fix(wecom): degrade stream replies to proactive msg when callback req_id stale

Turns that outlive the WeCom callback req_id window (e.g. approval
round-trips >5min) had their delta/finish frames silently rejected by
the gateway, so the user saw 'clicked approve, nothing happened' while
the agent kept working. Track req_id recv time; past a 240s TTL, stream
updates fall back to proactive aibot_send_msg. Also log non-zero
errcode ack frames so future gateway rejections are visible.
This commit is contained in:
openbot committed 2026-10-24 09:18:19 +08:00
1 parent e07f37fbb4
commit 59c44e94ca
3 files changed
+44 -10

No files matched your search

+28 -6
View File
@@ -57,9 +57,9 @@ class WecomAibotChannel:
self._ws = None # current socket (set by _connect_and_serve) self._ws = None # current socket (set by _connect_and_serve)
self.connection_state = "idle" self.connection_state = "idle"
self.connection_error: str = "" self.connection_error: str = ""
# chat -> streaming state; chat -> last callback req_id (respond must reuse it) # chat -> streaming state; chat -> (last callback req_id, recv time)
self._streams: dict[str, dict] = {} self._streams: dict[str, dict] = {}
self._req_ids: dict[str, str] = {} self._req_ids: dict[str, tuple[str, float]] = {}
self._stream_lock = threading.Lock() self._stream_lock = threading.Lock()
# media upload acks: req_id -> {event, msg} # media upload acks: req_id -> {event, msg}
self._pending_acks: dict[str, dict] = {} self._pending_acks: dict[str, dict] = {}
@@ -165,6 +165,8 @@ class WecomAibotChannel:
cmd = msg.get("cmd", "") cmd = msg.get("cmd", "")
if msg.get("errcode") is not None and not cmd: if msg.get("errcode") is not None and not cmd:
# ack/pong frame — route to a waiting media-upload _request() if any # ack/pong frame — route to a waiting media-upload _request() if any
if msg.get("errcode") != 0:
self._logger.warning(f"[wecom] gateway rejected frame: errcode={msg.get('errcode')} errmsg={msg.get('errmsg')} req_id={(msg.get('headers') or {}).get('req_id', '')}")
self._deliver_ack(msg) self._deliver_ack(msg)
return return
req_id = (msg.get("headers") or {}).get("req_id", "") req_id = (msg.get("headers") or {}).get("req_id", "")
@@ -216,7 +218,7 @@ class WecomAibotChannel:
# remember req_id for streaming replies keyed by chat # remember req_id for streaming replies keyed by chat
if req_id: if req_id:
with self._stream_lock: with self._stream_lock:
self._req_ids[chat_id] = req_id self._req_ids[chat_id] = (req_id, time.time())
text = clean_control_chars(text, max_len=100000).strip() if text else "" text = clean_control_chars(text, max_len=100000).strip() if text else ""
# group chat: only respond when @mentioned is handled by WeCom itself # group chat: only respond when @mentioned is handled by WeCom itself
if not text and not images and not files: if not text and not images and not files:
@@ -469,9 +471,21 @@ class WecomAibotChannel:
for chunk in overflow: for chunk in overflow:
self.reply(chat_id, chunk) self.reply(chat_id, chunk)
# The inbound callback's req_id is only accepted by the WeCom gateway for a
# limited window (~minutes). After an approval round-trip the turn can
# outlive that window; frames reusing the stale req_id are silently
# rejected, so the user sees "clicked approve, nothing happened". Past the
# TTL, stream updates degrade to proactive aibot_send_msg messages.
_CALLBACK_REQ_TTL_S = 240.0
def _reply_stream(self, chat_id: str, st: dict, text: str, finished: bool) -> None: def _reply_stream(self, chat_id: str, st: dict, text: str, finished: bool) -> None:
# official: reuse the CALLBACK's req_id in headers; stream.id identifies the bubble entry = self._req_ids.get(chat_id)
req_id = self._req_ids.get(chat_id, "") req_id = ""
stale = True
if entry:
req_id, ts = entry
stale = (time.time() - ts) > self._CALLBACK_REQ_TTL_S
if req_id and not stale:
self._send( self._send(
{ {
"cmd": "aibot_respond_msg", "cmd": "aibot_respond_msg",
@@ -484,8 +498,16 @@ class WecomAibotChannel:
) )
if finished: if finished:
st["sent"] = True st["sent"] = True
