aaugustin / django-userlog Goto Github PK

There are at least two cases where an HTTP request will fail and not be logged:

1. If the view function hangs or gets killed somehow, possibly because the application server enforces a timeout.

   To avoid that problem, I could log requests in process_request and update the log in process_response. However that increases the overhead from one to two cache writes per HTTP request for limited benefits.

   I don't think userlog should focus on error handling. That's the realm of Sentry.

2. If response middleware that runs before the userlog middleware raises an exception.

   The userlog middleware must run as late as possible because other response middleware may change the response's status code. If an exception happens in an earlier middleware, the only way for userlog to catch it is to connect to the got_request_exception signal. It's not obvious that this would be a good idea.

   I'm willing to accept this as a known limitation and push the responsibility to flaws in the design or the implementation of Django's middleware API.

Currently userlog only works for logged-in users.

There might be a use case for anonymous users. That would require logging requests by session id (easy) and finding a way to recover the session id (not obvious).

The session id could be obtained by sending the user to a specific view that would store the association between a known value and the session id in Redis.