clear target in the group server when worker fails to connect to peer and reverts back to targeting

src/group/libp2p_group_gossip_server.erl

@@ -147,6 +147,17 @@ handle_cast({add_handler, Key, Handler}, State=#state{handlers=Handlers}) ->
     {noreply, State#state{handlers=maps:put(Key, Handler, Handlers)}};
 handle_cast({request_target, inbound, WorkerPid, _Ref}, State=#state{}) ->
     {noreply, stop_inbound_worker(WorkerPid, State)};
+handle_cast({clear_target, _Kind, _WorkerPid, Ref}, State=#state{workers = Workers}) ->
+    lager:debug("clearing target for worker ~p ", [_WorkerPid]),
+    %% the ref is stable across restarts, so use that as the lookup key
+    case lookup_worker(Ref, #worker.ref, State) of
+        Worker=#worker{} ->
+            NewWorkers = lists:keyreplace(Ref, #worker.ref, Workers,
+                                          Worker#worker{target=undefined}),
+            {noreply, State#state{workers=NewWorkers}};
+        _ ->
+            {noreply, State}
+    end;
 handle_cast({request_target, peerbook, WorkerPid, Ref}, State=#state{tid=TID}) ->
     LocalAddr = libp2p_swarm:pubkey_bin(TID),
     PeerList = case libp2p_swarm:peerbook(TID) of
@@ -156,6 +167,7 @@ handle_cast({request_target, peerbook, WorkerPid, Ref}, State=#state{tid=TID}) -
                        WorkerAddrs = [ libp2p_crypto:p2p_to_pubkey_bin(W#worker.target) || W <- State#state.workers, W#worker.target /= undefined, W#worker.kind /= seed ],
                        try libp2p_peerbook:random(Peerbook, [LocalAddr|WorkerAddrs]) of
                            {Addr, _} ->
+                               lager:debug("found target ~p, assigning to worker ~p",[Addr, WorkerPid]),
                                [Addr];
                            false ->
                                lager:debug("cannot get target as no peers or already connected to all peers",[]),

src/group/libp2p_group_relcast_server.erl

@@ -213,6 +213,11 @@ handle_cast({request_target, Index, WorkerPid, _WorkerRef}, State=#state{tid=TID
                                              {keys, State#state.group_keys}]}},
     libp2p_group_worker:assign_target(WorkerPid, {Target, ClientSpec}),
     {noreply, NewState};
+handle_cast({clear_target, _Kind, _WorkerPid, _Ref}, State=#state{}) ->
+    %% relcast server's target assignment is fixed, the same group worker will always get the same target
+    %% as such clear target here is deliberately a noop, unlike gossip server
+    %% clear_target commands originate from the group workers
+    {noreply, State};
 handle_cast({handle_input, _Msg}, State=#state{close_state=closing}) ->
     {noreply, State};
 handle_cast({handle_input, _Msg}, State=#state{store=Bad}) when Bad == not_started orelse

src/group/libp2p_group_server.erl

@@ -1,11 +1,15 @@
 -module(libp2p_group_server).
 
--export([request_target/4, send_result/3, send_ready/4]).
+-export([request_target/4, clear_target/4, send_result/3, send_ready/4]).
 
 -spec request_target(Server::pid(), term(), Worker::pid(), Ref::reference()) -> ok.
 request_target(Pid, Kind, WorkerPid, Ref) ->
     gen_server:cast(Pid, {request_target, Kind, WorkerPid, Ref}).
 
+-spec clear_target(Server::pid(), term(), Worker::pid(), Ref::reference()) -> ok.
+clear_target(Pid, Kind, WorkerPid, Ref) ->
+    gen_server:cast(Pid, {clear_target, Kind, WorkerPid, Ref}).
+
 -spec send_result(Server::pid(), Ref::term(), Result::any()) -> ok.
 send_result(Pid, Ref, Result) ->
     gen_server:cast(Pid, {send_result, Ref, Result}).

src/group/libp2p_group_worker.erl

@@ -274,7 +274,8 @@ connecting(info, connect_retry_timeout, Data=#data{tid=TID,
             {keep_state, stop_connect_retry_timer(Data#data{connect_pid=Pid})};
         true ->
             lager:debug("max connect retries exceeded, going back to targeting"),
-            {next_state, targeting, cancel_connect_retry_timer(Data), ?TRIGGER_TARGETING}
+            libp2p_group_server:clear_target(Data#data.server, Data#data.kind, self(), Data#data.ref),
+            {next_state, targeting, cancel_connect_retry_timer(Data#data{target = undefined}), ?TRIGGER_TARGETING}
     end;
 connecting(EventType, Msg, Data) ->
     handle_event(EventType, Msg, Data).