A bit more faff, to deal with genuinely dying nodes.

2 files changed, 42 insertions, 16 deletions

diff --git a/src/dmon.erl b/src/dmon.erl
index dfb420c3..3f89c83a 100644
--- a/src/dmon.erl
+++ b/src/dmon.erl
@@ -16,8 +16,8 @@
 
 -module(dmon).
 
--export([new/0, monitor/2, monitor_all/2, demonitor/2, is_monitored/2, erase/2,
-         monitored/1, is_empty/1]).
+-export([new/0, monitor/2, monitor_all/2, demonitor/2, is_monitored/2,
+         monitored/1, monitored/2, is_empty/1]).
 
 -compile({no_auto_import, [monitor/2]}).
 
@@ -36,8 +36,8 @@
 -spec(monitor_all/2  :: ([item()], ?MODULE()) -> ?MODULE()).
 -spec(demonitor/2    :: (item(), ?MODULE()) -> ?MODULE()).
 -spec(is_monitored/2 :: (item(), ?MODULE()) -> boolean()).
--spec(erase/2        :: (item(), ?MODULE()) -> ?MODULE()).
 -spec(monitored/1    :: (?MODULE()) -> [item()]).
+-spec(monitored/2    :: (node(), ?MODULE()) -> [item()]).
 -spec(is_empty/1     :: (?MODULE()) -> boolean()).
 
 -endif.
@@ -45,9 +45,14 @@
 new() -> dict:new().
 
 monitor(Item, M) ->
-    case dict:is_key(Item, M) of
+    N = case dict:find(node(Item), M) of
+            {ok, N0} -> N0;
+            error    -> dict:new()
+        end,
+    case dict:is_key(Item, N) of
         true  -> M;
-        false -> dict:store(Item, delegate:monitor(Item), M)
+        false -> N2 = dict:store(Item, delegate:monitor(Item), N),
+                 dict:store(node(Item), N2, M)
     end.
 
 monitor_all([],     M) -> M;                %% optimisation
@@ -55,16 +60,30 @@ monitor_all([Item], M) -> monitor(Item, M); %% optimisation
 monitor_all(Items,  M) -> lists:foldl(fun monitor/2, M, Items).
 
 demonitor(Item, M) ->
-    case dict:find(Item, M) of
-        {ok, MRef} -> delegate:demonitor(Item, MRef),
-                      dict:erase(Item, M);
-        error      -> M
+    Node = node(Item),
+    case dict:find(Node, M) of
+        {ok, N} -> case dict:find(Item, N) of
+                       {ok, MRef} -> delegate:demonitor(Item, MRef),
+                                     N2 = dict:erase(Item, N),
+                                     case dict:size(N2) of
+                                         0 -> erlang:monitor_node(Node, false),
+                                              dict:erase(Node, M);
+                                         _ -> dict:store(Node, N2, M)
+                                     end;
+                       error      -> M
+                   end;
+        error   -> M
     end.
 
-is_monitored(Item, M) -> dict:is_key(Item, M).
+is_monitored(Item, M) -> dict:is_key(node(Item), M) andalso
+                             dict:is_key(Item, dict:fetch(node(Item), M)).
 
-erase(Item, M) -> dict:erase(Item, M).
+monitored(M) -> lists:flatten([dict:fetch_keys(dict:fetch(Node, M)) ||
+                                  Node <- dict:fetch_keys(M)]).
 
-monitored(M) -> dict:fetch_keys(M).
+monitored(Node, M) -> case dict:find(Node, M) of
+                          {ok, N} -> dict:fetch_keys(N);
+                          error   -> []
+                      end.
 
 is_empty(M) -> dict:size(M) == 0.
diff --git a/src/rabbit_mirror_queue_slave.erl b/src/rabbit_mirror_queue_slave.erl
index ca9418c6..b4c01440 100644
--- a/src/rabbit_mirror_queue_slave.erl
+++ b/src/rabbit_mirror_queue_slave.erl
@@ -273,7 +273,12 @@ handle_info({'DOWN', _MonitorRef, process, MPid, _Reason},
     noreply(State);
 
 handle_info({'DOWN', _MonitorRef, process, ChPid, _Reason}, State) ->
-    noreply(local_sender_death(ChPid, State));
+    local_sender_death(ChPid, State),
+    noreply(State);
+
+handle_info({node_down, Node}, State) ->
+    local_sender_node_death(Node, State),
+    noreply(State);
 
 handle_info({'EXIT', _Pid, Reason}, State) ->
     {stop, Reason, State};
@@ -604,7 +609,7 @@ stop_rate_timer(State) -> rabbit_misc:stop_timer(State, #state.rate_timer_ref).
 ensure_monitoring(ChPid, State = #state { known_senders = KS }) ->
     State #state { known_senders = dmon:monitor(ChPid, KS) }.
 
-local_sender_death(ChPid, State = #state { known_senders = KS }) ->
+local_sender_death(ChPid, #state { known_senders = KS }) ->
     %% The channel will be monitored iff we have received a delivery
     %% from it but not heard about its death from the master. So if it
     %% is monitored we need to point the death out to the master (see
@@ -612,8 +617,10 @@ local_sender_death(ChPid, State = #state { known_senders = KS }) ->
     ok = case dmon:is_monitored(ChPid, KS) of
              false -> ok;
              true  -> confirm_sender_death(ChPid)
-         end,
-    State.
+         end.
+
+local_sender_node_death(Node, State = #state { known_senders = KS }) ->
+    [local_sender_death(ChPid, State) || ChPid <- dmon:monitored(Node, KS)].
 
 confirm_sender_death(Pid) ->
     %% We have to deal with the possibility that we'll be promoted to