Merged bug24365 into default

2 files changed, 21 insertions, 4 deletions

diff --git a/src/rabbit_mnesia.erl b/src/rabbit_mnesia.erl
index 60dd0770..4d419fd9 100644
--- a/src/rabbit_mnesia.erl
+++ b/src/rabbit_mnesia.erl
@@ -732,18 +732,18 @@ reset(Force) ->
         false -> ok
     end,
     Node = node(),
+    Nodes = all_clustered_nodes() -- [Node],
     case Force of
         true  -> ok;
         false ->
             ensure_mnesia_dir(),
             start_mnesia(),
-            {Nodes, RunningNodes} =
+            RunningNodes =
                 try
                     %% Force=true here so that reset still works when clustered
                     %% with a node which is down
                     ok = init_db(read_cluster_nodes_config(), true),
-                    {all_clustered_nodes() -- [Node],
-                     running_clustered_nodes() -- [Node]}
+                    running_clustered_nodes() -- [Node]
                 after
                     stop_mnesia()
                 end,
@@ -751,6 +751,10 @@ reset(Force) ->
             rabbit_misc:ensure_ok(mnesia:delete_schema([Node]),
                                   cannot_delete_schema)
     end,
+    %% We need to make sure that we don't end up in a distributed
+    %% Erlang system with nodes while not being in an Mnesia cluster
+    %% with them. We don't handle that well.
+    [erlang:disconnect_node(N) || N <- Nodes],
     ok = delete_cluster_nodes_config(),
     %% remove persisted messages and any other garbage we find
     ok = rabbit_file:recursive_delete(filelib:wildcard(dir() ++ "/*")),
diff --git a/src/rabbit_tests.erl b/src/rabbit_tests.erl
index 7a96af26..433ed9cb 100644
--- a/src/rabbit_tests.erl
+++ b/src/rabbit_tests.erl
@@ -71,10 +71,13 @@ maybe_run_cluster_dependent_tests() ->
 run_cluster_dependent_tests(SecondaryNode) ->
     SecondaryNodeS = atom_to_list(SecondaryNode),
 
+    cover:stop(SecondaryNode),
     ok = control_action(stop_app, []),
     ok = control_action(reset, []),
     ok = control_action(cluster, [SecondaryNodeS]),
     ok = control_action(start_app, []),
+    cover:start(SecondaryNode),
+    ok = control_action(start_app, SecondaryNode, [], []),
 
     io:format("Running cluster dependent tests with node ~p~n", [SecondaryNode]),
     passed = test_delegates_async(SecondaryNode),
@@ -960,7 +963,9 @@ test_cluster_management2(SecondaryNode) ->
     ok = control_action(cluster, [SecondaryNodeS, NodeS]),
     ok = control_action(start_app, []),
     ok = control_action(stop_app, []),
+    cover:stop(SecondaryNode),
     ok = control_action(reset, []),
+    cover:start(SecondaryNode),
 
     %% attempt to leave cluster when no other node is alive
     ok = control_action(cluster, [SecondaryNodeS, NodeS]),
@@ -977,7 +982,15 @@ test_cluster_management2(SecondaryNode) ->
     %% leave system clustered, with the secondary node as a ram node
     ok = control_action(force_reset, []),
     ok = control_action(start_app, []),
-    ok = control_action(force_reset, SecondaryNode, [], []),
+    %% Yes, this is rather ugly. But since we're a clustered Mnesia
+    %% node and we're telling another clustered node to reset itself,
+    %% we will get disconnected half way through causing a
+    %% badrpc. This never happens in real life since rabbitmqctl is
+    %% not a clustered Mnesia node.
+    cover:stop(SecondaryNode),
+    {badrpc, nodedown} = control_action(force_reset, SecondaryNode, [], []),
+    pong = net_adm:ping(SecondaryNode),
+    cover:start(SecondaryNode),
     ok = control_action(cluster, SecondaryNode, [NodeS], []),
     ok = control_action(start_app, SecondaryNode, [], []),