Merge pull request #5442 from rabbitmq/prevent-overlapping-partitions

Set kernel param prevent_overlapping_partitions to true

7 files changed, 10 insertions, 82 deletions

diff --git a/deps/rabbit/apps/rabbitmq_prelaunch/src/rabbit_prelaunch_conf.erl b/deps/rabbit/apps/rabbitmq_prelaunch/src/rabbit_prelaunch_conf.erl
index 9e866c4f16..2980301b7b 100644
--- a/deps/rabbit/apps/rabbitmq_prelaunch/src/rabbit_prelaunch_conf.erl
+++ b/deps/rabbit/apps/rabbitmq_prelaunch/src/rabbit_prelaunch_conf.erl
@@ -24,7 +24,7 @@ setup(Context) ->
     %% TODO: Check if directories/files are inside Mnesia dir.
 
     ok = set_default_config(),
-    ok = disable_kernel_overlapping_partitions(),
+    ok = enable_kernel_overlapping_partitions(),
 
     AdditionalConfigFiles = find_additional_config_files(Context),
     AdvancedConfigFile = find_actual_advanced_config_file(Context),
@@ -575,8 +575,7 @@ get_input_iodevice() ->
             end
     end.
 
-disable_kernel_overlapping_partitions() ->
-    %% This new "fixed" behavior seriously affects our own partition handling,
-    %% and potentially even libraries such as Aten and Ra,
-    %% so disable this to be forward-compatible with Erlang 25
-    application:set_env(kernel, prevent_overlapping_partitions, false).
+enable_kernel_overlapping_partitions() ->
+    %% Kernel parameter prevent_overlapping_partitions got introduced
+    %% in Erlang 24.3 and is set to `true` by default in Erlang 25.
+    application:set_env(kernel, prevent_overlapping_partitions, true).
diff --git a/deps/rabbit/scripts/rabbitmq-server b/deps/rabbit/scripts/rabbitmq-server
index c9612d8946..303da6f534 100755
--- a/deps/rabbit/scripts/rabbitmq-server
+++ b/deps/rabbit/scripts/rabbitmq-server
@@ -81,7 +81,7 @@ start_rabbitmq_server() {
         ${RABBITMQ_SERVER_START_ARGS} \
         -syslog logger '[]' \
         -syslog syslog_error_logger false \
-        -kernel prevent_overlapping_partitions false \
+        -kernel prevent_overlapping_partitions true \
         "$@"
 }
 
diff --git a/deps/rabbit/scripts/rabbitmq-server.bat b/deps/rabbit/scripts/rabbitmq-server.bat
index 970a101479..19a250707f 100644
--- a/deps/rabbit/scripts/rabbitmq-server.bat
+++ b/deps/rabbit/scripts/rabbitmq-server.bat
@@ -70,7 +70,7 @@ if "!RABBITMQ_ALLOW_INPUT!"=="" (
 !RABBITMQ_SERVER_START_ARGS! ^

 -syslog logger [] ^

 -syslog syslog_error_logger false ^

--kernel prevent_overlapping_partitions false ^

+-kernel prevent_overlapping_partitions true ^

 !STAR!

 

 if ERRORLEVEL 1 (

diff --git a/deps/rabbit/scripts/rabbitmq-service.bat b/deps/rabbit/scripts/rabbitmq-service.bat
index 7870417b46..779bfc1ec2 100644
--- a/deps/rabbit/scripts/rabbitmq-service.bat
+++ b/deps/rabbit/scripts/rabbitmq-service.bat
@@ -200,7 +200,7 @@ set ERLANG_SERVICE_ARGUMENTS= ^
 !RABBITMQ_DIST_ARG! ^

 -syslog logger [] ^

 -syslog syslog_error_logger false ^

--kernel prevent_overlapping_partitions false ^

+-kernel prevent_overlapping_partitions true ^

 !STARVAR!

 

 set ERLANG_SERVICE_ARGUMENTS=!ERLANG_SERVICE_ARGUMENTS:\=\\!

