class WsClient:
r"""
飞书长连接（WebSocket）事件客户端。
作为 Webhook 接收器（[create_event_app][feishu.events.receiver.create_event_app] 等）的替代方案：
无需公网回调地址，应用主动与飞书建立一条持久 WebSocket 连接，事件经该连接推送，
处理结果通过 ACK 帧回传，对标 Slack 的 Socket Mode。
连接生命周期由 [start][feishu.ws.client.WsClient.start] 驱动：握手 -> 建连 -> 收发循环，
断线后按 [ClientConfig][feishu.ws.model.ClientConfig] 自动重连。事件解析与分发完全复用
[EventDispatcher][feishu.events.dispatcher.EventDispatcher]，因此 Webhook 与长连接两种接入
方式可共用同一套处理函数；分发结果会被编码进 ACK，供卡片回调等场景返回 `{toast, card}`。
为便于测试，HTTP 客户端与 websocket 连接器均可注入，默认实现仅在真正需要时才创建/导入。
Args:
app_id: 应用 App ID。
app_secret: 应用 App Secret。
dispatcher: 事件分发器。
region: 区域标识，`feishu` 或 `lark`，默认 `feishu`。
base_url: 自定义基础地址，传入时优先于 `region`。
auto_reconnect: 断线后是否自动重连，默认 `True`。
logger: 自定义日志器，缺省使用名为 `feishu` 的日志器。
http_client: 注入的 `httpx.AsyncClient`，用于握手；为 `None` 时每次握手临时创建并关闭。
connect: 注入的 websocket 连接器；为 `None` 时懒加载 `websockets`。
sleep: 注入的休眠函数，缺省使用 [asyncio.sleep][]。
max_partial_messages: 分片重组缓冲区可保留的未完成消息数量。
Raises:
ValueError: 当 `app_id` 或 `app_secret` 为空时抛出。
飞书文档:
[事件概述](https://open.feishu.cn/document/server-docs/event-subscription-guide/overview)
Examples:
>>> from feishu.events.dispatcher import EventDispatcher
>>> dispatcher = EventDispatcher()
>>> @dispatcher.on("im.message.receive_v1")
... async def on_message(event):
... print(event.event_id)
...
>>> ws = WsClient("cli_app", "secret", dispatcher)
>>> import asyncio
>>> asyncio.run(ws.start()) # doctest: +SKIP
"""
def __init__(
self,
app_id: str,
app_secret: str,
dispatcher: EventDispatcher,
*,
region: str = "feishu",
base_url: str | None = None,
auto_reconnect: bool = True,
logger: logging.Logger | None = None,
http_client: httpx.AsyncClient | None = None,
connect: Connect | None = None,
sleep: Callable[[float], Awaitable[Any]] | None = None,
max_partial_messages: int = _MAX_PARTIAL_MESSAGES,
card_ack_timeout: float | None = _CARD_ACK_TIMEOUT_SECONDS,
background_dispatch_drain_timeout: float | None = _BACKGROUND_DISPATCH_DRAIN_TIMEOUT_SECONDS,
) -> None:
if not app_id:
raise ValueError("app_id must not be empty")
if not app_secret:
raise ValueError("app_secret must not be empty")
if max_partial_messages < 1:
raise ValueError("max_partial_messages must be positive")
if card_ack_timeout is not None and card_ack_timeout < 0:
raise ValueError("card_ack_timeout must be non-negative or None")
if background_dispatch_drain_timeout is not None and background_dispatch_drain_timeout < 0:
raise ValueError("background_dispatch_drain_timeout must be non-negative or None")
self._app_id = app_id
self._app_secret = app_secret
self._dispatcher = dispatcher
self._base_url = resolve_base_url(region, base_url)
self._auto_reconnect = auto_reconnect
self.logger = logger or logging.getLogger("feishu")
self._http_client = http_client
self._connect: Connect = connect or _default_connect
self._sleep = sleep or asyncio.sleep
self._max_partial_messages = max_partial_messages
self._card_ack_timeout = card_ack_timeout
self._background_dispatch_drain_timeout = background_dispatch_drain_timeout
self._background_dispatches: set[asyncio.Task[Any]] = set()
