From 48952206db05b0ecedf9aeb11da33926f823aba0 Mon Sep 17 00:00:00 2001
From: Hyu <chenhyudada@gmail.com>
Date: Wed, 15 Jul 2026 19:53:48 +0800
Subject: [PATCH] fix(lark): keep connection lookup off event loop

Move the Lark SDK synchronous connection URL lookup off the main asyncio loop, serialize reconnects, and cover the incident with a non-blocking regression test.
---
 src/langbot/pkg/platform/sources/lark.py      | 31 ++++++++++++++-
 .../platform/test_lark_ws_nonblocking.py      | 39 +++++++++++++++++++
 2 files changed, 69 insertions(+), 1 deletion(-)
 create mode 100644 tests/unit_tests/platform/test_lark_ws_nonblocking.py

diff --git a/src/langbot/pkg/platform/sources/lark.py b/src/langbot/pkg/platform/sources/lark.py
index 17ef99c68..96e40469c 100644
--- a/src/langbot/pkg/platform/sources/lark.py
+++ b/src/langbot/pkg/platform/sources/lark.py
@@ -242,6 +242,33 @@ def _lark_extract_action_form_inputs(action: typing.Any, action_value_obj: dict)
     return form_inputs
 
 
+class NonBlockingLarkWSClient(lark_oapi.ws.Client):
+    """Keep the SDK's synchronous connection lookup off LangBot's event loop.
+
+    lark-oapi performs ``requests.post`` inside its async ``_connect`` method.
+    A stalled TLS handshake therefore freezes Quart and every other adapter in
+    the process. Pre-fetch the URL in a worker thread, then let the SDK finish
+    the WebSocket setup on the main loop with the already-resolved URL.
+    """
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        self._langbot_connect_lock = asyncio.Lock()
+
+    async def _connect(self) -> None:
+        async with self._langbot_connect_lock:
+            if self._conn is not None:
+                return
+
+            conn_url = await asyncio.to_thread(self._get_conn_url)
+            original_get_conn_url = self._get_conn_url
+            self._get_conn_url = lambda: conn_url
+            try:
+                await super()._connect()
+            finally:
+                self._get_conn_url = original_get_conn_url
+
+
 class AESCipher(object):
     def __init__(self, key):
         self.bs = AES.block_size
@@ -1240,7 +1267,9 @@ class LarkAdapter(abstract_platform_adapter.AbstractMessagePlatformAdapter):
         bot_account_id = config['bot_name']
 
         domain = self._resolve_domain(config)
-        bot = lark_oapi.ws.Client(config['app_id'], config['app_secret'], event_handler=event_handler, domain=domain)
+        bot = NonBlockingLarkWSClient(
+            config['app_id'], config['app_secret'], event_handler=event_handler, domain=domain
+        )
         api_client = self.build_api_client(config)
         cipher = AESCipher(config.get('encrypt-key', ''))
         self.request_app_ticket(api_client, config)
diff --git a/tests/unit_tests/platform/test_lark_ws_nonblocking.py b/tests/unit_tests/platform/test_lark_ws_nonblocking.py
new file mode 100644
index 000000000..e2d9f34b3
--- /dev/null
+++ b/tests/unit_tests/platform/test_lark_ws_nonblocking.py
@@ -0,0 +1,39 @@
+import asyncio
+import threading
+
+import pytest
+
+from langbot.pkg.platform.sources.lark import NonBlockingLarkWSClient
+
+
+@pytest.mark.asyncio
+async def test_lark_connection_url_lookup_does_not_block_main_event_loop(monkeypatch):
+    client = NonBlockingLarkWSClient('app-id', 'app-secret')
+    lookup_started = threading.Event()
+    release_lookup = threading.Event()
+    base_connect_urls: list[str] = []
+
+    def blocking_get_conn_url() -> str:
+        lookup_started.set()
+        if not release_lookup.wait(timeout=2):
+            raise TimeoutError('test did not release the connection lookup')
+        return 'wss://example.invalid/connect?device_id=device&service_id=service'
+
+    async def fake_base_connect(self):
+        base_connect_urls.append(self._get_conn_url())
+
+    monkeypatch.setattr(client, '_get_conn_url', blocking_get_conn_url)
+    monkeypatch.setattr('lark_oapi.ws.Client._connect', fake_base_connect)
+
+    connect_task = asyncio.create_task(client._connect())
+    await asyncio.wait_for(asyncio.to_thread(lookup_started.wait, 1), timeout=1)
+
+    # If the SDK's synchronous requests.post still runs on the event-loop
+    # thread, this sleep cannot complete until release_lookup is set.
+    await asyncio.wait_for(asyncio.sleep(0.01), timeout=0.1)
+    assert not connect_task.done()
+
+    release_lookup.set()
+    await asyncio.wait_for(connect_task, timeout=1)
+
+    assert base_connect_urls == ['wss://example.invalid/connect?device_id=device&service_id=service']