Servercrash with autobahn testcase 7.1.6 on trio

[Florian Apolloner]

The CI currently kills the server when running testcase 7.1.6:

[2023-12-29 12:12:57 +0100] [13470] [ERROR] Error in ASGI Framework
Traceback (most recent call last):
  File "/home/florian/sources/hypercorn/src/hypercorn/trio/task_group.py", line 26, in _handle
    await app(scope, receive, send, sync_spawn, call_soon)
  File "/home/florian/sources/hypercorn/src/hypercorn/app_wrappers.py", line 34, in __call__
    await self.app(scope, receive, send)
  File "/home/florian/sources/hypercorn/server.py", line 9, in app
    await send({
  File "/home/florian/sources/hypercorn/src/hypercorn/protocol/ws_stream.py", line 283, in app_send
    await self._send_wsproto_event(event)
  File "/home/florian/sources/hypercorn/src/hypercorn/protocol/ws_stream.py", line 336, in _send_wsproto_event
    data = self.connection.send(event)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/florian/sources/hypercorn/.venv/lib/python3.12/site-packages/wsproto/connection.py", line 107, in send
    raise LocalProtocolError(
wsproto.utilities.LocalProtocolError: Event TextMessage(data='Hello World!', frame_finished=True, message_finished=True) cannot be sent in state ConnectionState.CLOSED.
Process SpawnProcess-1:
Traceback (most recent call last):
  File "/usr/lib64/python3.12/multiprocessing/process.py", line 314, in _bootstrap
    self.run()
  File "/usr/lib64/python3.12/multiprocessing/process.py", line 108, in run
    self._target(*self._args, **self._kwargs)
  File "/home/florian/sources/hypercorn/src/hypercorn/trio/run.py", line 122, in trio_worker
    trio.run(partial(worker_serve, app, config, sockets=sockets, shutdown_trigger=shutdown_trigger))
  File "/home/florian/sources/hypercorn/.venv/lib/python3.12/site-packages/trio/_core/_run.py", line 2093, in run
    raise runner.main_task_outcome.error
  File "/home/florian/sources/hypercorn/src/hypercorn/trio/run.py", line 46, in worker_serve
    async with trio.open_nursery() as server_nursery:
  File "/home/florian/sources/hypercorn/.venv/lib/python3.12/site-packages/trio/_core/_run.py", line 881, in __aexit__
    raise combined_error_from_nursery
  File "/home/florian/sources/hypercorn/.venv/lib/python3.12/site-packages/trio/_highlevel_serve_listeners.py", line 25, in _run_handler
    await handler(stream)
  File "/home/florian/sources/hypercorn/src/hypercorn/trio/tcp_server.py", line 55, in run
    async with TaskGroup() as task_group:
  File "/home/florian/sources/hypercorn/src/hypercorn/trio/task_group.py", line 76, in __aexit__
    await self._nursery_manager.__aexit__(exc_type, exc_value, tb)
  File "/home/florian/sources/hypercorn/.venv/lib/python3.12/site-packages/trio/_core/_run.py", line 881, in __aexit__
    raise combined_error_from_nursery
  File "/home/florian/sources/hypercorn/src/hypercorn/trio/task_group.py", line 39, in _handle
    await send(None)
  File "/home/florian/sources/hypercorn/src/hypercorn/protocol/ws_stream.py", line 260, in app_send
    await self._send_wsproto_event(CloseConnection(code=CloseReason.INTERNAL_ERROR))
  File "/home/florian/sources/hypercorn/src/hypercorn/protocol/ws_stream.py", line 336, in _send_wsproto_event
    data = self.connection.send(event)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/florian/sources/hypercorn/.venv/lib/python3.12/site-packages/wsproto/connection.py", line 107, in send
    raise LocalProtocolError(
wsproto.utilities.LocalProtocolError: Event CloseConnection(code=<CloseReason.INTERNAL_ERROR: 1011>, reason=None) cannot be sent in state ConnectionState.CLOSED.

For trio the report says:

With asyncio the connection is closed and the server doesn't crash (though the "hello world" is not returned either):

[Florian Apolloner]

Thinking more about this, there are probably two separate issues:

• First one is the crash itself. I think hypercorn should never crash no matter whether the client is able to exploit a bug or not. Wouldn't it make sense to have a try/except at a very high level that allows to shut down a single connection in case of such errors.

• The shutting down of the connection in combination with the asgi app still sending messages results in the mentioned traceback.

[Phil Jones]

Fixed by 0bb4fb9

[Phil Jones]

Also note workers are now restored (repopulated) if they crash.

[Florian Apolloner]

This is lovely, thank you so much!