# Populated after the handshake: service frame field comes from the wss URL's service_id query param.
self._service_id = 0
self._ping_interval = ClientConfig().ping_interval
# Fragment reassembly buffer: message_id -> {seq: chunk}. This first version has no TTL eviction;
# if the upstream misses a fragment, the entry stays until a later timeout cleanup is added.
self._fragments: dict[str, dict[int, bytes]] = {}
# Runtime control.
self._stopped = False
self._websocket: Any = None
async def _handshake(self) -> tuple[str, ClientConfig]:
r"""
执行握手，换取 wss 连接地址与客户端配置。
向 `{base_url}/callback/ws/endpoint` POST 应用凭据（注意该端点不在 Open API 前缀下，
因此使用裸 httpx 而非 SDK 传输层）。成功后从返回的 wss URL 中解析出 `service_id`
并记录为后续出站帧的 `service` 字段。
Returns:
`(wss_url, client_config)` 二元组。
Raises:
FeishuServerError: 当握手返回 5xx 时抛出（可重试，由 [start][feishu.ws.client.WsClient.start] 退避重连）。
FeishuError: 当响应 `code` 非 0 时抛出（如鉴权/配置错误，不可重试）。
"""
client = self._http_client or httpx.AsyncClient()
try:
resp = await client.post(
f"{self._base_url}{_ENDPOINT_PATH}",
headers={"locale": "zh"},
json={"AppID": self._app_id, "AppSecret": self._app_secret},
)
# 5xx is a transient server-side failure -> surface as FeishuServerError so start() retries.
if resp.status_code >= 500:
raise FeishuServerError(resp.status_code, f"handshake failed: HTTP {resp.status_code}")
payload = resp.json()
finally:
if self._http_client is None:
await client.aclose()
code = payload.get("code", -1)
if code != 0:
raise FeishuError(code, payload.get("msg", ""), raw=payload)
data = payload.get("data") or {}
url = data["URL"]
self._service_id = _parse_service_id(url)
config = client_config_from_dict(data.get("ClientConfig") or {})
self._ping_interval = config.ping_interval
return url, config
def _ping_frame(self) -> Frame:
r"""构造一个心跳（ping）控制帧。"""
return Frame(
seq_id=0,
log_id=0,
service=self._service_id,
method=FRAME_TYPE_CONTROL,
headers=[Header("type", "ping")],
)
async def _ping_loop(self, websocket: Any, send_lock: asyncio.Lock) -> None:
r"""后台任务：按 `ping_interval` 周期发送心跳控制帧，直至被取消或连接断开。
连接在心跳发送期间断开会使 `websocket.send` 抛出异常；此处安静退出（`_serve` 会感知断开并触发
重连），避免该后台任务的异常无人取回而触发 asyncio 告警。取消（CancelledError）正常向上传播。
"""
try:
while True:
await self._sleep(self._ping_interval)
async with send_lock:
await websocket.send(encode_frame(self._ping_frame()))
except Exception: # noqa: BLE001 - a drop mid-ping is expected; exit quietly, don't leak the task exc
self.logger.debug("ws ping loop stopped", exc_info=True)
def _handle_control(self, frame: Frame) -> None:
r"""处理控制帧（心跳回复）：若回复携带 ClientConfig 则刷新心跳间隔。"""
if frame.payload:
# Keep the existing interval when a heartbeat reply has no valid ClientConfig.
with suppress(ValueError, KeyError):
self._ping_interval = client_config_from_dict(json.loads(frame.payload.decode("utf-8"))).ping_interval
async def _send_frame(self, websocket: Any, frame: Frame, send_lock: asyncio.Lock) -> None:
r"""在 `send_lock` 保护下回送一帧（ACK），保证并发任务间的发送不交错。"""
async with send_lock:
await websocket.send(encode_frame(frame))
def _reassemble(self, frame: Frame) -> bytes | None:
r"""
按 `sum`/`seq` 头重组分片帧。
`sum <= 1` 时帧自身即完整载荷，直接返回其 `payload`。否则按 `message_id` 缓存各 `seq`
分片，集齐 `sum` 个后按序拼接并清理缓冲返回；尚未集齐时返回 `None`。
Args:
frame: 数据帧。
Returns:
完整的载荷字节；分片尚未集齐时返回 `None`。
"""
total = int(frame.header("sum") or "1")
payload = frame.payload or b""
if total <= 1:
return payload
message_id = frame.header("message_id") or ""
seq = int(frame.header("seq") or "0")
chunks = self._fragments.get(message_id)
if chunks is None:
if len(self._fragments) >= self._max_partial_messages:
