Protect against accidental multiple concurrent connections

src/gmhc_watchdog.erl

@@ -6,6 +6,7 @@
         , ping/0 ]).
 
 -export([ note_started/2
+        , remove_restart_info/1
         , get_restart_info/0 ]).
 
 -export([ start_link/0
@@ -41,6 +42,11 @@ note_started(Id, Info) ->
 get_restart_info() ->
     gen_server:call(?MODULE, get_restart_info).
 
+remove_restart_info([]) ->
+    ok;
+remove_restart_info(IDs) ->
+    gen_server:call(?MODULE, {remove_restart_info, IDs}).
+
 start_link() ->
     gen_server:start_link({local, ?MODULE}, ?MODULE, [], []).
 
@@ -50,6 +56,9 @@ init([]) ->
 handle_call({note_started, Id, Info}, _From, S) ->
     update_pt(Id, Info),
     {reply, ok, S};
+handle_call({remove_restart_info, IDs}, _From, S) ->
+    remove_restart_info_(IDs),
+    {reply, ok, S};
 handle_call(get_restart_info, _From, S) ->
     {reply, get_pt(), S};
 handle_call({watch, Pid, Interval, N}, _From, S) ->
@@ -65,9 +74,9 @@ handle_cast(Msg, S) ->
     ?LOG_DEBUG("Unknown cast: ~p", [Msg]),
     {noreply, S}.
 
-handle_info({timeout, _, Pid}, S) ->
+handle_info({timeout, TRef, Pid}, S) ->
     ?LOG_INFO("Timeout for pid ~p", [Pid]),
-    {noreply, ping_timeout(Pid, S)};
+    {noreply, ping_timeout(Pid, TRef, S)};
 handle_info({'DOWN', _, process, Pid, _}, S) ->
     {noreply, delete_watch(Pid, S)};
 handle_info(Msg, S) ->
@@ -108,9 +117,9 @@ delete_watch(Pid, #st{services = Svcs} = S) ->
             S
     end.
 
-ping_timeout(Pid, #st{services = Svcs} = S) ->
+ping_timeout(Pid, TRef, #st{services = Svcs} = S) ->
     case maps:find(Pid, Svcs) of
-        {ok, #svc{ n = N } = Svc} ->
+        {ok, #svc{ n = N, tref = TRef} = Svc} ->
             N1 = N-1,
             if N1 =< 0 ->
                     ?LOG_ERROR("Will exit Pid ~p", [Pid]),
@@ -120,7 +129,10 @@ ping_timeout(Pid, #st{services = Svcs} = S) ->
                     Svc1 = restart_timer(Pid, Svc#svc{n = N1}),
                     S#st{services = Svcs#{Pid := Svc1}}
             end;
-        error ->
+        {ok, _} ->
+            ?LOG_DEBUG("Timeout didn't match TRef - ignoring", []),
+            S;
+        _ ->
             S
     end.
 
@@ -136,6 +148,15 @@ update_pt(Id, Info) ->
     Pt = get_pt(),
     put_pt(Pt#{Id => Info}).
 
+remove_restart_info_(IDs) ->
+    RI = get_pt(),
+    case maps:without(IDs, RI) of
+        RI ->
+            ok;
+        NewRI ->
+            put_pt(NewRI)
+    end.
+
 get_pt() ->
     persistent_term:get(pt_key(), #{}).