test(e2e): ретраи WS-коннекта — транзиентный таймаут после deploy-stage

В run 621 после фикса OOM остался один упавший кейс:
admin_websocket_tests падал на gun:await_up(5000) — одиночный
TCP/TLS-таймаут в окно стабилизации traefik/бэкенда сразу после
рестарта контейнера. ws_connect_up/4 теперь ретраит коннект
3 раза с паузой 2s; одинаково в admins- и users-хелперах.

test/api/admins/admin_websocket_tests.erl

@@ -205,8 +205,7 @@ test_ws_connect_debug(Url, Token) ->
   ct:pal(" Host: ~s", [Host]),
   ct:pal(" Port: ~p", [Port]),
   ct:pal(" Path: ~s", [Path]),
-  {ok, ConnPid} = gun:open(Host, Port, Opts),
+  ConnPid = ws_connect_up(Host, Port, Opts, 3),
-  {ok, http}    = gun:await_up(ConnPid, 5000),
   Headers = [{<<"host">>, list_to_binary(Host ++ ":" ++ integer_to_list(Port))}],
   StreamRef = gun:ws_upgrade(ConnPid, Path, Headers),
   receive
@@ -243,6 +242,31 @@ test_ws_connect_debug(Url, Token) ->
     {error, timeout}
   end.
 
+%% @doc TCP/TLS-коннект с ретраями: сразу после deploy-stage traefik/бэкенд
+%% могут быть нестабильны, и одиночный 5s-таймаут ронял сьюит
+%% (badmatch {error,timeout}).
+-spec ws_connect_up(string(), integer(), map(), non_neg_integer()) -> pid().
+ws_connect_up(Host, Port, Opts, Retries) ->
+  Result =
+    case gun:open(Host, Port, Opts) of
+      {ok, ConnPid} ->
+        case gun:await_up(ConnPid, 5000) of
+          {ok, http} -> {ok, ConnPid};
+          {error, Reason} -> gun:close(ConnPid), {error, Reason}
+        end;
+      {error, Reason} -> {error, Reason}
+    end,
+  case Result of
+    {ok, Pid} -> Pid;
+    {error, Err} when Retries > 1 ->
+      ct:pal(" WS connect failed (~p), retry in 2s (~p attempts left)",
+        [Err, Retries - 1]),
+      timer:sleep(2000),
+      ws_connect_up(Host, Port, Opts, Retries - 1);
+    {error, Err} ->
+      error({ws_connect_failed, Err})
+  end.
+
 test_ws_send(ConnPid, Data) ->
   Msg = jsx:encode(Data),
   ct:pal(" Sending: ~s", [Msg]),

test/api/users/user_websocket_tests.erl

@@ -91,8 +91,7 @@ test_ws_connect_debug(Url, Token) ->
   ct:pal("    Host: ~s", [Host]),
   ct:pal("    Port: ~p", [Port]),
   ct:pal("    Path: ~s", [Path]),
-  {ok, ConnPid} = gun:open(Host, Port, Opts),
+  ConnPid = ws_connect_up(Host, Port, Opts, 3),
-  {ok, http}    = gun:await_up(ConnPid, 5000),
   Headers = [{<<"host">>, list_to_binary(Host ++ ":" ++ integer_to_list(Port))}],
   StreamRef = gun:ws_upgrade(ConnPid, Path, Headers),
   receive
@@ -129,6 +128,31 @@ test_ws_connect_debug(Url, Token) ->
     {error, timeout}
   end.
 
+%% @doc TCP/TLS-коннект с ретраями: сразу после deploy-stage traefik/бэкенд
+%% могут быть нестабильны, и одиночный 5s-таймаут ронял сьюит
+%% (badmatch {error,timeout}).
+-spec ws_connect_up(string(), integer(), map(), non_neg_integer()) -> pid().
+ws_connect_up(Host, Port, Opts, Retries) ->
+  Result =
+    case gun:open(Host, Port, Opts) of
+      {ok, ConnPid} ->
+        case gun:await_up(ConnPid, 5000) of
+          {ok, http} -> {ok, ConnPid};
+          {error, Reason} -> gun:close(ConnPid), {error, Reason}
+        end;
+      {error, Reason} -> {error, Reason}
+    end,
+  case Result of
+    {ok, Pid} -> Pid;
+    {error, Err} when Retries > 1 ->
+      ct:pal("    WS connect failed (~p), retry in 2s (~p attempts left)",
+        [Err, Retries - 1]),
+      timer:sleep(2000),
+      ws_connect_up(Host, Port, Opts, Retries - 1);
+    {error, Err} ->
+      error({ws_connect_failed, Err})
+  end.
+
 test_ws_send(ConnPid, Data) ->
   Msg = jsx:encode(Data),
   ct:pal("    Sending: ~s", [Msg]),