# Drop the oldest unfinished message (dict insertion order) to keep growth bounded.
self._fragments.pop(next(iter(self._fragments)), None)
chunks = self._fragments[message_id] = {}
chunks[seq] = payload
if len(chunks) < total:
return None
ordered = b"".join(chunks[i] for i in range(total))
del self._fragments[message_id]
return ordered
async def _serve(self, websocket: Any) -> None:
r"""
在已建立的连接上收发，直至连接关闭。
启动心跳后台任务并循环接收、解码帧。控制帧（心跳）与分片重组在收发循环内同步处理；
每条完整消息的分发与 ACK 回送则派生独立任务并发执行，从而避免某个耗时的处理函数阻塞
后续帧的接收（所有发送经 `send_lock` 串行化以保证帧不交错）。连接关闭（websockets 抛出
`ConnectionClosed`）时退出循环，并在 `finally` 中取消心跳、等待在途分发任务收尾。
Args:
websocket: 已连接的 websocket 对象。
"""
import websockets
self._websocket = websocket
send_lock = asyncio.Lock()
ping_task = asyncio.ensure_future(self._ping_loop(websocket, send_lock))
pending: set[asyncio.Task[None]] = set()
try:
while True:
try:
raw = await websocket.recv()
except websockets.ConnectionClosed:
break
if isinstance(raw, str):
raw = raw.encode("utf-8")
frame = decode_frame(raw)
if frame.method == FRAME_TYPE_CONTROL:
self._handle_control(frame)
continue
if frame.method != FRAME_TYPE_DATA:
continue
# Reassemble fragments synchronously in this loop to avoid competing writes to the fragment
# buffer. Only complete-message dispatch and ACK sending run in separate tasks.
payload = self._reassemble(frame)
if payload is None:
continue
task = asyncio.ensure_future(self._handle_frame(websocket, frame, payload, send_lock))
pending.add(task)
task.add_done_callback(pending.discard)
finally:
ping_task.cancel()
# Wait for in-flight dispatch tasks after close; their ACKs may fail on the closed connection.
if pending:
await asyncio.gather(*pending, return_exceptions=True)
await self._drain_background_dispatches()
self._websocket = None
async def _drain_background_dispatches(self) -> None:
r"""On connection shutdown, give ACK-detached card handlers a bounded chance to finish."""
tasks = {task for task in self._background_dispatches if not task.done()}
if not tasks:
return
_done, pending = await asyncio.wait(tasks, timeout=self._background_dispatch_drain_timeout)
if not pending:
return
self.logger.warning("ws background dispatch drain timed out; cancelling %d pending task(s)", len(pending))
for task in pending:
task.cancel()
await asyncio.gather(*pending, return_exceptions=True)
async def _handle_frame(self, websocket: Any, frame: Frame, payload: bytes, send_lock: asyncio.Lock) -> None:
r"""
处理一条完整入站消息：解析事件、回送 ACK、交由分发器处理（作为独立任务并发执行）。
卡片回调（[_SYNC_ACK_EVENT_TYPES][feishu.ws.client._SYNC_ACK_EVENT_TYPES]）须把处理结果
（toast / 更新后的卡片）编码进 ACK 帧，故「先分发、后 ACK」；其余事件（尤其是可能触发慢速
Agent 循环的 `im.message.receive_v1`）则「先即时 ACK、再后台分发」，避免慢处理函数迟迟不 ACK
被飞书按「至少一次」语义重投而重复处理同一条消息。
作为脱离收发循环的独立任务运行，其异常不会冒泡到 `_serve`；因此在此捕获并记录
（载荷解析失败、处理函数异常或连接已关闭导致的发送失败），避免在连接存活期间被静默吞掉。
"""
try:
event = Event.from_payload(json.loads(payload.decode("utf-8")))
if event.event_type in _SYNC_ACK_EVENT_TYPES:
if self._card_ack_timeout == 0:
await self._send_frame(websocket, _build_ack(frame, _CARD_ACK_TIMEOUT_RESULT), send_lock)
background_dispatch = asyncio.ensure_future(self._dispatcher.dispatch(event))
self._background_dispatches.add(background_dispatch)
background_dispatch.add_done_callback(self._background_dispatches.discard)
background_dispatch.add_done_callback(_log_background_dispatch_error(self.logger, event.event_id))
