Ray Serve部署FastAPI WebSocket服务时超5连接触发"ASGI callable returned without sending handshake"错误的排查求助

我正在开发一个基于FastAPI的WebSocket服务器，用Ray Serve部署来处理多WebSocket连接，目标是支持最高5万并发用户。但集成Ray Serve做多进程处理后，遇到了一个棘手的问题：

前5个用户连接时一切正常，第6个用户连接时就触发了错误。

错误日志详情 #

ERROR:    ASGI callable returned without sending handshake.
(ProxyActor pid=1264842) Task exception was never retrieved
(ProxyActor pid=1264842) future: <Task finished name='Task-740' coro=<ProxyResponseGenerator._await_response_anext() done, defined at /home/user/lib/python3.11/site-packages/ray/serve/_private/proxy_response_generator.py:115> exception=RequestCancelledError('e85273c8-f582-49a4-8dbe-0a86b464ad90')>
(ProxyActor pid=1264842) Traceback (most recent call last):
(ProxyActor pid=1264842)   File "/home/user/lib/python3.11/site-packages/ray/serve/_private/proxy_response_generator.py", line 116, in _await_response_anext
(ProxyActor pid=1264842)     return await self._response.__anext__()
(ProxyActor pid=1264842)            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(ProxyActor pid=1264842)   File "/home/user/lib/python3.11/site-packages/ray/serve/handle.py", line 559, in __anext__
(ProxyActor pid=1264842)     replica_result = await self._fetch_future_result_async()
(ProxyActor pid=1264842)                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(ProxyActor pid=1264842)   File "/home/user/lib/python3.11/site-packages/ray/serve/handle.py", line 280, in _fetch_future_result_async
(ProxyActor pid=1264842)     raise RequestCancelledError(self.request_id) from None
(ProxyActor pid=1264842) ray.serve.exceptions.RequestCancelledError: Request e85273c8-f582-49a4-8dbe-0a86b464ad90 was cancelled.
(ProxyActor pid=1264842) Task exception was never retrieved
(ProxyActor pid=1264842) future: <Task finished name='Task-738' coro=<WebSocketProtocol.run_asgi() done, defined at /home/user/lib/python3.11/site-packages/uvicorn/protocols/websockets/websockets_impl.py:236> exception=RuntimeError('unable to perform operation on <TCPTransport closed=True reading=False 0x7f0420176800>; the handler is closed')>
(ProxyActor pid=1264842) Traceback (most recent call last):
(ProxyActor pid=1264842)   File "/home/user/lib/python3.11/site-packages/uvicorn/protocols/websockets/websockets_impl.py", line 258, in run_asgi
(ProxyActor pid=1264842)     self.send_500_response()
(ProxyActor pid=1264842)   File "/home/user/lib/python3.11/site-packages/uvicorn/protocols/websockets/websockets_impl.py", line 222, in send_500_response
(ProxyActor pid=1264842)     self.transport.write(b"".join(content))
(ProxyActor pid=1264842)   File "uvloop/handles/stream.pyx", line 678, in uvloop.loop.UVStream.write
(ProxyActor pid=1264842)   File "uvloop/handles/handle.pyx", line 159, in uvloop.loop.UVHandle._ensure_alive
(ProxyActor pid=1264842) RuntimeError: unable to perform operation on <TCPTransport closed=True reading=False 0x7f0420176800>; the handler is closed

相关代码 #

主应用与Ray Serve部署代码

from fastapi import FastAPI, WebSocket
from ray import serve

app = FastAPI()

@serve.deployment
@serve.ingress(app)
class FastAPIDeployment:
    def __init__(self):
        self.connections = {}

    @app.websocket("/ws")
    async def websocket_endpoint(self, websocket: WebSocket):
        await websocket.accept()
        # Connection handling logic

WebSocket管理器

我用了一个基于Redis的管理器来追踪所有WebSocket连接。

环境版本 #

• Python 3.11
• ray[serve]==2.40.0
• fastapi==0.115.5
• uvicorn==0.32.1
• asyncio==3.4.3

已尝试的操作 #

1. 验证了Redis连接的可用性，以及活跃WebSocket连接的管理逻辑，暂时没发现问题；
2. 修改了serve.start()的http_options参数，调整了连接相关配置：

serve.start(http_options={
    "host": "0.0.0.0",
    "port": 8000,
    "keep_alive_timeout": 60,
    "max_keep_alive_requests": 10
})

求助问题 #

• Ray Serve或FastAPI是否存在我没注意到的并发WebSocket连接限制？
• 该如何调试并解决这个"ASGI callable returned without sending handshake"错误？
• 这个问题会不会和Redis中WebSocket连接状态的管理方式，或者Ray Serve的状态管理逻辑有关？

备注：内容来源于stack exchange，提问作者Musab