Promote down slave to master if the master gets forgotten.bug26254

2 files changed, 29 insertions, 15 deletions

diff --git a/src/rabbit_amqqueue.erl b/src/rabbit_amqqueue.erl
index 7a10d239..4e23dbd2 100644
--- a/src/rabbit_amqqueue.erl
+++ b/src/rabbit_amqqueue.erl
@@ -258,15 +258,16 @@ declare(QueueName, Durable, AutoDelete, Args, Owner) ->
 declare(QueueName, Durable, AutoDelete, Args, Owner, Node) ->
     ok = check_declare_arguments(QueueName, Args),
     Q = rabbit_queue_decorator:set(
-          rabbit_policy:set(#amqqueue{name            = QueueName,
-                                      durable         = Durable,
-                                      auto_delete     = AutoDelete,
-                                      arguments       = Args,
-                                      exclusive_owner = Owner,
-                                      pid             = none,
-                                      slave_pids      = [],
-                                      sync_slave_pids = [],
-                                      gm_pids         = []})),
+          rabbit_policy:set(#amqqueue{name             = QueueName,
+                                      durable          = Durable,
+                                      auto_delete      = AutoDelete,
+                                      arguments        = Args,
+                                      exclusive_owner  = Owner,
+                                      pid              = none,
+                                      slave_pids       = [],
+                                      sync_slave_pids  = [],
+                                      down_slave_nodes = [],
+                                      gm_pids          = []})),
     Node = rabbit_mirror_queue_misc:initial_queue_node(Q, Node),
     gen_server2:call(start_queue_process(Node, Q), {init, new}, infinity).
 
@@ -666,15 +667,23 @@ forget_all_durable(Node) ->
           fun () ->
                   Qs = mnesia:match_object(rabbit_durable_queue,
                                            #amqqueue{_ = '_'}, write),
-                  [rabbit_binding:process_deletions(
-                     internal_delete1(Name)) ||
-                      #amqqueue{name = Name, pid = Pid} = Q <- Qs,
-                      node(Pid) =:= Node,
-                      rabbit_policy:get(<<"ha-mode">>, Q) =:= undefined],
+                  [forget_node_for_queue(Q) || #amqqueue{pid = Pid} = Q <- Qs,
+                      node(Pid) =:= Node],
                   ok
           end),
     ok.
 
+forget_node_for_queue(#amqqueue{name             = Name,
+                                down_slave_nodes = []}) ->
+    %% No slaves to recover from, queue is gone
+    rabbit_binding:process_deletions(internal_delete1(Name));
+
+forget_node_for_queue(Q = #amqqueue{down_slave_nodes = [H|T]}) ->
+    %% Promote a slave while down - it'll happily recover as a master
+    Q1 = Q#amqqueue{pid              = rabbit_misc:node_to_fake_pid(H),
+                    down_slave_nodes = T},
+    ok = mnesia:write(rabbit_durable_queue, Q1, write).
+
 run_backing_queue(QPid, Mod, Fun) ->
     gen_server2:cast(QPid, {run_backing_queue, Mod, Fun}).
 
diff --git a/src/rabbit_misc.erl b/src/rabbit_misc.erl
index 6f353da5..fd4b7b11 100644
--- a/src/rabbit_misc.erl
+++ b/src/rabbit_misc.erl
@@ -45,7 +45,7 @@
 -export([with_local_io/1, local_info_msg/2]).
 -export([unfold/2, ceil/1, queue_fold/3]).
 -export([sort_field_table/1]).
--export([pid_to_string/1, string_to_pid/1]).
+-export([pid_to_string/1, string_to_pid/1, node_to_fake_pid/1]).
 -export([version_compare/2, version_compare/3]).
 -export([version_minor_equivalent/2]).
 -export([dict_cons/3, orddict_cons/3, gb_trees_cons/3]).
@@ -193,6 +193,7 @@
         (rabbit_framing:amqp_table()) -> rabbit_framing:amqp_table()).
 -spec(pid_to_string/1 :: (pid()) -> string()).
 -spec(string_to_pid/1 :: (string()) -> pid()).
+-spec(node_to_fake_pid/1 :: (atom()) -> pid()).
 -spec(version_compare/2 :: (string(), string()) -> 'lt' | 'eq' | 'gt').
 -spec(version_compare/3 ::
         (string(), string(), ('lt' | 'lte' | 'eq' | 'gte' | 'gt'))
@@ -709,6 +710,10 @@ string_to_pid(Str) ->
             throw(Err)
     end.
 
+%% node(node_to_fake_pid(Node)) =:= Node.
+node_to_fake_pid(Node) ->
+    string_to_pid(format("<~s.0.0.0>", [Node])).
+
 version_compare(A, B, lte) ->
     case version_compare(A, B) of
         eq -> true;