Merge pull request #752 from davidvossel/pcmk-remote-failcount

High: pengine: clear failures after reconnect_interval expires
ClusterLabs · Jul 9, 2015 · ef86b49 · ef86b49
2 parents d81f5f1 + a109cee
commit ef86b49
Show file tree

Hide file tree

Showing 2 changed files with 35 additions and 3 deletions.
diff --git a/lib/pengine/unpack.c b/lib/pengine/unpack.c
@@ -2834,8 +2834,9 @@ static bool check_operation_expiry(resource_t *rsc, node_t *node, int rc, xmlNod
 
             node_t *remote_node = pe_find_node(data_set->nodes, rsc->id);
             if (remote_node && remote_node->details->remote_was_fenced == 0) {
-
-                crm_info("Waiting to clear monitor failure for remote node %s until fencing has occured", rsc->id); 
+                if (strstr(ID(xml_op), "last_failure")) {
+                    crm_info("Waiting to clear monitor failure for remote node %s until fencing has occured", rsc->id); 
+                }
                 /* disabling failure timeout for this operation because we believe
                  * fencing of the remote node should occur first. */ 
                 failure_timeout = 0;
@@ -2866,6 +2867,9 @@ static bool check_operation_expiry(resource_t *rsc, node_t *node, int rc, xmlNod
                 } else {
                     expired = FALSE;
                 }
+            } else if (rsc->remote_reconnect_interval && strstr(ID(xml_op), "last_failure")) {
+                /* always clear last failure when reconnect interval is set */
+                clear_failcount = 1;
             }
         }
 

diff --git a/pengine/allocate.c b/pengine/allocate.c
@@ -1681,10 +1681,38 @@ apply_remote_node_ordering(pe_working_set_t *data_set)
         resource_t *remote_rsc = NULL;
         resource_t *container = NULL;
 
+        if (action->rsc == NULL) {
+            continue;
+        }
+
+        /* Special case. */
+        if (action->rsc &&
+            action->rsc->is_remote_node &&
+            safe_str_eq(action->task, CRM_OP_CLEAR_FAILCOUNT)) {
+
+            /* if we are clearing the failcount of an actual remote node connect
+             * resource, then make sure this happens before allowing the connection
+             * to start if we are planning on starting the connection during this
+             * transition */ 
+            custom_action_order(action->rsc,
+                NULL,
+                action,
+                action->rsc,
+                generate_op_key(action->rsc->id, RSC_START, 0),
+                NULL,
+                pe_order_optional,
+                data_set);
+
+                continue;
+        }
+
+        /* detect if the action occurs on a remote node. if so create
+         * ordering constraints that guarantee the action occurs while
+         * the remote node is active (after start, before stop...) things
+         * like that */ 
         if (action->node == NULL ||
             is_remote_node(action->node) == FALSE ||
             action->node->details->remote_rsc == NULL ||
-            action->rsc == NULL ||
             is_set(action->flags, pe_action_pseudo)) {
             continue;
         }