Merge bug26265

5 files changed, 71 insertions, 22 deletions

diff --git a/packaging/RPMS/Fedora/rabbitmq-server.spec b/packaging/RPMS/Fedora/rabbitmq-server.spec
index f363bce6..c77d7e9d 100644
--- a/packaging/RPMS/Fedora/rabbitmq-server.spec
+++ b/packaging/RPMS/Fedora/rabbitmq-server.spec
@@ -130,6 +130,9 @@ done
 rm -rf %{buildroot}
 
 %changelog
+* Wed Nov 26 2014 simon@rabbitmq.com 3.4.2-1
+- New Upstream Release
+
 * Wed Oct 29 2014 simon@rabbitmq.com 3.4.1-1
 - New Upstream Release
 
diff --git a/packaging/debs/Debian/debian/changelog b/packaging/debs/Debian/debian/changelog
index a47caa0e..5e3744fd 100644
--- a/packaging/debs/Debian/debian/changelog
+++ b/packaging/debs/Debian/debian/changelog
@@ -1,3 +1,9 @@
+rabbitmq-server (3.4.2-1) unstable; urgency=low
+
+  * New Upstream Release
+
+ -- Simon MacMullen <simon@rabbitmq.com>  Wed, 26 Nov 2014 12:11:12 +0000
+
 rabbitmq-server (3.4.1-1) unstable; urgency=low
 
   * New Upstream Release
diff --git a/src/rabbit_amqqueue.erl b/src/rabbit_amqqueue.erl
index 2e5a5e8c..0dfca854 100644
--- a/src/rabbit_amqqueue.erl
+++ b/src/rabbit_amqqueue.erl
@@ -466,6 +466,7 @@ check_arguments(QueueName, Args, Validators) ->
 declare_args() ->
     [{<<"x-expires">>,                 fun check_expires_arg/2},
      {<<"x-message-ttl">>,             fun check_message_ttl_arg/2},
+     {<<"x-dead-letter-exchange">>,    fun check_dlxname_arg/2},
      {<<"x-dead-letter-routing-key">>, fun check_dlxrk_arg/2},
      {<<"x-max-length">>,              fun check_non_neg_int_arg/2},
      {<<"x-max-length-bytes">>,        fun check_non_neg_int_arg/2}].
@@ -502,6 +503,11 @@ check_message_ttl_arg({Type, Val}, Args) ->
         Error -> Error
     end.
 
+%% Note that the validity of x-dead-letter-exchange is already verified
+%% by rabbit_channel's queue.declare handler.
+check_dlxname_arg({longstr, _}, _) -> ok;
+check_dlxname_arg({Type,    _}, _) -> {error, {unacceptable_type, Type}}.
+
 check_dlxrk_arg({longstr, _}, Args) ->
     case rabbit_misc:table_lookup(Args, <<"x-dead-letter-exchange">>) of
         undefined -> {error, routing_key_but_no_dlx_defined};
diff --git a/src/rabbit_mnesia.erl b/src/rabbit_mnesia.erl
index fa51dd70..80fbcd68 100644
--- a/src/rabbit_mnesia.erl
+++ b/src/rabbit_mnesia.erl
@@ -174,18 +174,25 @@ join_cluster(DiscoveryNode, NodeType) ->
     {ClusterNodes, _, _} = discover_cluster([DiscoveryNode]),
     case me_in_nodes(ClusterNodes) of
         false ->
-            %% reset the node. this simplifies things and it will be needed in
-            %% this case - we're joining a new cluster with new nodes which
-            %% are not in synch with the current node. I also lifts the burden
-            %% of reseting the node from the user.
-            reset_gracefully(),
-
-            %% Join the cluster
-            rabbit_log:info("Clustering with ~p as ~p node~n",
-                            [ClusterNodes, NodeType]),
-            ok = init_db_with_mnesia(ClusterNodes, NodeType, true, true),
-            rabbit_node_monitor:notify_joined_cluster(),
-            ok;
+            case check_cluster_consistency(DiscoveryNode, false) of
+                {ok, _} ->
+                    %% reset the node. this simplifies things and it
+                    %% will be needed in this case - we're joining a new
+                    %% cluster with new nodes which are not in synch
+                    %% with the current node. It also lifts the burden
+                    %% of resetting the node from the user.
+                    reset_gracefully(),
+
+                    %% Join the cluster
+                    rabbit_log:info("Clustering with ~p as ~p node~n",
+                                    [ClusterNodes, NodeType]),
+                    ok = init_db_with_mnesia(ClusterNodes, NodeType,
+                                             true, true),
+                    rabbit_node_monitor:notify_joined_cluster(),
+                    ok;
+                {error, Reason} ->
+                    {error, Reason}
+            end;
         true ->
             rabbit_log:info("Already member of cluster: ~p~n", [ClusterNodes]),
             {ok, already_member}
@@ -545,7 +552,7 @@ maybe_force_load() ->
 check_cluster_consistency() ->
     %% We want to find 0 or 1 consistent nodes.
     case lists:foldl(
-           fun (Node,  {error, _})    -> check_cluster_consistency(Node);
+           fun (Node,  {error, _})    -> check_cluster_consistency(Node, true);
                (_Node, {ok, Status})  -> {ok, Status}
            end, {error, not_found}, nodes_excl_me(cluster_nodes(all)))
     of
@@ -575,17 +582,22 @@ check_cluster_consistency() ->
             throw(E)
     end.
 
