Fix empty recv() (zero-length WSS frame) JSONDecodeError in _sub_routing_loop (OPS-3597)

CHANGELOG.md

@@ -1,5 +1,9 @@
 # CHANGELOG
 
+## [3.8.3] - 2026-06-24
+
+- [Fixed] `_sub_routing_loop` (and thus all websocket subscriptions) now tolerates empty `recv()` results from the WSS (zero-length frames), which previously produced `orjson.JSONDecodeError: Input is a zero-length, empty document` and an ERROR log that killed the router thread. Empty (and any other) `orjson` decode failures on the wire are now routed through `TRANSIENT_WS_ERRORS` so they log at WARNING ("WSS connection reset or closed by peer") and trigger the existing reconnect path. (OPS-3597)
+
 ## [3.8.2] - 2026-06-23
 
 - [Fixed] `_new_conn` now closes the previous WSS connection before opening a new one, instead of overwriting `self._conn` and abandoning the old socket. On reconnect (when `_sub_routing_loop` sets `wss_conn_halted` and calls `_new_conn`) the orphaned socket lingered half-open on the server — no TCP FIN was ever sent — so the gateway kept fanning subscription events into a dead connection, accumulating an unbounded send buffer/mailbox until it OOMed. A new private `_close_conn` helper best-effort closes `self._conn` and clears the handle; it runs at the top of `_new_conn` so the server receives a FIN and tears down the stale subscription, and is reused by `close()` to drop the duplicated teardown.

pygqlc/GraphQLClient.py

@@ -37,6 +37,7 @@
     BrokenPipeError,
     ConnectionAbortedError,
     websocket.WebSocketConnectionClosedException,
+    orjson.JSONDecodeError,
 )
 
 # * Custom Exception class for GraphQL responses

pygqlc/__version__.py

@@ -1 +1 @@
-__version__ = "3.8.2"
+__version__ = "3.8.3"

tests/pygqlc/gql_client/test_subscribe.py

@@ -366,3 +366,28 @@ def make_conn(*_args, **_kwargs):
     )
     gql.closing = True
     Singleton._instances.pop(GraphQLClient, None)
+
+
+def test_sub_routing_loop_empty_recv_logged_as_warning(routing_client):
+    """OPS-3597: empty recv() result (zero-length WSS frame -> orjson.loads(b'')) must
+    not crash the routing loop with JSONDecodeError; treat as transient close (like
+    ConnectionResetError), log at WARNING, set halted, and trigger reconnect path."""
+    gql = routing_client
+    gql._conn.recv.return_value = b""
+
+    with (
+        patch.object(gql, "_new_conn", side_effect=_stop_loop_on(gql)) as new_conn,
+        _capture_logs() as records,
+    ):
+        _run_routing_loop(gql)
+
+    assert gql.wss_conn_halted is True
+    assert new_conn.called
+    assert any(
+        level == LogLevel.WARNING and "reset or closed by peer" in msg
+        for level, msg in records
+    )
+    assert not any(
+        level == LogLevel.ERROR and "Some error trying to receive WSS" in msg
+        for level, msg in records
+    )