Socket Mode: ENOTSOCK every 300 minutes on Windows

Edit: updated reproducible versions to latest.

Functionality is not impacted; the reconnect happens properly. This bug is about log spam.

Reproducible in:

slack-sdk-3.7.0 slack-bolt-1.6.1 Python 3.7.8 MINGW64_NT-10.0-18363 Microsoft Windows [Version 10.0.18363.1621]

Steps to reproduce:

app = slack_bolt.App(token=<OAUTH_TOKEN>, signing_secret=<SIGN_SECRET>, logger=...)
handler = slack_bolt.adapter.socket_mode.SocketModeHandler(app, <APP_TOKEN>)

Run, then wait 300 minutes.

Actual result:

Every 300 minutes, this error is logged.

2021-07-06 18:53:32,090 S.Sok ERROR on_error invoked (session id: 2585b6bb-07ea-470e-be5c-449bf120f013, error: OSError, message: [WinError 10038] An operation was attempted on something that is not a socket)
Traceback (most recent call last):
  File "...\slack_sdk\socket_mode\builtin\connection.py", line 289, in run_until_completion
    all_message_trace_enabled=self.all_message_trace_enabled,
  File "...\slack_sdk\socket_mode\builtin\internals.py", line 235, in _receive_messages
    logger=logger,
  File "...\slack_sdk\socket_mode\builtin\internals.py", line 251, in _fetch_messages
    remaining_bytes = receive()
  File "...\slack_sdk\socket_mode\builtin\internals.py", line 226, in receive
    raise e
  File "...\slack_sdk\socket_mode\builtin\internals.py", line 217, in receive
    received_bytes = sock.recv(size)
  File "C:\python3\lib\ssl.py", line 1058, in recv
    return super().recv(buflen, flags)
OSError: [WinError 10038] An operation was attempted on something that is not a socket

Analysis

10038 is WSAENOTSOCK and (in Python on Windows) also errno.ENOTSOCK. It’s roughly comparable to EBADF (see for example https://bugs.python.org/issue13344) Should it be caught in _fetch_messages?

            except OSError as e:
                if e.errno == errno.EBADF:  # if e.errno in (errno.EBADF, errno.ENOTSOCK)
                    logger.debug("The connection seems to be already closed.")
                    return bytes()

Other info

Every now and then (~25% of the time?) this error is logged immediately afterwards.

2021-07-06 18:53:35,412 S.Sok ERROR Failed to establish a connection (session id: f1c1ee8e-a386-4e8a-8bd4-81b911c4a813, error: The read operation timed out)
Traceback (most recent call last):
  File "...\slack_sdk\socket_mode\builtin\connection.py", line 139, in connect
    status, headers, text = _parse_handshake_response(sock)
  File "...\slack_sdk\socket_mode\builtin\internals.py", line 150, in _parse_handshake_response
    line = _read_http_response_line(sock)
  File "...\slack_sdk\socket_mode\builtin\internals.py", line 129, in _read_http_response_line
    c: str = sock.recv(1).decode("utf-8")
  File "C:\python3\lib\ssl.py", line 1056, in recv
    return self.read(buflen)
  File "C:\python3\lib\ssl.py", line 931, in read
    return self._sslobj.read(len)
socket.timeout: The read operation timed out

[seratch]

Hi @dubois, we’ve just released v3.9.0rc2, which includes the fix for this issue. If you have a chance, try the version out 👋 https://pypi.org/project/slack-sdk/3.9.0rc2/

[seratch]

@dubois Thanks for taking the time to try the RC1 version! And, I’m sorry for bothering you as the updates on the WebClient does not help you resolve the issue described here.

Also, the analysis in your description looks valid. I will come up with updates on the relevant part and will release a new RC version including the change. Thanks for being patient and helping us resolve the issue.