Correctly detect and handle oversubscription for comm_spawn

cmr=v1.7.4:reviewer=jsquyres:subject=Correctly detect and handle oversubscription for comm_spawn

This commit was SVN r30186.

orte/mca/rmaps/base/rmaps_base_support_fns.c

@@ -490,22 +490,13 @@ int orte_rmaps_base_get_target_nodes(opal_list_t *allocated_nodes, orte_std_cntr
                                  node->name, node->slots, node->slots_inuse));
             opal_list_remove_item(allocated_nodes, item);
             OBJ_RELEASE(item);  /* "un-retain" it */
-        } else {
+        } else if (node->slots > node->slots_inuse) {
-            if (node->slots > node->slots_inuse) {
                 /* add the available slots */
                 OPAL_OUTPUT_VERBOSE((5, orte_rmaps_base_framework.framework_output,
                                      "%s node %s has %d slots available",
                                      ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
-                                     node->name, node->slots));
+                                     node->name, node->slots - node->slots_inuse));
                 num_slots += node->slots - node->slots_inuse;
-            } else {
-                /* always allocate at least one */
-                OPAL_OUTPUT_VERBOSE((5, orte_rmaps_base_framework.framework_output,
-                                     "%s node %s has %d slots %d used",
-                                     ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
-                                     node->name, node->slots, node->slots_inuse));
-                num_slots++;
-            }
         }
 
         /** go on to next item */

orte/mca/rmaps/round_robin/rmaps_rr_mappers.c

@@ -98,6 +98,7 @@ int orte_rmaps_rr_byslot(orte_job_t *jdata,
         opal_output_verbose(2, orte_rmaps_base_framework.framework_output,
                             "mca:rmaps:rr:slot assigning %d procs to node %s",
                             (int)num_procs_to_assign, node->name);
+
         for (i=0; i < num_procs_to_assign && nprocs_mapped < app->num_procs; i++) {
             /* add this node to the map - do it only once */
             if (!node->mapped) {
@@ -507,16 +508,6 @@ int orte_rmaps_rr_byobj(orte_job_t *jdata,
                            true, node->name);
             return ORTE_ERR_SILENT;
         }
-        /* add this node to the map, if reqd */
-        if (!node->mapped) {
-            if (ORTE_SUCCESS > (idx = opal_pointer_array_add(jdata->map->nodes, (void*)node))) {
-                ORTE_ERROR_LOG(idx);
-                return idx;
-            }
-            node->mapped = true;
-            OBJ_RETAIN(node);  /* maintain accounting on object */
-            ++(jdata->map->num_nodes);
-        }
 
         /* compute the number of procs to go on this node */
         if (add_one) {
@@ -538,6 +529,29 @@ int orte_rmaps_rr_byobj(orte_job_t *jdata,
             }
         }
 
+        /* if this would oversubscribe the node and the user hasn't permitted
+         * oversubscription, then don't use it - since the total number of
+         * slots is adequate for this app, there should be room somewhere else
+         */
+        if (node->slots < (node->slots_inuse + num_procs_to_assign) &&
+            ORTE_MAPPING_NO_OVERSUBSCRIBE & ORTE_GET_MAPPING_DIRECTIVE(jdata->map->mapping)) {
+            opal_output_verbose(2, orte_rmaps_base_framework.framework_output,
+                                "mca:rmaps:rr: mapping no-span would oversubscribe node %s - ignoring it",
+                                node->name);
+            continue;
+        }
+
+        /* add this node to the map, if reqd */
+        if (!node->mapped) {
+            if (ORTE_SUCCESS > (idx = opal_pointer_array_add(jdata->map->nodes, (void*)node))) {
+                ORTE_ERROR_LOG(idx);
+                return idx;
+            }
+            node->mapped = true;
+            OBJ_RETAIN(node);  /* maintain accounting on object */
+            ++(jdata->map->num_nodes);
+        }
+
         /* get the number of objects of this type on this node */
         nobjs = opal_hwloc_base_get_nbobjs_by_type(node->topology, target, cache_level, OPAL_HWLOC_AVAILABLE);
         opal_output_verbose(2, orte_rmaps_base_framework.framework_output,