Support timeout values when performing connect/accept operations. Bump default timeout to 10 minutes so folks have time to start the partnering application

ompi/communicator/comm_cid.c

@@ -875,7 +875,7 @@ static int ompi_comm_allreduce_pmix_reduce_complete (ompi_comm_request_t *reques
 
     /* this macro is not actually non-blocking. if a non-blocking version becomes available this function
      * needs to be reworked to take advantage of it. */
-    OPAL_PMIX_EXCHANGE(rc, &info, &pdat, 60);
+    OPAL_PMIX_EXCHANGE(rc, &info, &pdat, 600);  // give them 10 minutes
     OBJ_DESTRUCT(&info);
     if (OPAL_SUCCESS != rc) {
         OBJ_DESTRUCT(&pdat);

ompi/dpm/dpm.c

@@ -211,7 +211,7 @@ int ompi_dpm_connect_accept(ompi_communicator_t *comm, int root,
         info.data.string = opal_argv_join(members, ':');
         pdat.value.type = OPAL_STRING;
 
-        OPAL_PMIX_EXCHANGE(rc, &info, &pdat, 60);
+        OPAL_PMIX_EXCHANGE(rc, &info, &pdat, 600);  // give them 10 minutes
         OBJ_DESTRUCT(&info);
         if (OPAL_SUCCESS != rc) {
             OBJ_DESTRUCT(&pdat);

opal/mca/pmix/base/pmix_base_fns.c

@@ -162,8 +162,7 @@ int opal_pmix_base_exchange(opal_value_t *indat,
     info->type = OPAL_BOOL;
     info->data.flag = true;
     opal_list_append(&mlist, &info->super);
-    if (0 < timeout) {
-	/* give it a decent timeout as we don't know when
+    /* pass along the given timeout as we don't know when
      * the other side will publish - it doesn't
      * have to be simultaneous */
     info = OBJ_NEW(opal_value_t);
@@ -171,7 +170,6 @@ int opal_pmix_base_exchange(opal_value_t *indat,
     info->type = OPAL_INT;
     info->data.integer = timeout;
     opal_list_append(&mlist, &info->super);
-    }
 
     /* if a non-blocking version of lookup isn't
      * available, then use the blocking version */

opal/mca/pmix/pmix2x/pmix2x_client.c

@@ -437,7 +437,8 @@ int pmix2x_get(const opal_process_name_t *proc, const char *key,
             n=0;
             OPAL_LIST_FOREACH(ival, info, opal_value_t) {
                 (void)strncpy(pinfo[n].key, ival->key, PMIX_MAX_KEYLEN);
-		pmix2x_value_load(&pinfo[n++].value, ival);
+                pmix2x_value_load(&pinfo[n].value, ival);
+                ++n;
             }
         } else {
             pinfo = NULL;
@@ -534,6 +535,7 @@ int pmix2x_getnb(const opal_process_name_t *proc, const char *key,
             OPAL_LIST_FOREACH(ival, info, opal_value_t) {
                 (void)strncpy(op->info[n].key, ival->key, PMIX_MAX_KEYLEN);
                 pmix2x_value_load(&op->info[n].value, ival);
+                ++n;
             }
         }
     }

orte/orted/pmix/pmix_server.c

@@ -151,11 +151,18 @@ static void eviction_cbfunc(struct opal_hotel_t *hotel,
                             int room_num, void *occupant)
 {
     pmix_server_req_t *req = (pmix_server_req_t*)occupant;
+    bool timeout = false;
     int rc;
 
     /* decrement the request timeout */
     req->timeout -= orte_pmix_server_globals.timeout;
-    if (0 < req->timeout) {
+    if (req->timeout > 0) {
+        req->timeout -= orte_pmix_server_globals.timeout;
+        if (0 >= req->timeout) {
+            timeout = true;
+        }
+    }
+    if (!timeout) {
         /* not done yet - check us back in */
         if (OPAL_SUCCESS == (rc = opal_hotel_checkin(&orte_pmix_server_globals.reqs, req, &req->room_num))) {
             return;

orte/orted/pmix/pmix_server_pub.c

@@ -156,13 +156,18 @@ int pmix_server_publish_fn(opal_process_name_t *proc,
         return rc;
     }
 
-    /* if we have items, pack those too - ignore persistence
+    /* if we have items, pack those too - ignore persistence, timeout
      * and range values */
     OPAL_LIST_FOREACH(iptr, info, opal_value_t) {
         if (0 == strcmp(iptr->key, OPAL_PMIX_RANGE) ||
             0 == strcmp(iptr->key, OPAL_PMIX_PERSISTENCE)) {
             continue;
         }
+        if (0 == strcmp(iptr->key, OPAL_PMIX_TIMEOUT)) {
+            /* record the timeout value, but don't pack it */
+            req->timeout = iptr->data.integer;
+            continue;
+        }
         opal_output_verbose(5, orte_pmix_server_globals.output,
                             "%s publishing data %s of type %d from source %s",
                             ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), iptr->key, iptr->type,
@@ -257,11 +262,16 @@ int pmix_server_lookup_fn(opal_process_name_t *proc, char **keys,
         }
     }
 
-    /* if we have items, pack those too - ignore range value */
+    /* if we have items, pack those too - ignore range and timeout value */
     OPAL_LIST_FOREACH(iptr, info, opal_value_t) {
         if (0 == strcmp(iptr->key, OPAL_PMIX_RANGE)) {
             continue;
         }
+        if (0 == strcmp(iptr->key, OPAL_PMIX_TIMEOUT)) {
+            /* record the timeout value, but don't pack it */
+            req->timeout = iptr->data.integer;
+            continue;
+        }
         if (OPAL_SUCCESS != (rc = opal_dss.pack(&req->msg, &iptr, 1, OPAL_VALUE))) {
             ORTE_ERROR_LOG(rc);
             OBJ_RELEASE(req);
@@ -347,11 +357,16 @@ int pmix_server_unpublish_fn(opal_process_name_t *proc, char **keys,
         }
     }
 
-    /* if we have items, pack those too - ignore range value */
+    /* if we have items, pack those too - ignore range and timeout value */
     OPAL_LIST_FOREACH(iptr, info, opal_value_t) {
         if (0 == strcmp(iptr->key, OPAL_PMIX_RANGE)) {
             continue;
         }
+        if (0 == strcmp(iptr->key, OPAL_PMIX_TIMEOUT)) {
+            /* record the timeout value, but don't pack it */
+            req->timeout = iptr->data.integer;
+            continue;
+        }
         if (OPAL_SUCCESS != (rc = opal_dss.pack(&req->msg, &iptr, 1, OPAL_VALUE))) {
             ORTE_ERROR_LOG(rc);
             OBJ_RELEASE(req);