Clean up a couple of minor typos. Bring the new bproc-related RAS components online.

This commit was SVN r15328.

orte/mca/ras/bjs/ras_bjs_component.c

@@ -110,7 +110,6 @@ static orte_ras_base_module_t *orte_ras_bjs_init(int* priority)
         return NULL;
     }
     
-#if 0
     /* see if bjs is running */
     if (getenv("BJS_SOCKET") == NULL) {
         return NULL;
@@ -122,7 +121,6 @@ static orte_ras_base_module_t *orte_ras_bjs_init(int* priority)
     if (getenv("NODES") == NULL) {
         return NULL;
     }
-#endif
     
     *priority = mca_ras_bjs_component.priority;
     return &orte_ras_bjs_module;

orte/mca/ras/bproc_raw/.ompi_unignore

@@ -1,2 +0,0 @@
-rhc
-jsquyres

orte/mca/ras/bproc_raw/ras_bproc_raw.c

@@ -91,7 +91,7 @@ static int orte_ras_bproc_raw_allocate(orte_jobid_t jobid, opal_list_t *attribut
     }
     
     /* if no nodes available, let the user know and return error */
-    if (0 == ns->size) {
+    if (0 == ns.size) {
         opal_show_help("help-ras-broc-raw.txt", "no-nodes-found", true);
         return ORTE_ERR_NOT_AVAILABLE;
     }
@@ -100,8 +100,8 @@ static int orte_ras_bproc_raw_allocate(orte_jobid_t jobid, opal_list_t *attribut
     OBJ_CONSTRUCT(&nodes, opal_list_t);
 
     /* cycle through the list */
-    for (i=0; i < ns->size; i++) {
-        ni = &ns->node[i];
+    for (i=0; i < ns.size; i++) {
+        ni = &ns.node[i];
         
         /* check that the node is alive */
         if(orte_ras_bproc_raw_node_state(ni->node) != ORTE_NODE_STATE_UP) {
@@ -118,7 +118,15 @@ static int orte_ras_bproc_raw_allocate(orte_jobid_t jobid, opal_list_t *attribut
         /* okay, we have access and it is alive - create a new node entry */
         node = OBJ_NEW(orte_ras_node_t);
         asprintf(&node->node_name, "%d", ni->node);
-        node->node_state = ni->status;
+        if (strcmp(ni->status, "up") == 0) {
+            node->node_state = ORTE_NODE_STATE_UP;
+        } else if (strcmp(ni->status, "down") == 0) {
+            node->node_state = ORTE_NODE_STATE_DOWN;
+        } else if (strcmp(ni->status, "boot") == 0) {
+            node->node_state = ORTE_NODE_STATE_REBOOT;
+        } else {
+            node->node_state = ORTE_NODE_STATE_UNKNOWN;
+        }
         /* RHC - until we can find some way of querying bproc for the number of
          * available slots, just assume two
          */

orte/mca/ras/bproc_raw/ras_bproc_raw_component.c

@@ -18,10 +18,14 @@
 
 #include "orte_config.h"
 #include "orte/orte_constants.h"
+
+#include <sys/bproc.h>
+
 #include "opal/mca/base/base.h"
 #include "opal/mca/base/mca_base_param.h"
 #include "orte/util/proc_info.h"
 #include "opal/util/output.h"
+
 #include "ras_bproc_raw.h"
 
 /*

orte/mca/ras/lanl_bproc/.ompi_unignore

@@ -1,2 +0,0 @@
-rhc
-jsquyres

orte/mca/ras/lanl_bproc/ras_lanl_bproc.c

@@ -107,10 +107,7 @@ static int orte_ras_lanl_bproc_node_resolve(char* node_name, int* node_num)
  *  - check for additional nodes that have already been allocated
  */
 
-static int orte_ras_lanl_bproc_discover(
-    opal_list_t* nodelist,
-    orte_app_context_t** context,
-    size_t num_context)
+static int orte_ras_lanl_bproc_discover(opal_list_t* nodelist)
 {
     char* nodes;
     char* ptr;
@@ -138,6 +135,7 @@ static int orte_ras_lanl_bproc_discover(
             continue;
         }
 
+        /* ensure that the node is "up" */
         if(orte_ras_lanl_bproc_node_state(node_num) != ORTE_NODE_STATE_UP) {
             opal_list_remove_item(nodelist,item);
             OBJ_DESTRUCT(item);
@@ -145,6 +143,7 @@ static int orte_ras_lanl_bproc_discover(
             continue;
         }
 
+        /* ensure that we have execution authority on it */
         if(bproc_access(node_num, BPROC_X_OK) != 0) {
             opal_list_remove_item(nodelist,item);
             OBJ_DESTRUCT(item);
@@ -186,11 +185,14 @@ static int orte_ras_lanl_bproc_discover(
             continue;
         }
 
+        /* check that the node is alive */
         if(ORTE_NODE_STATE_UP != (node_state = orte_ras_lanl_bproc_node_state(node_num))) {
             opal_output(0, "error: a specified node (%d) is not up.\n", node_num);
             rc = ORTE_ERROR;
             goto cleanup;
         }
+
+        /* check that we have execution authority on it */
         if(bproc_access(node_num, BPROC_X_OK) != 0) {
             opal_output(0, "error: a specified node (%d) is not accessible.\n", node_num);
             rc = ORTE_ERROR;
@@ -239,18 +241,10 @@ static int orte_ras_lanl_bproc_allocate(orte_jobid_t jobid, opal_list_t *attribu
     opal_list_t nodes;
     opal_list_item_t* item;
     int rc;
-    orte_app_context_t **context = NULL;
-    orte_std_cntr_t i, num_context = 0;
 
     OBJ_CONSTRUCT(&nodes, opal_list_t);
     
-    rc = orte_rmgr.get_app_context(jobid, &context, &num_context);
-    if(ORTE_SUCCESS != rc) {
-        ORTE_ERROR_LOG(rc);
-        goto cleanup;
-    }
-
-    if(ORTE_SUCCESS != (rc = orte_ras_lanl_bproc_discover(&nodes, context, num_context))) {
+    if(ORTE_SUCCESS != (rc = orte_ras_lanl_bproc_discover(&nodes))) {
         ORTE_ERROR_LOG(rc);
         goto cleanup;
     }
@@ -265,12 +259,6 @@ cleanup:
         OBJ_RELEASE(item);
     }
     OBJ_DESTRUCT(&nodes);
-    for(i=0; i<num_context; i++) {
-        OBJ_RELEASE(context[i]);
-    }
-    if (NULL != context) {
-        free(context);
-    }
     return rc;
 }
 

orte/mca/ras/lanl_bproc/ras_lanl_bproc_component.c

@@ -18,10 +18,15 @@
 
 #include "orte_config.h"
 #include "orte/orte_constants.h"
+
+#include <sys/bproc.h>
+
 #include "opal/mca/base/base.h"
 #include "opal/mca/base/mca_base_param.h"
-#include "orte/util/proc_info.h"
 #include "opal/util/output.h"
+
+#include "orte/util/proc_info.h"
+
 #include "ras_lanl_bproc.h"
 
 /*