Base the whole thing off net_adm:ping/1 - because we might see other nodes come back but also be waiting (in the no-majority case, and RAM nodes). Better to detect they exist and come back than to stay stuck because they don't happen to be running Mnesia.

2 files changed, 10 insertions, 11 deletions

diff --git a/src/rabbit_mnesia.erl b/src/rabbit_mnesia.erl
index ecb03f54..c39e898c 100644
--- a/src/rabbit_mnesia.erl
+++ b/src/rabbit_mnesia.erl
@@ -26,7 +26,6 @@
 
          status/0,
          is_clustered/0,
-         majority/0,
          cluster_nodes/1,
          node_type/0,
          dir/0,
@@ -68,7 +67,6 @@
 -spec(status/0 :: () -> [{'nodes', [{node_type(), [node()]}]} |
                          {'running_nodes', [node()]} |
                          {'partitions', [{node(), [node()]}]}]).
--spec(majority/0 :: () -> boolean()).
 -spec(is_clustered/0 :: () -> boolean()).
 -spec(cluster_nodes/1 :: ('all' | 'disc' | 'ram' | 'running') -> [node()]).
 -spec(node_type/0 :: () -> node_type()).
@@ -340,10 +338,6 @@ status() ->
             false -> []
         end.
 
-majority() ->
-    ensure_mnesia_running(),
-    (length(cluster_nodes(running)) / length(cluster_nodes(all))) > 0.5.
-
 mnesia_partitions(Nodes) ->
     {Replies, _BadNodes} = rpc:multicall(
                              Nodes, rabbit_node_monitor, partitions, []),
diff --git a/src/rabbit_node_monitor.erl b/src/rabbit_node_monitor.erl
index 42df6e5d..249c17a4 100644
--- a/src/rabbit_node_monitor.erl
+++ b/src/rabbit_node_monitor.erl
@@ -281,7 +281,7 @@ handle_dead_rabbit(Node) ->
 %% down - otherwise we have a race.
 handle_dead_according_to_mnesia_rabbit() ->
     case application:get_env(rabbit, cluster_cp_mode) of
-        {ok, true}  -> case rabbit_mnesia:majority() of
+        {ok, true}  -> case majority() of
                            true  -> ok;
                            false -> await_cluster_recovery()
                        end;
@@ -289,6 +289,13 @@ handle_dead_according_to_mnesia_rabbit() ->
     end,
     ok.
 
+majority() ->
+    Nodes = rabbit_mnesia:cluster_nodes(all),
+    Alive = [Status || N      <- Nodes,
+                       Status <- [net_adm:ping(N)],
+                       Status =:= pong],
+    length(Alive) / length(Nodes) > 0.5.
+
 await_cluster_recovery() ->
     rabbit_log:warning("Cluster minority status detected - awaiting recovery~n",
                        []),
@@ -303,11 +310,9 @@ await_cluster_recovery() ->
 
 wait_for_cluster_recovery(Nodes) ->
     [erlang:disconnect_node(Node) || Node <- Nodes],
-    mnesia:start(),
-    case rabbit_mnesia:majority() of
+    case majority() of
         true  -> rabbit:start();
-        false -> mnesia:stop(),
-                 timer:sleep(1000),
+        false -> timer:sleep(1000),
                  wait_for_cluster_recovery(Nodes)
     end.