KeyError: 'Session is disconnected'

[fabswt]

I'm getting these tracebacks:

Sep 29 10:42:21Z gliglish app/web.5 Traceback (most recent call last):
Sep 29 10:42:21Z gliglish app/web.5 File "/app/.heroku/python/lib/python3.11/site-packages/gunicorn/workers/base_async.py", line 55, in handle
Sep 29 10:42:21Z gliglish app/web.5 self.handle_request(listener_name, req, client, addr)
Sep 29 10:42:21Z gliglish app/web.5 File "/app/.heroku/python/lib/python3.11/site-packages/gunicorn/workers/base_async.py", line 108, in handle_request
Sep 29 10:42:21Z gliglish app/web.5 respiter = self.wsgi(environ, resp.start_response)
Sep 29 10:42:21Z gliglish app/web.5 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Sep 29 10:42:21Z gliglish app/web.5 File "/app/.heroku/python/lib/python3.11/site-packages/flask/app.py", line 2213, in __call__
Sep 29 10:42:21Z gliglish app/web.5 return self.wsgi_app(environ, start_response)
Sep 29 10:42:21Z gliglish app/web.5 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Sep 29 10:42:21Z gliglish app/web.5 File "/app/.heroku/python/lib/python3.11/site-packages/flask_socketio/__init__.py", line 43, in __call__
Sep 29 10:42:21Z gliglish app/web.5 return super(_SocketIOMiddleware, self).__call__(environ,
Sep 29 10:42:21Z gliglish app/web.5 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Sep 29 10:42:21Z gliglish app/web.5 File "/app/.heroku/python/lib/python3.11/site-packages/engineio/middleware.py", line 63, in __call__
Sep 29 10:42:21Z gliglish app/web.5 return self.engineio_app.handle_request(environ, start_response)
Sep 29 10:42:21Z gliglish app/web.5 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Sep 29 10:42:21Z gliglish app/web.5 File "/app/.heroku/python/lib/python3.11/site-packages/socketio/server.py", line 607, in handle_request
Sep 29 10:42:21Z gliglish app/web.5 return self.eio.handle_request(environ, start_response)
Sep 29 10:42:21Z gliglish app/web.5 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Sep 29 10:42:21Z gliglish app/web.5 File "/app/.heroku/python/lib/python3.11/site-packages/engineio/server.py", line 418, in handle_request
Sep 29 10:42:21Z gliglish app/web.5 socket = self._get_socket(sid)
Sep 29 10:42:21Z gliglish app/web.5 ^^^^^^^^^^^^^^^^^^^^^
Sep 29 10:42:21Z gliglish app/web.5 File "/app/.heroku/python/lib/python3.11/site-packages/engineio/server.py", line 663, in _get_socket
Sep 29 10:42:21Z gliglish app/web.5 raise KeyError('Session is disconnected')
Sep 29 10:42:21Z gliglish app/web.5 KeyError: 'Session is disconnected'

Here's a similar one with logging enabled for engineio:

