feat(cluster-link): preserve replication actor state in pdict

apps/emqx_cluster_link/src/emqx_cluster_link.erl

@@ -86,14 +86,15 @@ should_route_to_external_dests(_Msg) ->
 on_message_publish(
     #message{topic = <<?ROUTE_TOPIC_PREFIX, ClusterName/binary>>, payload = Payload} = Msg
 ) ->
-    _ =
     case emqx_cluster_link_mqtt:decode_route_op(Payload) of
-            {actor_init, InitInfoMap} ->
+        {actor_init, Actor, InitInfo} ->
-                actor_init(ClusterName, emqx_message:get_header(properties, Msg), InitInfoMap);
+            Result = actor_init(ClusterName, Actor, InitInfo),
-            {route_updates, #{actor := Actor, incarnation := Incr}, RouteOps} ->
+            _ = actor_init_ack(Actor, Result, Msg),
-                update_routes(ClusterName, Actor, Incr, RouteOps);
+            ok;
-            {heartbeat, #{actor := Actor, incarnation := Incr}} ->
+        {route_updates, #{actor := Actor}, RouteOps} ->
-                actor_heartbeat(ClusterName, Actor, Incr)
+            ok = update_routes(ClusterName, Actor, RouteOps);
+        {heartbeat, #{actor := Actor}} ->
+            ok = actor_heartbeat(ClusterName, Actor)
     end,
     {stop, []};
 on_message_publish(#message{topic = <<?MSG_TOPIC_PREFIX, ClusterName/binary>>, payload = Payload}) ->
@@ -117,6 +118,9 @@ delete_hook() ->
 %% Internal functions
 %%--------------------------------------------------------------------
 
+-define(PD_EXTROUTER_ACTOR, '$clink_extrouter_actor').
+-define(PD_EXTROUTER_ACTOR_STATE, '$clink_extrouter_actor_state').
+
 maybe_push_route_op(Op, Topic, RouteID) ->
     maybe_push_route_op(Op, Topic, RouteID, push).
 
@@ -143,24 +147,18 @@ topic_intersect_any(_Topic, []) ->
 
 actor_init(
     ClusterName,
-    #{'Correlation-Data' := ReqId, 'Response-Topic' := RespTopic},
+    #{actor := Actor, incarnation := Incr},
     #{
-        actor := Actor,
+        target_cluster := TargetCluster,
-        incarnation := Incr,
-        cluster := TargetCluster,
         proto_ver := _
     }
 ) ->
-    Res =
     case emqx_cluster_link_config:link(ClusterName) of
         undefined ->
-                ?SLOG(
+            ?SLOG(error, #{
-                    error,
-                    #{
                 msg => "init_link_request_from_unknown_cluster",
                 link_name => ClusterName
-                    }
+            }),
-                ),
             %% Avoid atom error reasons, since they can be sent to the remote cluster,
             %% which will use safe binary_to_term decoding
             %% TODO: add error details?
@@ -172,7 +170,11 @@ actor_init(
             case MyClusterName of
                 TargetCluster ->
                     Env = #{timestamp => erlang:system_time(millisecond)},
-                        emqx_cluster_link_extrouter:actor_init(ClusterName, Actor, Incr, Env);
+                    {ok, ActorSt} = emqx_cluster_link_extrouter:actor_init(
+                        ClusterName, Actor, Incr, Env
+                    ),
+                    undefined = set_actor_state(ClusterName, Actor, ActorSt),
+                    ok;
                 _ ->
                     %% The remote cluster uses a different name to refer to this cluster
                     ?SLOG(error, #{
@@ -186,27 +188,38 @@ actor_init(
                     }),
                     {error, <<"bad_remote_cluster_link_name">>}
             end
-        end,
+    end.
-    _ = actor_init_ack(Actor, Res, ReqId, RespTopic),
-    {stop, []}.
 
-actor_init_ack(Actor, Res, ReqId, RespTopic) ->
+actor_init_ack(#{actor := Actor}, Res, MsgIn) ->
-    RespMsg = emqx_cluster_link_mqtt:actor_init_ack_resp_msg(Actor, Res, ReqId, RespTopic),
+    RespMsg = emqx_cluster_link_mqtt:actor_init_ack_resp_msg(Actor, Res, MsgIn),
     emqx_broker:publish(RespMsg).
 
