type_t;
 
-/* Macro that returns FTB event name given the FTB event code */
+/* Returns the FTB event name (as a string) given the event code */
-#define FTB_ERROR(errnum) #errnum
+#define FTB_EVENT(errnum) #errnum
 
 END_C_DECLS
 

orte/mca/notifier/ftb/notifier_ftb_module.c

@@ -119,30 +119,31 @@ static const char* get_ftb_event_severity(orte_notifier_base_severity_t severity
 static const char* get_ftb_event_name(int errnum)
 {
     /* If it an OMPI error, translate it to an equivalent FTB event */
-    if (OPAL_SUCCESS > errnum) {
+    if (ORTE_SUCCESS > errnum) {
         switch (errnum) {
-        case ORTE_ERR_OUT_OF_RESOURCE:
+
-        case ORTE_ERR_TEMP_OUT_OF_RESOURCE:
+        case ORTE_SNAPC_CKPT_STATE_ESTABLISHED:
-            return FTB_ERROR(MPI_OUT_OF_RESOURCE);
+        case ORTE_SNAPC_CKPT_STATE_RECOVERED:
+            return FTB_EVENT(FTB_MPI_PROCS_CKPTED);
+
+        case ORTE_SNAPC_CKPT_STATE_NO_CKPT:
+        case ORTE_SNAPC_CKPT_STATE_ERROR:
+            return FTB_EVENT(FTB_MPI_PROCS_CKPT_FAIL);
 
         case ORTE_ERR_CONNECTION_REFUSED:
         case ORTE_ERR_CONNECTION_FAILED:
         case ORTE_ERR_UNREACH:
-            return FTB_ERROR(MPI_NODE_DEAD);
+            return FTB_EVENT(FTB_MPI_PROCS_UNREACHABLE);
 
         case ORTE_ERR_COMM_FAILURE:
-            return FTB_ERROR(MPI_COMM_FAILURE);
+            return FTB_EVENT(FTB_MPI_PROCS_COMM_ERROR);
 
-        case ORTE_ERR_PROC_DEAD:
-            return FTB_ERROR(MPI_RANK_DEAD);
-
-        case ORTE_ERR_FATAL:
         default:
-            return FTB_ERROR(MPI_UNKNOWN_ERROR);
+            return NULL;
         }
     }
 
-    return FTB_ERROR(MPI_UNKNOWN_ERROR);
+    return NULL;
 }
 
 static void publish_ftb_event(orte_notifier_base_severity_t severity, int errcode, char *payload)
@@ -162,12 +163,14 @@ static void publish_ftb_event(orte_notifier_base_severity_t severity, int errcod
 
     /* Publish the event to the Fault Tolerant Backplane */
     event_name = get_ftb_event_name(errcode);
+    if (NULL != event_name) {
         ret = FTB_Publish(ftb_client_handle, event_name, &eprop, &ehandle);
         if (FTB_SUCCESS != ret) {
             orte_show_help("help-orte-notifier-ftb.txt", "publish failed", true,
                            "FTB_Publish() failed", ret, get_ftb_event_severity(severity),
                            event_name, payload, errcode);
         }
+    }
 }
 
 static void ftb_log(orte_notifier_base_severity_t severity, int errcode, const char *msg,
@@ -175,7 +178,6 @@ static void ftb_log(orte_notifier_base_severity_t severity, int errcode, const c
 {
     char *payload;
 
-    /* If there was a message, output it */
     vasprintf(&payload, msg, ap);
     if (NULL != payload) {
         publish_ftb_event(severity, errcode, payload);