-check_cluster_consistency(Node) ->
+check_cluster_consistency(Node, CheckNodesConsistency) ->
     case rpc:call(Node, rabbit_mnesia, node_info, []) of
         {badrpc, _Reason} ->
             {error, not_found};
         {_OTP, _Rabbit, {error, _}} ->
             {error, not_found};
-        {OTP, Rabbit, {ok, Status}} ->
+        {OTP, Rabbit, {ok, Status}} when CheckNodesConsistency ->
             case check_consistency(OTP, Rabbit, Node, Status) of
                 {error, _} = E -> E;
                 {ok, Res}      -> {ok, Res}
             end;
+        {OTP, Rabbit, {ok, Status}} ->
+            case check_consistency(OTP, Rabbit) of
+                {error, _} = E -> E;
+                ok             -> {ok, Status}
+            end;
         {_OTP, Rabbit, _Hash, _Status} ->
             %% delegate hash checking implies version mismatch
             version_error("Rabbit", rabbit_misc:version(), Rabbit)
diff --git a/src/rabbit_node_monitor.erl b/src/rabbit_node_monitor.erl
index a4ae2a5e..82a7a89b 100644
--- a/src/rabbit_node_monitor.erl
+++ b/src/rabbit_node_monitor.erl
@@ -342,7 +342,7 @@ handle_cast({check_partial_partition, _Node, _Reporter,
     {noreply, State};
 
 handle_cast({partial_partition, NotReallyDown, Proxy, MyGUID},
-            State = #state{guid = MyGUID, partitions = Partitions}) ->
+            State = #state{guid = MyGUID}) ->
     FmtBase = "Partial partition detected:~n"
         " * We saw DOWN from ~s~n"
         " * We can still see ~s which can see ~s~n",
@@ -359,17 +359,22 @@ handle_cast({partial_partition, NotReallyDown, Proxy, MyGUID},
             rabbit_log:error(
               FmtBase ++ "We will therefore intentionally disconnect from ~s~n",
               ArgsBase ++ [Proxy]),
-            erlang:disconnect_node(Proxy),
-            %% In the event of explicitly disconnecting from a node,
-            %% sometimes Mnesia does not log that we were partitioned
-            %% - so note it here.
-            Partitions1 = lists:usort([Proxy | Partitions]),
-            {noreply, State#state{partitions = Partitions1}}
+            cast(Proxy, {partial_partition_disconnect, node()}),
+            disconnect(Proxy),
+            {noreply, State}
     end;
 
 handle_cast({partial_partition, _GUID, _Reporter, _Proxy}, State) ->
     {noreply, State};
 
+%% Sometimes it appears the Erlang VM does not give us nodedown
+%% messages reliably when another node disconnects from us. Therefore
+%% we are told just before the disconnection so we can reciprocate.
+handle_cast({partial_partition_disconnect, Other}, State) ->
+    rabbit_log:error("Partial partition disconnect from ~s~n", [Other]),
+    disconnect(Other),
+    {noreply, State};
+
 %% Note: when updating the status file, we can't simply write the
 %% mnesia information since the message can (and will) overtake the
 %% mnesia propagation.
@@ -448,6 +453,10 @@ handle_info({nodedown, Node, Info}, State = #state{guid       = MyGUID,
     end,
     {noreply, handle_dead_node(Node, State)};
 
+handle_info({nodeup, Node, _Info}, State) ->
+    rabbit_log:info("node ~p up~n", [Node]),
+    {noreply, State};
+
 handle_info({mnesia_system_event,
              {inconsistent_database, running_partitioned_network, Node}},
             State = #state{partitions = Partitions,
@@ -650,6 +659,19 @@ del_node(Node, Nodes) -> Nodes -- [Node].
 
 cast(Node, Msg) -> gen_server:cast({?SERVER, Node}, Msg).
 
+%% When we call this, it's because we want to force Mnesia to detect a
+%% partition. But if we just disconnect_node/1 then Mnesia won't
+%% detect a very short partition. So we want to force a slightly
+%% longer disconnect. Unfortunately we don't have a way to blacklist
+%% individual nodes; the best we can do is turn off auto-connect
+%% altogether.
+disconnect(Node) ->
+    application:set_env(kernel, dist_auto_connect, never),
+    erlang:disconnect_node(Node),
+    timer:sleep(1000),
+    application:unset_env(kernel, dist_auto_connect),
+    ok.
+
 %%--------------------------------------------------------------------
 
 %% mnesia:system_info(db_nodes) (and hence