return
# Card actions: the ACK carries the toast / updated card, so dispatch first.
result, pending_dispatch = await self._dispatch_card_action_for_ack(event)
await self._send_frame(websocket, _build_ack(frame, result), send_lock)
if pending_dispatch is not None:
self._background_dispatches.add(pending_dispatch)
pending_dispatch.add_done_callback(self._background_dispatches.discard)
pending_dispatch.add_done_callback(_log_background_dispatch_error(self.logger, event.event_id))
else:
# ACK immediately so the broker can't redeliver while a slow handler runs, then dispatch.
await self._send_frame(websocket, _build_ack(frame, None), send_lock)
await self._dispatcher.dispatch(event)
except Exception: # noqa: BLE001 - a per-message task failure must be logged, not silently dropped
self.logger.exception("ws ack/dispatch failed for frame seq_id=%s", frame.seq_id)
async def _dispatch_card_action_for_ack(
self, event: Event
) -> tuple[dict[str, Any] | None, asyncio.Task[dict | None] | None]:
r"""Dispatch a card action, but do not let a slow handler delay the Feishu ACK indefinitely."""
task = asyncio.ensure_future(self._dispatcher.dispatch(event))
if self._card_ack_timeout is None:
return await task, None
if self._card_ack_timeout == 0:
return _CARD_ACK_TIMEOUT_RESULT, task
try:
result = await asyncio.wait_for(asyncio.shield(task), timeout=self._card_ack_timeout)
return result, None
except asyncio.TimeoutError:
return _CARD_ACK_TIMEOUT_RESULT, task
async def start(self) -> None:
r"""
启动长连接并阻塞运行，直至 [aclose][feishu.ws.client.WsClient.aclose] 被调用。
循环执行「握手 -> 建连 -> 收发」；连接断开后，若开启了自动重连且仍有重连次数，
则等待 `reconnect_interval` 秒后重试（`reconnect_count == -1` 表示无限重连）。
重连次数耗尽则停止。握手本身的瞬时失败（网络错误 / 5xx）同样按此退避重试，并计入重连预算；
鉴权 / 配置等不可重试错误则直接抛出。
Examples:
>>> ws = WsClient("cli_app", "secret", EventDispatcher())
>>> import asyncio
>>> asyncio.run(ws.start()) # doctest: +SKIP
"""
# Transient connect/serve failures are retried on the reconnect budget below, not escaped.
# websockets is optional; include its base exception only when importable so its connect /
# WS-upgrade / abnormal-close errors count too (ImportError from a missing dep still propagates).
transient: tuple[type[BaseException], ...] = (OSError, asyncio.TimeoutError)
try:
import websockets
transient += (websockets.WebSocketException,)
except ImportError:
pass
attempts = 0
config = ClientConfig() # defaults until the first successful handshake (drives early backoff)
while not self._stopped:
try:
url, config = await self._handshake()
except (httpx.RequestError, FeishuServerError) as exc:
# Transient handshake failure (network blip / 5xx): back off and retry rather than
# aborting the whole connection loop. Non-transient errors (auth/config) propagate.
if self._stopped or not self._auto_reconnect:
raise
if config.reconnect_count != -1 and attempts >= config.reconnect_count:
self.logger.warning("ws handshake retries exhausted (%d)", config.reconnect_count)
raise
attempts += 1
self.logger.warning("ws handshake failed (%s); retrying", exc)
await self._sleep(self._reconnect_delay(config))
continue
try:
async with self._connect(url) as websocket:
await self._serve(websocket)
except transient as exc:
# Transient failure opening/serving the socket after a good handshake (DNS blip,
# refused, WS-upgrade error, abnormal close). Retry on the same reconnect budget
# below rather than escaping start(); non-transient errors (bugs/auth) still propagate.
if self._stopped or not self._auto_reconnect:
raise
self.logger.warning("ws connection failed (%s); retrying", exc)
if self._stopped or not self._auto_reconnect:
return
if config.reconnect_count != -1 and attempts >= config.reconnect_count:
self.logger.warning("ws reconnect attempts exhausted (%d)", config.reconnect_count)
return
attempts += 1
await self._sleep(self._reconnect_delay(config))
def _reconnect_delay(self, config: ClientConfig) -> float:
r"""重连等待时长：在 `reconnect_interval` 之上叠加 `[0, reconnect_nonce)` 的随机抖动，避免雪崩式重连。"""
return config.reconnect_interval + random.uniform(0, config.reconnect_nonce)
async def aclose(self) -> None:
r"""
请求停止：置位停止标志使 [start][feishu.ws.client.WsClient.start] 的循环退出，并关闭活动连接。
"""
self._stopped = True
websocket = self._websocket
if websocket is not None:
await websocket.close()