- get rid of rpc:call to avoid group leader inheritance

- do not log migration errors
- remove stopping node from cluster hashing explicitly

src/ejabberd_app.erl

@@ -80,6 +80,7 @@ start(_, _) ->
 %% This function is called when an application is about to be stopped,
 %% before shutting down the processes of the application.
 prep_stop(State) ->
+    ejabberd_cluster:shutdown(),
     stop_modules(),
     ejabberd_admin:stop(),
     broadcast_c2s_shutdown(),

src/ejabberd_cluster.erl

@@ -10,7 +10,7 @@
 -behaviour(gen_server).
 
 %% API
--export([start_link/0, get_node/1, get_node_new/1, announce/0,
+-export([start_link/0, get_node/1, get_node_new/1, announce/0, shutdown/0,
 	 node_id/0, get_node_by_id/1, get_nodes/0, rehash_timeout/0]).
 
 %% gen_server callbacks
@@ -68,6 +68,14 @@ get_node_by_id(NodeID) ->
 	    node()
     end.
 
+shutdown() ->
+    lists:foreach(
+      fun(Node) when Node /= node() ->
+	      {ejabberd_cluster, Node} ! {node_down, node()};
+	 (_) ->
+	      ok
+      end, get_nodes()).
+
 %%====================================================================
 %% gen_server callbacks
 %%====================================================================
@@ -118,6 +126,10 @@ handle_info({node_ready, Node}, State) ->
     ?INFO_MSG("adding node ~p to hash", [Node]),
     append_node(?HASHTBL, Node),
     {noreply, State};
+handle_info({node_down, Node}, State) ->
+    delete_node(?HASHTBL, Node),
+    delete_node(?HASHTBL_NEW, Node),
+    {noreply, State};
 handle_info({nodedown, Node, _}, State) ->
     ?INFO_MSG("node ~p goes down", [Node]),
     delete_node(?HASHTBL, Node),

src/p1_fsm.erl

@@ -525,8 +525,8 @@ relay_messages(MRef, TRef, Clone, Queue) ->
 
 relay_messages(MRef, TRef, Clone) ->
     receive
-	{'DOWN', MRef, process, Clone, Reason} ->
-	    Reason;
+	{'DOWN', MRef, process, Clone, _Reason} ->
+	    normal;
 	{'EXIT', _Parent, _Reason} ->
 	    {migrated, Clone};
 	{timeout, TRef, timeout} ->
@@ -563,15 +563,7 @@ handle_msg(Msg, Parent, Name, StateName, StateData, Mod, _Time,
 				 Time1
 			 end,
 	    Now = now(),
-	    Reason = case catch rpc:call(Node, M, F, A, RPCTimeout) of
-			 {badrpc, timeout} ->
-			     normal;
-			 {badrpc, _} = Err ->
-			     {migration_error, Err};
-			 {'EXIT', _} = Err ->
-			     {migration_error, Err};
-			 {error, _} = Err ->
-			     {migration_error, Err};
+	    Reason = case catch rpc_call(Node, M, F, A, RPCTimeout) of
 			 {ok, Clone} ->
 			     process_flag(trap_exit, true),
 			     MRef = erlang:monitor(process, Clone),
@@ -579,8 +571,8 @@ handle_msg(Msg, Parent, Name, StateName, StateData, Mod, _Time,
 			     TimeLeft = lists:max([Time1 - NowDiff, 0]),
 			     TRef = erlang:start_timer(TimeLeft, self(), timeout),
 			     relay_messages(MRef, TRef, Clone, Queue);
-			 Reply ->
-			     {migration_error, {bad_reply, Reply}}
+			 _ ->
+			     normal
 		     end,
 	    terminate(Reason, Name, Msg, Mod, StateName, NStateData, []);
 	{stop, Reason, NStateData} ->
@@ -628,15 +620,7 @@ handle_msg(Msg, Parent, Name, StateName, StateData,
 				 Time1
 			 end,
 	    Now = now(),
-	    Reason = case catch rpc:call(Node, M, F, A, RPCTimeout) of
-			 {badrpc, timeout} ->
-			     normal;
-			 {badrpc, R} ->
-			     {migration_error, R};
-			 {'EXIT', R} ->
-			     {migration_error, R};
-			 {error, R} ->
-			     {migration_error, R};
+	    Reason = case catch rpc_call(Node, M, F, A, RPCTimeout) of
 			 {ok, Clone} ->
 			     process_flag(trap_exit, true),
 			     MRef = erlang:monitor(process, Clone),
@@ -644,8 +628,8 @@ handle_msg(Msg, Parent, Name, StateName, StateData,
 			     TimeLeft = lists:max([Time1 - NowDiff, 0]),
 			     TRef = erlang:start_timer(TimeLeft, self(), timeout),
 			     relay_messages(MRef, TRef, Clone, Queue);
-			 Reply ->
-			     {migration_error, {bad_reply, Reply}}
+			 _ ->
+			     normal
 		     end,
 	    terminate(Reason, Name, Msg, Mod, StateName, NStateData, Debug);
 	{stop, Reason, NStateData} ->
@@ -832,3 +816,34 @@ message_queue_len(#limits{max_queue = MaxQueue}, QueueLen) ->
 	_ ->
 	    ok
     end.
+
+rpc_call(Node, Mod, Fun, Args, Timeout) ->
+    Ref = make_ref(),
+    Caller = self(),
+    F = fun() ->
+		group_leader(whereis(user), self()),
+		case catch apply(Mod, Fun, Args) of
+		    {'EXIT', _} = Err ->
+			Caller ! {Ref, {badrpc, Err}};
+		    Result ->
+			Caller ! {Ref, Result}
+		end
+	end,
+    Pid = spawn(Node, F),
+    MRef = erlang:monitor(process, Pid),
+    receive
+	{Ref, Result} ->
+	    erlang:demonitor(MRef, [flush]),
+	    Result;
+	{'DOWN', MRef, _, _, Reason} ->
+	    {badrpc, Reason}
+    after Timeout ->
+	    erlang:demonitor(MRef, [flush]),
+	    catch exit(Pid, kill),
+	    receive
+		{Ref, Result} ->
+		    Result
+	    after 0 ->
+		    {badrpc, timeout}
+	    end
+    end.