Solve startup conflict between nodes by restarting slave runtime

1 files changed, 29 insertions, 18 deletions

diff --git a/src/rabbit_mirror_queue_slave.erl b/src/rabbit_mirror_queue_slave.erl
index 54ee7129..ec6e500f 100644
--- a/src/rabbit_mirror_queue_slave.erl
+++ b/src/rabbit_mirror_queue_slave.erl
@@ -137,28 +137,39 @@ init(Q = #amqqueue { name = QName }) ->
             {stop, {duplicate_live_master, Node}};
         existing ->
             gm:leave(GM),
+            ignore;
+        master_recovery_detected ->
+            init:restart(),
             ignore
     end.
 
 init_it(Self, GM, Node, QName) ->
-    [Q = #amqqueue { pid = QPid, slave_pids = SPids, gm_pids = GMPids }] =
-        mnesia:read({rabbit_queue, QName}),
-    case [Pid || Pid <- [QPid | SPids], node(Pid) =:= Node] of
-        []     -> add_slave(Q, Self, GM),
-                  {new, QPid, GMPids};
-        [QPid] -> case rabbit_misc:is_process_alive(QPid) of
-                      true  -> duplicate_live_master;
-                      false -> {stale, QPid}
-                  end;
-        [SPid] -> case rabbit_misc:is_process_alive(SPid) of
-                      true  -> existing;
-                      false -> Q1 = Q#amqqueue {
-                                      slave_pids = SPids -- [SPid],
-                                      gm_pids    = [T || T = {_, S} <- GMPids,
-                                                         S =/= SPid] },
-                               add_slave(Q1, Self, GM),
-                               {new, QPid, GMPids}
-                  end
+    case mnesia:read({rabbit_queue, QName}) of
+        [Q = #amqqueue { pid = QPid, slave_pids = SPids, gm_pids = GMPids }] ->
+            case [Pid || Pid <- [QPid | SPids], node(Pid) =:= Node] of
+                [] ->
+                    add_slave(Q, Self, GM),
+                    {new, QPid, GMPids};
+                [QPid] ->
+                    case rabbit_misc:is_process_alive(QPid) of
+                        true  -> duplicate_live_master;
+                        false -> {stale, QPid}
+                    end;
+                [SPid] ->
+                    case rabbit_misc:is_process_alive(SPid) of
+                        true ->
+                            existing;
+                        false ->
+                            Q1 = Q#amqqueue {
+                                   slave_pids = SPids -- [SPid],
+                                   gm_pids    = [T || T = {_, S} <- GMPids,
+                                                      S =/= SPid] },
+                            add_slave(Q1, Self, GM),
+                            {new, QPid, GMPids}
+                    end
+            end;
+        [] ->
+            master_recovery_detected
     end.
 
 %% Add to the end, so they are in descending order of age, see