Define a new key and store the process name of the local_rank=0 process on each node so that the MPI layer can retrieve it as desired.

This commit was SVN r30759.

opal/dss/dss_types.h

@@ -53,6 +53,7 @@ typedef struct {
 
 /* define an RTE-agnostic process identifier */
 typedef uint64_t opal_identifier_t;
+#define OPAL_ID_T OPAL_UINT64
 
 /* Type defines for packing and unpacking */
 #define    OPAL_UNDEF               (opal_data_type_t)    0 /**< type hasn't been defined yet */

opal/mca/db/db_types.h

@@ -30,6 +30,7 @@ BEGIN_C_DECLS
 #define OPAL_DB_JOB_SDIR    "opal.job.session.dir"
 #define OPAL_DB_MY_SDIR     "opal.my.session.dir"
 #define OPAL_DB_LOCALRANK   "opal.local.rank"
+#define OPAL_DB_LOCALLDR    "opal.local.ldr"
 
 END_C_DECLS
 

orte/mca/ess/pmi/ess_pmi_module.c

@@ -95,6 +95,7 @@ static int rte_init(void)
     int *ranks=NULL;
     orte_jobid_t jobid;
     char *rmluri;
+    orte_process_name_t ldr;
 
     /* run the prolog */
     if (ORTE_SUCCESS != (ret = orte_ess_base_std_prolog())) {
@@ -333,6 +334,14 @@ static int rte_init(void)
                 break;
             }
         }
+        /* store the name of the local leader */
+        ldr.jobid = ORTE_PROC_MY_NAME->jobid;
+        ldr.vpid = ranks[0];
+        if (ORTE_SUCCESS != (ret = opal_db.store((opal_identifier_t*)&ldr, OPAL_SCOPE_INTERNAL,
+                                                 OPAL_DB_LOCALLDR, (opal_identifier_t*)&ldr, OPAL_ID_T))) {
+            error = "storing local leader";
+            goto error;
+        }
         free(ranks);
 
         /* setup process binding */

orte/util/nidmap.c

@@ -979,6 +979,14 @@ int orte_util_decode_pidmap(opal_byte_object_t *bo)
                 /* set mine */
                 orte_process_info.my_local_rank = local_rank;
                 orte_process_info.my_node_rank = node_rank;
+                /* if we are the local leader (i.e., local_rank=0), then record it */
+                if (0 == local_rank) {
+                    if (ORTE_SUCCESS != (rc = opal_db.store((opal_identifier_t*)&proc, OPAL_SCOPE_INTERNAL,
+                                                            OPAL_DB_LOCALLDR, (opal_identifier_t*)&proc, OPAL_ID_T))) {
+                        ORTE_ERROR_LOG(rc);
+                        goto cleanup;
+                    }
+                }
 #if OPAL_HAVE_HWLOC
                 if (NULL != cpu_bitmap) {
                     orte_process_info.cpuset = strdup(cpu_bitmap);
@@ -988,6 +996,14 @@ int orte_util_decode_pidmap(opal_byte_object_t *bo)
                        dmn.vpid == ORTE_PROC_MY_DAEMON->vpid) {
                 /* if we share a daemon, then add to my local peers */
                 orte_process_info.num_local_peers++;
+                /* if this is the local leader (i.e., local_rank=0), then record it */
+                if (0 == local_rank) {
+                    if (ORTE_SUCCESS != (rc = opal_db.store((opal_identifier_t*)&proc, OPAL_SCOPE_INTERNAL,
+                                                            OPAL_DB_LOCALLDR, (opal_identifier_t*)&proc, OPAL_ID_T))) {
+                        ORTE_ERROR_LOG(rc);
+                        goto cleanup;
+                    }
+                }
             }
             /* apps don't need the rest of the data in the buffer for this proc,
              * but we have to unpack it anyway to stay in sync