Stu Hood

So the relevant aspect of #16277 not triggering in the repro on #16327 is that it means that the `test` goal is _not_ retrying the invalidated process. That might be...

https://github.com/pantsbuild/pants/runs/7661011168?check_suite_focus=true

Note that this issue can still occur with `cache_content_behavior=validate` if an artifact disappears after `validate`, but before it is actually fetched.

> @jyggen sorry for not merging this earlier, we'd like to include this in the future 2.15 release, would you mind rebasing this with `main` as there have some changes...

@jyggen: Note that this doesn't add the backend to `src/python/pants/bin:plugins`, so it won't yet actually go in the `pantsbuild.pants` dist. If you're ready for that, would you mind opening another...

Although we need to do this for `go`, do you think that it is actually worthwhile for terraform? * Does it make a noticeable difference in performance? * Do we...

This should also ensure that the local cache runner uses the computed `eager_fetch` value.

This exhibited in #15885, with multiple downstream consumers of a process racing to restart it.

It's additionally the case that the backtracking warning is fairly noisy. Unlike cache warnings in general, these are not backed off, and if both the local and remote caches are...

#16009 is also related, in that the backtrack attempts that we see are never "repaired", and so are noisier in CI than they would otherwise be.