- Change a few calls from exit() to orte_abort() so that we get

session directory cleanup (among other things)
- When we get an abnormal exit in orterun (i.e., timeout expires and
  we haven't gotten termination notices from all processes), print a
  better message an exit in a better way (which includes session
  directory cleanup)
- Fix tm and poe pls's to not exit() but rather propagate the error up
  the stack (where relevant)

This commit was SVN r7058.

orte/mca/errmgr/base/errmgr_base_fns.c

@@ -75,7 +75,7 @@ void orte_errmgr_base_abort()
     orte_wait_kill(9);
 
     /* abnormal exit */
-    exit(-1);
+    orte_abort(-1, NULL);
 }
 
 int orte_errmgr_base_register_job(orte_jobid_t job)

orte/mca/pls/poe/pls_poe_module.c

@@ -224,7 +224,7 @@ int pls_poe_launch_interactive_orted(orte_jobid_t jobid)
         rc = orte_ns.get_proc_name_string(&name_string, name);
         if(ORTE_SUCCESS != rc) {
             opal_output(0, "orte_pls_poe: unable to create process name");
-            exit(-1);
+            return rc;
         }
         argv[proc_name_index] = name_string;
         for(i=0;i<argc;i++) {

orte/mca/pls/tm/pls_tm_module.c

@@ -239,7 +239,7 @@ pls_tm_launch(orte_jobid_t jobid)
         rc = orte_ns.get_proc_name_string(&name_string, name);
         if (ORTE_SUCCESS != rc) {
             opal_output(0, "pls:tm: unable to create process name");
-            exit(-1);
+            return rc;
         }
         argv[proc_name_index] = name_string;
 

orte/tools/orteprobe/orteprobe.c

@@ -322,13 +322,13 @@ int main(int argc, char *argv[])
 
         if (ORTE_SUCCESS != (ret = orte_dps.pack(&buffer, &orted_uri_ptr, 1, ORTE_STRING))) {
             fprintf(stderr, "orteprobe: failed to pack contact info for existing universe\n");
-            exit(1);
+            orte_abort(1, NULL);
         }
 
         if (0 > orte_rml.send_buffer(&requestor, &buffer, ORTE_RML_TAG_PROBE, 0)) {
             fprintf(stderr, "orteprobe: comm failure when sending contact info for existing univ back to requestor\n");
             OBJ_DESTRUCT(&buffer);
-            exit(1);
+            orte_abort(1, NULL);
         }
 
         OBJ_DESTRUCT(&buffer);
@@ -355,7 +355,7 @@ int main(int argc, char *argv[])
 
             if (0 > asprintf(&orte_universe_info.name, "%s-%d", universe, pid)) {
                 fprintf(stderr, "orteprobe: failed to create unique universe name");
-                exit(1);
+                orte_abort(1, NULL);
             }
         }
 
@@ -363,7 +363,7 @@ int main(int argc, char *argv[])
         /* setup the pipe to get the contact info back */
         if (pipe(orted_pipe)) {
             fprintf (stderr, "orteprobe: Pipe failed\n");
-            exit(1);
+            orte_abort(1, NULL);
         }
         
         /* get name of orted application - just in case user specified something different */
@@ -378,7 +378,7 @@ int main(int argc, char *argv[])
         ortedargc = opal_argv_count(ortedargv);
         if (ortedargc <= 0) {
             fprintf(stderr, "orteprobe: could not initialize argv array for daemon\n");
-            exit(1);
+            orte_abort(1, NULL);
         }
         
         /* setup the path */
@@ -415,11 +415,11 @@ int main(int argc, char *argv[])
                 Close read end first. */
             execv(path, ortedargv);
             fprintf(stderr, "orteprobe: execv failed with errno=%d\n", errno);
-            exit(1);
+            orte_abort(1, NULL);
         } else if (pid < (pid_t) 0) {
             /* The fork failed. */
             fprintf (stderr, "orteprobe: Fork failed\n");
-            exit(1);
+            orte_abort(1, NULL);
         } else {
             /* This is the parent process.
                 Close write end first. */
@@ -437,13 +437,13 @@ int main(int argc, char *argv[])
 
             if (ORTE_SUCCESS != (ret = orte_dps.pack(&buffer, &orted_uri_ptr[0], 1, ORTE_STRING))) {
                 fprintf(stderr, "orteprobe: failed to pack daemon uri\n");
-                exit(1);
+                orte_abort(1, NULL);
             }
 
             if (0 > orte_rml.send_buffer(&requestor, &buffer, ORTE_RML_TAG_PROBE, 0)) {
                 fprintf(stderr, "orteprobe: could not send daemon uri info back to probe\n");
                 OBJ_DESTRUCT(&buffer);
-                exit(1);
+                orte_abort(1, NULL);
             }
 
             OBJ_DESTRUCT(&buffer);

orte/tools/orterun/help-orterun.txt

@@ -81,4 +81,8 @@ in the environment. Returned value %d instead of ORTE_SUCCESS.
 [orterun:proc-aborted]
 %s noticed that job rank %lu with PID %lu on node "%s" exited on signal %d.
 [orterun:abnormal-exit]
-%s encountered an abnormal exit.
+WARNING: %s encountered an abnormal exit.
+
+This means that %s exited before it received notification that all
+started processes had terminated.  You should double check and ensure
+that there are no runaway processes still executing.

orte/tools/orterun/orterun.c

@@ -565,7 +565,12 @@ static void exit_callback(int fd, short event, void *arg)
 {
     opal_show_help("help-orterun.txt", "orterun:abnormal-exit",
                    true, orterun_basename);
-    exit(1);
+
+    /* Trigger the normal exit conditions */
+
+    orterun_globals.exit = true;
+    orterun_globals.exit_status = 1;
+    opal_condition_signal(&orterun_globals.cond);
 }