CI keeps failing due to timeout

[rlycx]

When i open a PR, i often have to run the job a couple of times before it completes in time.

We've had issues before with our integration tests being flaky and becoming deadlocked indefinitely. #6 introduced the 20 second timeout for integration tests, so they fail more quickly once they become deadlocked.

It's also difficult to see what exactly causes the failure. A first step towards resolving this issue could be to see what we can do to improve the output that we get when a test times out on CI, because at time of writing this is just a massive stack trace with mostly callsites originating from pytest plugins and the like.

Either the time-out is just too short for CI, or we're still getting deadlocks. So far, i've not been able to really reproduce any deadlocks by running the tests locally. We could try to relax that timeout a bit more, but not by too much. As the comment above the timeout configuration states, we're already being quite generous with our time limit.

Examples

The last four PRs have all seen at least one failure of the integration tests due to it exceeding the time limit:

• Update cargo dependencies #19
  • failed push
• Clarify consumer log message when chunk reservation expired #18
  • failed push
  • failed re-run
• Allow http(s) scheme in the producer client host string #17
  • failed push
• Allow configuring the max number of chunk retries #16
  • failed merge by Hoff