Unresponsive without internet connection

[georgeto]

When SIPA loses its Internet access, it becomes unresponsive. Maybe because it tries to fetch other services, such as the Pycroft backend or PBX?

[lukasjuhrich]

I think I have a hunch going through the logs: They are full of:

Sun Jan 24 20:54:35 2021 - *** uWSGI listen queue of socket "0.0.0.0:5000" (fd: 3) full !!! (101/100) ***
Sun Jan 24 20:54:36 2021 - *** uWSGI listen queue of socket "0.0.0.0:5000" (fd: 3) full !!! (101/100) ***
Sun Jan 24 20:54:37 2021 - *** uWSGI listen queue of socket "0.0.0.0:5000" (fd: 3) full !!! (101/100) ***

Now add the following to the mix:

WSGI app 0 (mountpoint='/sipa') ready in 134 seconds on interpreter 0x55ebba9f86b0 pid: 95268 (default app)

This together with Git fetch failed suggests to me that on startup, SIPA waits for the fetch until it times out. In thea meantime, a lot of requests (even if it's just HEAD / from a proxy) build up and flood the listen queue so the worker commits HARAKIRI.

There are multiple options for this:

• Quickfix: Figure out whether we can decrease the timeout for the git fetch and increase the listen queue capacity
• Proper fix: Establish some kind of „health check“ either before all of the start startup, or before trying to fetch the git repo. This may e.g. just be a ping to the git server.