Some cleanups for staged execution

This commit was SVN r27317.

orte/mca/rmaps/staged/rmaps_staged.c

@@ -224,6 +224,11 @@ static int staged_mapper(orte_job_t *jdata)
 	    /* track number of procs on node and number of slots used */
             node->num_procs++;
             node->slots_inuse++;
+            opal_output_verbose(10, orte_rmaps_base.rmaps_output,
+                                "%s Proc %s on node %s: slots %d inuse %d",
+                                ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
+                                ORTE_NAME_PRINT(&proc->name), node->name,
+                                (int)node->slots, (int)node->slots_inuse);
             if (node->slots_inuse == node->slots) {
                 opal_list_remove_item(&node_list, &node->super);
                 OBJ_RELEASE(node);

orte/mca/state/orted/state_orted_component.c

@@ -57,7 +57,7 @@ orte_state_base_component_t mca_state_orted_component =
     },
 };
 
-static int my_priority=1000;
+static int my_priority=100;
 
 static int state_orted_open(void) 
 {

orte/mca/state/staged_hnp/state_staged_hnp.c

@@ -270,6 +270,11 @@ static void cleanup_node(orte_proc_t *proc)
     }
     node->num_procs--;
     node->slots_inuse--;
+    OPAL_OUTPUT_VERBOSE((5, orte_state_base_output,
+                         "%s state:staged_hnp:track_procs proc %s termed node %s has %d slots, %d slots inuse",
+                         ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
+                         ORTE_NAME_PRINT(&proc->name), node->name,
+                         (int)node->slots, (int)node->slots_inuse));
     for (i=0; i < node->procs->size; i++) {
         if (NULL == (p = (orte_proc_t*)opal_pointer_array_get_item(node->procs, i))) {
             continue;
@@ -281,10 +286,6 @@ static void cleanup_node(orte_proc_t *proc)
             break;
         }
     }
-    OPAL_OUTPUT_VERBOSE((5, orte_state_base_output,
-                         "%s state:staged_hnp:track_procs node %s has %d slots, %d slots inuse",
-                         ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), node->name,
-                         (int)node->slots, (int)node->slots_inuse));
 }
 
 static void track_procs(int fd, short args, void *cbdata)

orte/mca/state/staged_orted/state_staged_orted.c

@@ -164,10 +164,11 @@ static void track_procs(int fd, short argc, void *cbdata)
     orte_process_name_t *proc = &caddy->name;
     orte_proc_state_t state = caddy->proc_state;
     orte_job_t *jdata;
-    orte_proc_t *pdata;
+    orte_proc_t *pdata, *pptr;
     opal_buffer_t *alert;
     int rc;
     orte_plm_cmd_flag_t cmd;
+    int i;
 
     OPAL_OUTPUT_VERBOSE((5, orte_state_base_output,
                          "%s state:staged_orted:track_procs called for proc %s state %s",
@@ -238,7 +239,20 @@ static void track_procs(int fd, short argc, void *cbdata)
                                                   orte_rml_send_callback, NULL))) {
                 ORTE_ERROR_LOG(rc);
             }
-        }
+            /* find this proc in the children array and remove it so
+             * we don't keep telling the HNP that it died
+             */
+            for (i=0; i < orte_local_children->size; i++) {
+                if (NULL == (pptr = (orte_proc_t*)opal_pointer_array_get_item(orte_local_children, i))) {
+                    continue;
+                }
+                if (pptr == pdata) {
+                    opal_pointer_array_set_item(orte_local_children, i, NULL);
+                    OBJ_RELEASE(pdata);
+                    break;
+                }
+            }
+         }
         /* Release the stdin IOF file descriptor for this child, if one
          * was defined. File descriptors for the other IOF channels - stdout,
          * stderr, and stddiag - were released when their associated pipes
@@ -287,6 +301,19 @@ static void track_procs(int fd, short argc, void *cbdata)
                                                   orte_rml_send_callback, NULL))) {
                 ORTE_ERROR_LOG(rc);
             }
+            /* find this proc in the children array and remove it so
+             * we don't keep telling the HNP that it died
+             */
+            for (i=0; i < orte_local_children->size; i++) {
+                if (NULL == (pptr = (orte_proc_t*)opal_pointer_array_get_item(orte_local_children, i))) {
+                    continue;
+                }
+                if (pptr == pdata) {
+                    opal_pointer_array_set_item(orte_local_children, i, NULL);
+                    OBJ_RELEASE(pdata);
+                    break;
+                }
+            }
         }
         break;