fix: prevent a node from discovering and re-joining the same cluster after it has (manually) left it.

apps/emqx/rebar.config

@@ -28,7 +28,7 @@
     {gproc, {git, "https://github.com/emqx/gproc", {tag, "0.9.0.1"}}},
     {cowboy, {git, "https://github.com/emqx/cowboy", {tag, "2.9.2"}}},
     {esockd, {git, "https://github.com/emqx/esockd", {tag, "5.11.1"}}},
-    {ekka, {git, "https://github.com/emqx/ekka", {tag, "0.19.1"}}},
+    {ekka, {git, "https://github.com/emqx/ekka", {tag, "0.19.2"}}},
     {gen_rpc, {git, "https://github.com/emqx/gen_rpc", {tag, "3.3.1"}}},
     {hocon, {git, "https://github.com/emqx/hocon.git", {tag, "0.42.1"}}},
     {emqx_http_lib, {git, "https://github.com/emqx/emqx_http_lib.git", {tag, "0.5.3"}}},

apps/emqx_management/src/emqx_management.app.src

@@ -2,7 +2,7 @@
 {application, emqx_management, [
     {description, "EMQX Management API and CLI"},
     % strict semver, bump manually!
-    {vsn, "5.1.0"},
+    {vsn, "5.1.1"},
     {modules, []},
     {registered, [emqx_management_sup]},
     {applications, [

apps/emqx_management/src/emqx_mgmt_cli.erl

@@ -108,6 +108,7 @@ cluster(["join", SNode]) ->
             emqx_ctl:print("Failed to join the cluster: ~0p~n", [Error])
     end;
 cluster(["leave"]) ->
+    _ = maybe_disable_autocluster(),
     case mria:leave() of
         ok ->
             emqx_ctl:print("Leave the cluster successfully.~n"),
@@ -139,12 +140,15 @@ cluster(["status"]) ->
 cluster(["status", "--json"]) ->
     Info = sort_map_list_fields(cluster_info()),
     emqx_ctl:print("~ts~n", [emqx_logger_jsonfmt:best_effort_json(Info)]);
+cluster(["discovery", "enable"]) ->
+    enable_autocluster();
 cluster(_) ->
     emqx_ctl:usage([
         {"cluster join <Node>", "Join the cluster"},
         {"cluster leave", "Leave the cluster"},
         {"cluster force-leave <Node>", "Force the node leave from cluster"},
-        {"cluster status [--json]", "Cluster status"}
+        {"cluster status [--json]", "Cluster status"},
+        {"cluster discovery enable", "Enable and run automatic cluster discovery (if configured)"}
     ]).
 
 %% sort lists for deterministic output
@@ -163,6 +167,25 @@ sort_map_list_field(Field, Map) ->
         _ -> Map
     end.
 
+enable_autocluster() ->
+    ok = ekka:enable_autocluster(),
+    _ = ekka:autocluster(emqx),
+    emqx_ctl:print("Automatic cluster discovery enabled.~n").
+
+maybe_disable_autocluster() ->
+    case ekka:autocluster_enabled() of
+        true ->
+            ok = ekka:disable_autocluster(),
+            emqx_ctl:print(
+                "Automatic cluster discovery is disabled on this node: ~p to avoid"
+                " re-joining the same cluster again, if the node is not stopped soon."
+                " To enable it run: 'emqx ctl cluster discovery enable' or restart the node.~n",
+                [node()]
+            );
+        false ->
+            ok
+    end.
+
 %%--------------------------------------------------------------------
 %% @doc Query clients
 

apps/emqx_management/test/emqx_mgmt_cli_SUITE.erl

@@ -19,6 +19,7 @@
 -compile(nowarn_export_all).
 
 -include_lib("eunit/include/eunit.hrl").
+-include_lib("common_test/include/ct.hrl").
 
 all() ->
     emqx_common_test_helpers:all(?MODULE).
@@ -31,6 +32,47 @@ init_per_suite(Config) ->
 end_per_suite(_) ->
     emqx_mgmt_api_test_util:end_suite([emqx_management, emqx_conf]).
 