Sep 29 11:26:16Z gliglish app/web.16 10.1.43.18 - - [29/Sep/2023:11:26:15 +0000] "POST /socket.io/?EIO=4&transport=polling&t=OhWi7R0&sid=G59Rf-ezQhOMZCPoAAAy HTTP/1.1" 500 0 "-" "-"
 Sep 29 11:26:16Z gliglish app/web.16 [2023-09-29 11:26:15 +0000] [7] [ERROR] Error handling request /socket.io/?EIO=4&transport=polling&t=OhWi7Qk&sid=gX6-qEK2MvQ1GV0OAAAx
 Sep 29 11:26:16Z gliglish app/web.16 Traceback (most recent call last):
 Sep 29 11:26:16Z gliglish app/web.16 File "/app/.heroku/python/lib/python3.11/site-packages/gunicorn/workers/base_async.py", line 55, in handle
 Sep 29 11:26:16Z gliglish app/web.16 self.handle_request(listener_name, req, client, addr)
 Sep 29 11:26:16Z gliglish app/web.16 File "/app/.heroku/python/lib/python3.11/site-packages/gunicorn/workers/base_async.py", line 108, in handle_request
 Sep 29 11:26:16Z gliglish app/web.16 respiter = self.wsgi(environ, resp.start_response)
 Sep 29 11:26:16Z gliglish app/web.16 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 Sep 29 11:26:16Z gliglish app/web.16 File "/app/.heroku/python/lib/python3.11/site-packages/flask/app.py", line 2213, in __call__
 Sep 29 11:26:16Z gliglish app/web.16 return self.wsgi_app(environ, start_response)
 Sep 29 11:26:16Z gliglish app/web.16 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 Sep 29 11:26:16Z gliglish app/web.16 File "/app/.heroku/python/lib/python3.11/site-packages/flask_socketio/__init__.py", line 43, in __call__
 Sep 29 11:26:16Z gliglish app/web.16 return super(_SocketIOMiddleware, self).__call__(environ,
 Sep 29 11:26:16Z gliglish app/web.16 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 Sep 29 11:26:16Z gliglish app/web.16 File "/app/.heroku/python/lib/python3.11/site-packages/engineio/middleware.py", line 63, in __call__
 Sep 29 11:26:16Z gliglish app/web.16 return self.engineio_app.handle_request(environ, start_response)
 Sep 29 11:26:16Z gliglish app/web.16 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 Sep 29 11:26:16Z gliglish app/web.16 File "/app/.heroku/python/lib/python3.11/site-packages/socketio/server.py", line 607, in handle_request
 Sep 29 11:26:16Z gliglish app/web.16 return self.eio.handle_request(environ, start_response)
 Sep 29 11:26:16Z gliglish app/web.16 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 Sep 29 11:26:16Z gliglish app/web.16 File "/app/.heroku/python/lib/python3.11/site-packages/engineio/server.py", line 438, in handle_request
 Sep 29 11:26:16Z gliglish app/web.16 socket = self._get_socket(sid)
 Sep 29 11:26:16Z gliglish app/web.16 ^^^^^^^^^^^^^^^^^^^^^
 Sep 29 11:26:16Z gliglish app/web.16 File "/app/.heroku/python/lib/python3.11/site-packages/engineio/server.py", line 663, in _get_socket
 Sep 29 11:26:16Z gliglish app/web.16 raise KeyError('Session is disconnected')
 Sep 29 11:26:16Z gliglish app/web.16 KeyError: 'Session is disconnected'

Because there's no reference to my own codebase in it, I'm having a hard time troubleshooting it.

In a similar thread, @miguelgrinberg wrote "This error occurs when a client take too long to send its ping messages." but it's not clear to me whether this referred to all "KeyError: 'Session is disconnected'" errors.

I hit major traffic to my project (100K pageviews per day) but I think I had these before the traffic spike. (The project is hosted in Europe on Heroku with most of the traffic gain coming from India and Pakistan.)

Thanks for the help! 🙏

[miguelgrinberg]

The meaning of "session is disconnected" is this: the client was connected, but at some point became unresponsive and the server decided to remove it. After the removal, the client showed up again and attempted to continue using the connection.

If you enable Engine.IO logs and look at previous log entries that match the sid you will see a "client is gone" message, which is when the server decided that the client wasn't responding and should be removed.

The most straightforward definition of "unresponsive client" is one that fails to respond in a timely manner when the server pings it. But there is another possibility which is common, which is that your server became slow due to managing too many connections, and due to the slow processing it fails to handle responses to pings in time, even if the client sends them promptly.

[fabswt]

Thank you. What's the recommended way to handle them?

If the client merely became unresponsive I could let them be. (Or maybe I should detect when clients close a tab, for example, and then ask the client to disconnect from the server?)

If it's a case of the server being overloaded… I'll have to think about it, but your opinion would be most welcome.

[miguelgrinberg]

There isn't really much information or context I can use, but in general it is bad for servers to overload, so you should provision your server adequately for the traffic that you expect to have. If you don't know if the server was overloaded or not that is also a problem, you should know. Consider monitoring your server.