# proactive follow-ups after the stream need aibot_send_msg with self._stream_lock:
self._req_ids.pop(chat_id, None) self._req_ids.pop(chat_id, None)
return
# no live callback req_id: fall back to a proactive message
self.reply(chat_id, text) if text else None
if finished:
st["sent"] = True
with self._stream_lock:
self._req_ids.pop(chat_id, None)
self._logger.info(f"[wecom] stream finished via proactive message (callback req_id stale or absent) chat={chat_id}")
def _send(self, obj: dict) -> None: def _send(self, obj: dict) -> None:
raw = json.dumps(obj, ensure_ascii=False) raw = json.dumps(obj, ensure_ascii=False)
+14 -2
View File
@@ -77,7 +77,7 @@ def test_wecom_dispatch_text_and_permission():
"from": {"userid": "u1"}, "text": {"content": "hello\u200b"}}} "from": {"userid": "u1"}, "text": {"content": "hello\u200b"}}}
ch._dispatch(frame, ws) ch._dispatch(frame, ws)
assert len(got) == 1 and got[0].text == "hello" and got[0].chat_id == "c1" assert len(got) == 1 and got[0].text == "hello" and got[0].chat_id == "c1"
assert ch._req_ids["c1"] == "r1" # captured for the streaming reply assert ch._req_ids["c1"][0] == "r1" # captured for the streaming reply
def test_wecom_allowed_users_drops_others(): def test_wecom_allowed_users_drops_others():
@@ -91,7 +91,8 @@ def test_wecom_allowed_users_drops_others():
def test_wecom_streaming_delta_coalesce_and_final(): def test_wecom_streaming_delta_coalesce_and_final():
ch, ws = _make_wecom() ch, ws = _make_wecom()
ch._ws = ws # pretend connected (stream sends go through the socket) ch._ws = ws # pretend connected (stream sends go through the socket)
ch._req_ids["c1"] = "req-9" # as set by an inbound callback ch._req_ids["c1"] = ("req-9", __import__("time").time()) # fresh inbound callback
ch._req_ids["stale"] = ("req-old", __import__("time").time() - 600) # expired callback
ch.push_event("c1", "delta", {"text": "hello "}) ch.push_event("c1", "delta", {"text": "hello "})
ch.push_event("c1", "delta", {"text": "world"}) ch.push_event("c1", "delta", {"text": "world"})
ch.push_event("c1", "final", {"text": "hello world"}) ch.push_event("c1", "final", {"text": "hello world"})
@@ -171,3 +172,14 @@ def test_weixin_no_edit_deltas_dropped_final_sent():
assert not http.posts assert not http.posts
ch.push_event("u", "final", {"text": "full text"}) ch.push_event("u", "final", {"text": "full text"})
assert http.posts[-1][1]["text"]["content"] == "full text" assert http.posts[-1][1]["text"]["content"] == "full text"
def test_wecom_stream_falls_back_when_callback_req_stale():
ch, ws = _make_wecom()
ch._ws = ws
ch._req_ids["c1"] = ("req-old", __import__("time").time() - 600) # beyond TTL
ch.push_event("c1", "final", {"text": "late answer"})
respond = [m for m in ws.sent if m["cmd"] == "aibot_respond_msg"]
send = [m for m in ws.sent if m["cmd"] == "aibot_send_msg"]
assert respond == [] # stale req_id must NOT be reused
assert send and "late answer" in json.dumps(send[-1])
+2 -2
View File
@@ -37,7 +37,7 @@ def _channel() -> tuple[WecomAibotChannel, MockWS]:
def test_small_final_single_stream_frame(): def test_small_final_single_stream_frame():
ch, ws = _channel() ch, ws = _channel()
ch._req_ids["c1"] = "r1" ch._req_ids["c1"] = ("r1", __import__("time").time())
ch._stream_close("c1", "hello") ch._stream_close("c1", "hello")
streams = [m for m in ws.sent if m.get("cmd") == "aibot_respond_msg"] streams = [m for m in ws.sent if m.get("cmd") == "aibot_respond_msg"]
assert len(streams) == 1 assert len(streams) == 1
@@ -47,7 +47,7 @@ def test_small_final_single_stream_frame():
def test_large_final_splits_into_stream_plus_markdown(): def test_large_final_splits_into_stream_plus_markdown():
ch, ws = _channel() ch, ws = _channel()
ch._req_ids["c1"] = "r1" ch._req_ids["c1"] = ("r1", __import__("time").time())
big = "字" * 12000 # 3 bytes/char -> 36000 bytes > 19000 limit big = "字" * 12000 # 3 bytes/char -> 36000 bytes > 19000 limit
ch._stream_close("c1", big) ch._stream_close("c1", big)
streams = [m for m in ws.sent if m.get("cmd") == "aibot_respond_msg"] streams = [m for m in ws.sent if m.get("cmd") == "aibot_respond_msg"]