Revisit: Retrying in the merge group
When monorepo was started, Mac bots were highly flakey in the merge group. We have automatic retries in these cases and it was very helpful to retry one or two bots.
Mac bots are way more stable now thanks to the teams efforts (🥳) but now the retries are blocking the merge group when there are real failures (e.g. https://github.com/flutter/flutter/pull/169187)
We should probably:
- Reduce the retry count in the merge group
- Attempt to see if errors are real and fail fast
- If we see multiple errors from different builders, fail fast.
Thoughts @matanlurey @zanderso ?
Reducing retries if the stability problems they were hacking around are gone sgtm.
Reducing retries if the stability problems they were hacking around are gone sgtm.
That sound be easy to verify. I'd like to also have orchestrators fail fast if a sub build fails in the merge group. Anything to get positive failures back to the PR sooner (and let good PRs through).
Let's give adjusting a shot!