+init_per_testcase(t_autocluster_leave = TC, Config) ->
+    [Core1, Core2, Core3, Repl] =
+        Nodes = [
+            t_autocluster_leave_core1,
+            t_autocluster_leave_core2,
+            t_autocluster_leave_core3,
+            t_autocluster_leave_replicant
+        ],
+
+    NodeNames = [emqx_cth_cluster:node_name(N) || N <- Nodes],
+    AppSpec = [
+        emqx,
+        {emqx_conf, #{
+            config => #{
+                cluster => #{
+                    discovery_strategy => static,
+                    static => #{seeds => NodeNames}
+                }
+            }
+        }},
+        emqx_management
+    ],
+    Cluster = emqx_cth_cluster:start(
+        [
+            {Core1, #{role => core, apps => AppSpec}},
+            {Core2, #{role => core, apps => AppSpec}},
+            {Core3, #{role => core, apps => AppSpec}},
+            {Repl, #{role => replicant, apps => AppSpec}}
+        ],
+        #{work_dir => emqx_cth_suite:work_dir(TC, Config)}
+    ),
+    [{cluster, Cluster} | Config];
+init_per_testcase(_TC, Config) ->
+    Config.
+
+end_per_testcase(_TC, Config) ->
+    case ?config(cluster, Config) of
+        undefined -> ok;
+        Cluster -> emqx_cth_cluster:stop(Cluster)
+    end.
+
 t_status(_Config) ->
     emqx_ctl:run_command([]),
     emqx_ctl:run_command(["status"]),
@@ -263,3 +305,44 @@ t_admin(_Config) ->
     %% admins passwd <Username> <Password>            # Reset dashboard user password
     %% admins del <Username>                          # Delete dashboard user
     ok.
+
+t_autocluster_leave(Config) ->
+    [Core1, Core2, Core3, Repl] = Cluster = ?config(cluster, Config),
+    %% Mria membership updates are async, makes sense to wait a little
+    timer:sleep(300),
+    ClusterView = [lists:sort(rpc:call(N, emqx, running_nodes, [])) || N <- Cluster],
+    [View1, View2, View3, View4] = ClusterView,
+    ?assertEqual(lists:sort(Cluster), View1),
+    ?assertEqual(View1, View2),
+    ?assertEqual(View1, View3),
+    ?assertEqual(View1, View4),
+
+    rpc:call(Core3, emqx_mgmt_cli, cluster, [["leave"]]),
+    timer:sleep(1000),
+    %% Replicant node may still discover and join Core3 which is now split from [Core1, Core2],
+    %% but it's expected to choose a bigger cluster of [Core1, Core2]..
+    ?assertMatch([Core3], rpc:call(Core3, emqx, running_nodes, [])),
+    ?assertEqual(undefined, rpc:call(Core1, erlang, whereis, [ekka_autocluster])),
+    ?assertEqual(lists:sort([Core1, Core2, Repl]), rpc:call(Core1, emqx, running_nodes, [])),
+    ?assertEqual(lists:sort([Core1, Core2, Repl]), rpc:call(Core2, emqx, running_nodes, [])),
+    ?assertEqual(lists:sort([Core1, Core2, Repl]), rpc:call(Repl, emqx, running_nodes, [])),
+
+    rpc:call(Repl, emqx_mgmt_cli, cluster, [["leave"]]),
+    timer:sleep(1000),
+    ?assertEqual(lists:sort([Core1, Core2]), rpc:call(Core1, emqx, running_nodes, [])),
+    ?assertEqual(lists:sort([Core1, Core2]), rpc:call(Core2, emqx, running_nodes, [])),
+
+    rpc:call(Core3, emqx_mgmt_cli, cluster, [["discovery", "enable"]]),
+    rpc:call(Repl, emqx_mgmt_cli, cluster, [["discovery", "enable"]]),
+    %% core nodes will join and restart asyncly, may need more time to re-cluster
+    ?assertEqual(
+        ok,
+        emqx_common_test_helpers:wait_for(
+            ?FUNCTION_NAME,
+            ?LINE,
+            fun() ->
+                [lists:sort(rpc:call(N, emqx, running_nodes, [])) || N <- Cluster] =:= ClusterView
+            end,
+            10_000
+        )
+    ).

changes/ce/fix-12802.en.md

@@ -0,0 +1,3 @@
+Improve cluster discovery behaviour when a node is manually removed from a cluster using 'emqx ctl cluster leave' command.
+Previously, if the configured cluster 'discovery_strategy' was not 'manual', the left node might re-discover and re-join the same cluster shortly after it left (unless it was stopped).
+After this change, 'cluster leave' command disables automatic cluster_discovery, so that the left node won't re-join the same cluster again. Cluster discovery can be re-enabled by running 'emqx ctl discovery enable` or by restarting the left node.

mix.exs

@@ -55,7 +55,7 @@ defmodule EMQXUmbrella.MixProject do
       {:cowboy, github: "emqx/cowboy", tag: "2.9.2", override: true},
       {:esockd, github: "emqx/esockd", tag: "5.11.1", override: true},
       {:rocksdb, github: "emqx/erlang-rocksdb", tag: "1.8.0-emqx-2", override: true},
-      {:ekka, github: "emqx/ekka", tag: "0.19.1", override: true},
+      {:ekka, github: "emqx/ekka", tag: "0.19.2", override: true},
       {:gen_rpc, github: "emqx/gen_rpc", tag: "3.3.1", override: true},
       {:grpc, github: "emqx/grpc-erl", tag: "0.6.12", override: true},
       {:minirest, github: "emqx/minirest", tag: "1.4.0", override: true},

rebar.config

@@ -83,7 +83,7 @@
     {cowboy, {git, "https://github.com/emqx/cowboy", {tag, "2.9.2"}}},
     {esockd, {git, "https://github.com/emqx/esockd", {tag, "5.11.1"}}},
     {rocksdb, {git, "https://github.com/emqx/erlang-rocksdb", {tag, "1.8.0-emqx-2"}}},
-    {ekka, {git, "https://github.com/emqx/ekka", {tag, "0.19.1"}}},
+    {ekka, {git, "https://github.com/emqx/ekka", {tag, "0.19.2"}}},
     {gen_rpc, {git, "https://github.com/emqx/gen_rpc", {tag, "3.3.1"}}},
     {grpc, {git, "https://github.com/emqx/grpc-erl", {tag, "0.6.12"}}},
     {minirest, {git, "https://github.com/emqx/minirest", {tag, "1.4.0"}}},