diff --git a/deps/rabbit/src/rabbit_ff_controller.erl b/deps/rabbit/src/rabbit_ff_controller.erl
index 7b005c5db7..f8dd874dc6 100644
--- a/deps/rabbit/src/rabbit_ff_controller.erl
+++ b/deps/rabbit/src/rabbit_ff_controller.erl
@@ -268,7 +268,7 @@ register_globally() ->
        "Feature flags: [global sync] @ ~s",
        [node()],
        #{domain => ?RMQLOG_DOMAIN_FEAT_FLAGS}),
-    ok = rabbit_node_monitor:global_sync(),
+    ok = global:sync(),
     ?LOG_DEBUG(
        "Feature flags: [global register] @ ~s",
        [node()],
diff --git a/deps/rabbit/src/rabbit_mnesia.erl b/deps/rabbit/src/rabbit_mnesia.erl
index 66362bdfc7..e802065b57 100644
--- a/deps/rabbit/src/rabbit_mnesia.erl
+++ b/deps/rabbit/src/rabbit_mnesia.erl
@@ -84,8 +84,7 @@ init() ->
     %% We intuitively expect the global name server to be synced when
     %% Mnesia is up. In fact that's not guaranteed to be the case -
     %% let's make it so.
-    ok = rabbit_node_monitor:global_sync(),
-    ok.
+    ok = global:sync().
 
 init_with_lock() ->
     {Retries, Timeout} = rabbit_peer_discovery:locking_retry_timeout(),
diff --git a/deps/rabbit/src/rabbit_node_monitor.erl b/deps/rabbit/src/rabbit_node_monitor.erl
index 447dde637f..6955307cac 100644
--- a/deps/rabbit/src/rabbit_node_monitor.erl
+++ b/deps/rabbit/src/rabbit_node_monitor.erl
@@ -19,7 +19,6 @@
 -export([notify_node_up/0, notify_joined_cluster/0, notify_left_cluster/1]).
 -export([partitions/0, partitions/1, status/1, subscribe/1]).
 -export([pause_partition_guard/0]).
--export([global_sync/0]).
 
 %% gen_server callbacks
 -export([init/1, handle_call/3, handle_cast/2, handle_info/2, terminate/2,
@@ -269,75 +268,6 @@ pause_if_all_down_guard(PreferredNodes, LastNodes, LastState) ->
     end.
 
 %%----------------------------------------------------------------------------
-%% "global" hang workaround.
-%%----------------------------------------------------------------------------
-
-%% This code works around a possible inconsistency in the "global"
-%% state, causing global:sync/0 to never return.
-%%
-%%     1. A process is spawned.
-%%     2. If after 10", global:sync() didn't return, the "global"
-%%        state is parsed.
-%%     3. If it detects that a sync is blocked for more than 10",
-%%        the process sends fake nodedown/nodeup events to the two
-%%        nodes involved (one local, one remote).
-%%     4. Both "global" instances restart their synchronisation.
-%%     5. global:sync() finally returns.
-%%
-%% FIXME: Remove this workaround, once we got rid of the change to
-%% "dist_auto_connect" and fixed the bugs uncovered.
-
-global_sync() ->
-    Pid = spawn(fun workaround_global_hang/0),
-    ok = global:sync(),
-    Pid ! global_sync_done,
-    ok.
-
-workaround_global_hang() ->
-    receive
-        global_sync_done ->
-            ok
-    after 10_000 ->
-            find_blocked_global_peers()
-    end.
-
-find_blocked_global_peers() ->
-    Snapshot1 = snapshot_global_dict(),
-    timer:sleep(10_000),
-    Snapshot2 = snapshot_global_dict(),
-    find_blocked_global_peers1(Snapshot2, Snapshot1).
-
-snapshot_global_dict() ->
-    {status, _, _, [Dict | _]} = sys:get_status(global_name_server),
-    [E || {{sync_tag_his, _}, _} = E <- Dict].
-
-find_blocked_global_peers1([{{sync_tag_his, Peer}, _} = Item | Rest],
-  OlderSnapshot) ->
-    case lists:member(Item, OlderSnapshot) of
-        true  -> unblock_global_peer(Peer);
-        false -> ok
-    end,
-    find_blocked_global_peers1(Rest, OlderSnapshot);
-find_blocked_global_peers1([], _) ->
-    ok.
-
-unblock_global_peer(PeerNode) ->
-    ThisNode = node(),
-    PeerState = rpc:call(PeerNode, sys, get_status, [global_name_server]),
-    logger:debug(
-      "Global hang workaround: global state on ~s seems inconsistent~n"
-      " * Peer global state:  ~p~n"
-      " * Local global state: ~p~n"
-      "Faking nodedown/nodeup between ~s and ~s",
-      [PeerNode, PeerState, sys:get_status(global_name_server),
-       PeerNode, ThisNode]),
-    {global_name_server, ThisNode} ! {nodedown, PeerNode},
-    {global_name_server, PeerNode} ! {nodedown, ThisNode},
-    {global_name_server, ThisNode} ! {nodeup, PeerNode},
-    {global_name_server, PeerNode} ! {nodeup, ThisNode},
-    ok.
-
-%%----------------------------------------------------------------------------
 %% gen_server callbacks
 %%----------------------------------------------------------------------------