Two months ago I switched from Travis CI to GitHub actions to run my acceptance tests (with Cucumber and Capybara).
Whereas tests were “flaky” on the former, they became completely predictable since the switch (for two months).
Last evening, they started to fail (even with retries). With those kind of messages that I interpret as timeouts:
Request to http://fv-az186-255:3000/item/Vitraux - Bénel/D65963E34F30C462B600A7AF6DDC3F0B73F9 reached server, but there are still pending connections: http://fv-az186-255:3000/item/Vitraux%20-%20Bénel/D65963E34F30C462B600A7AF6DDC3F0B73F9 (Ferrum::StatusError)
When I “re-run” the job, sometimes it succeeds, but most of the time it doesn’t. They are not predictable anymore
Could it be related:
- to a change in the infrastructure (but I use mainly dockerized apps),
- to a high load on GH Actions (I’ve read about the cryptominers attack one week ago… but GH status tells everything is OK right now),
- to an error of mine (but that does not explain why it succeeds from time to time)?
I really hope to find a solution before tomorrow: my students will discover continuous integration, BDD and TDD, and I wouldn’t like them to have a bad opinion of it because of this problem.