-update_routes(ClusterName, Actor, Incarnation, RouteOps) ->
+update_routes(ClusterName, Actor, RouteOps) ->
-    ActorState = emqx_cluster_link_extrouter:actor_state(ClusterName, Actor, Incarnation),
+    ActorSt = get_actor_state(ClusterName, Actor),
     lists:foreach(
         fun(RouteOp) ->
-            emqx_cluster_link_extrouter:actor_apply_operation(RouteOp, ActorState)
+            emqx_cluster_link_extrouter:actor_apply_operation(RouteOp, ActorSt)
         end,
         RouteOps
     ).
 
-actor_heartbeat(ClusterName, Actor, Incarnation) ->
+actor_heartbeat(ClusterName, Actor) ->
     Env = #{timestamp => erlang:system_time(millisecond)},
-    ActorState = emqx_cluster_link_extrouter:actor_state(ClusterName, Actor, Incarnation),
+    ActorSt0 = get_actor_state(ClusterName, Actor),
-    _State = emqx_cluster_link_extrouter:actor_apply_operation(heartbeat, ActorState, Env).
+    ActorSt = emqx_cluster_link_extrouter:actor_apply_operation(heartbeat, ActorSt0, Env),
+    _ = update_actor_state(ActorSt),
+    ok.
+
+get_actor_state(ClusterName, Actor) ->
+    {ClusterName, Actor} = erlang:get(?PD_EXTROUTER_ACTOR),
+    erlang:get(?PD_EXTROUTER_ACTOR_STATE).
+
+set_actor_state(ClusterName, Actor, ActorSt) ->
+    undefined = erlang:put(?PD_EXTROUTER_ACTOR, {ClusterName, Actor}),
+    update_actor_state(ActorSt).
+
+update_actor_state(ActorSt) ->
+    erlang:put(?PD_EXTROUTER_ACTOR_STATE, ActorSt).
 
 %% let it crash if extra is not a map,
 %% we don't expect the message to be forwarded from an older EMQX release,

apps/emqx_cluster_link/src/emqx_cluster_link_mqtt.erl

@@ -35,7 +35,7 @@
 
 -export([
     publish_actor_init_sync/6,
-    actor_init_ack_resp_msg/4,
+    actor_init_ack_resp_msg/3,
     publish_route_sync/4,
     publish_heartbeat/3,
     encode_field/2
@@ -277,13 +277,17 @@ publish_actor_init_sync(ClientPid, ReqId, RespTopic, TargetCluster, Actor, Incar
     },
     emqtt:publish(ClientPid, ?ROUTE_TOPIC, Properties, ?ENCODE(Payload), [{qos, ?QOS_1}]).
 
-actor_init_ack_resp_msg(Actor, InitRes, ReqId, RespTopic) ->
+actor_init_ack_resp_msg(Actor, InitRes, MsgIn) ->
     Payload = #{
         ?F_OPERATION => ?OP_ACTOR_INIT_ACK,
         ?F_PROTO_VER => ?PROTO_VER,
         ?F_ACTOR => Actor
     },
     Payload1 = with_res_and_bootstrap(Payload, InitRes),
+    #{
+        'Response-Topic' := RespTopic,
+        'Correlation-Data' := ReqId
+    } = emqx_message:get_header(properties, MsgIn),
     emqx_message:make(
         undefined,
         ?QOS_1,
@@ -334,12 +338,11 @@ decode_route_op1(#{
     ?F_ACTOR := Actor,
     ?F_INCARNATION := Incr
 }) ->
-    {actor_init, #{
+    Info = #{
-        actor => Actor,
+        target_cluster => TargetCluster,
-        incarnation => Incr,
-        cluster => TargetCluster,
         proto_ver => ProtoVer
-    }};
+    },
+    {actor_init, #{actor => Actor, incarnation => Incr}, Info};
 decode_route_op1(#{
     ?F_OPERATION := ?OP_ROUTE,
     ?F_ACTOR := Actor,