Some checkpoint/restart fixes in response to r18338 (changes in modex).

Things should be working now.

This commit was SVN r18348.

The following SVN revision numbers were found above:
  r18338 --> open-mpi/ompi@3e55fe6f6d

ompi/mca/pml/ob1/pml_ob1.c

@@ -536,17 +536,7 @@ int mca_pml_ob1_ft_event( int state )
                         ret);
             return ret;
         }
-#if 0
-        /*
-         * Fill in remote proc information
-         */
-        if (OMPI_SUCCESS != (ret = ompi_proc_get_info())) {
-            opal_output(0,
-                        "pml:ob1: ft_event(Restart): Failed ompi_proc_get_info() = %d",
-                        ret);
-            return ret;
-        }
-#endif
+
         /*
          * Startup the PML stack now that the modex is running again
          * Add the new procs (BTLs redo modex recv's)

ompi/proc/proc.c

@@ -278,6 +278,9 @@ int ompi_proc_refresh(void) {
          item  = opal_list_get_next(item), ++i ) {
         proc = (ompi_proc_t*)item;
 
+        /* Does not change: proc->proc_name.vpid */
+        proc->proc_name.jobid = ORTE_PROC_MY_NAME->jobid;
+
         if (i == ORTE_PROC_MY_NAME->vpid) {
             ompi_proc_local_proc = proc;
             proc->proc_flags |= OMPI_PROC_FLAG_LOCAL;

orte/mca/ess/env/ess_env_module.c

@@ -109,7 +109,7 @@ orte_ess_base_module_t orte_ess_env_module = {
 };
 
 static opal_pointer_array_t nidmap;
-static orte_pmap_t *pmap;
+static orte_pmap_t *pmap = NULL;
 static orte_vpid_t nprocs;
 
 static int rte_init(char flags)
@@ -198,11 +198,13 @@ static int rte_finalize(void)
             }
             if (NULL != nids[i]->name) {
                 free(nids[i]->name);
+                nids[i]->name = NULL;
             }
         }
         OBJ_DESTRUCT(&nidmap);
         free(pmap);
-        
+        pmap = NULL;
+
         /* use the default procedure to finish */
         if (ORTE_SUCCESS != (ret = orte_ess_base_app_finalize())) {
             ORTE_ERROR_LOG(ret);
@@ -341,6 +343,8 @@ static int rte_ft_event(int state)
     int ret, exit_status = ORTE_SUCCESS;
     char * procid_str = NULL;
     char * jobid_str  = NULL;
+    orte_nid_t **nids = NULL;
+    int32_t i;
 
     /******** Checkpoint Prep ********/
     if(OPAL_CRS_CHECKPOINT == state) {
@@ -416,6 +420,23 @@ static int rte_ft_event(int state)
          * This should follow the ess init() function
          */
 
+        /*
+         * Clear nidmap
+         */
+        nids = (orte_nid_t**)nidmap.addr;
+        for (i=0; i < nidmap.size; i++) {
+            if (NULL == nids[i]) {
+                break;
+            }
+            if (NULL != nids[i]->name) {
+                free(nids[i]->name);
+                nids[i]->name = NULL;
+            }
+        }
+        OBJ_DESTRUCT(&nidmap);
+        free(pmap);
+        pmap = NULL;
+
         /*
          * - Reset Contact information
          */
@@ -541,6 +562,20 @@ static int rte_ft_event(int state)
             exit_status = ret;
             goto cleanup;
         }
+
+        /*
+         * Refresh nidmap structure
+         */
+        OBJ_CONSTRUCT(&nidmap, opal_pointer_array_t);
+        opal_pointer_array_init(&nidmap, 8, INT32_MAX, 8);
+
+        /* if one was provided, build my nidmap */
+        if (ORTE_SUCCESS != (ret = orte_ess_base_build_nidmap(orte_process_info.sync_buf,
+                                                              &nidmap, &pmap, &nprocs))) {
+            ORTE_ERROR_LOG(ret);
+            exit_status = ret;
+            goto cleanup;
+        }
     }
     else if (OPAL_CRS_TERM == state ) {
         /* Nothing */

orte/util/proc_info.c

@@ -208,6 +208,9 @@ int orte_proc_info_finalize(void)
     orte_process_info.singleton = false;
     orte_process_info.daemon = false;
     
+    OBJ_RELEASE(orte_process_info.sync_buf);
+    orte_process_info.sync_buf = NULL;
+
     init = false;
     return ORTE_SUCCESS;
 }