Do not fail fast

[cstamas]

By default the matrix will fail fast, but in this case it makes no sense, and is even better to make it non fail fast.

Ref:
https://github.com/apache/maven-gh-actions-shared/blob/3b6a3fd53a6cc1ea2f5f25625cc9fd6a9d6ebe22/.github/workflows/maven-verify.yml#L162

[desruisseaux]

It looks fine to me. But wouldn't "fail fast" be nicer for the carbon footprint?

[cstamas]

Certainly it is. It was just an idea to point out this possibility as well, as currently whole CI run can be thrown away, while in that mode one can "rerun failed jobs" via GH UI.

[olamy]

If we keep running the workflow while we know it doesn't make sense, as something has already failed.
It sounds bad to me, especially as we share GHA runners with the entire Apache organisation, and so we will potentially consume running time for nothing useful. We had issue in the past as running too much (nothing is free with GHA)

[desruisseaux]

If we have a limited amount of time (in addition to concern for carbon footprint), then indeed it seems to me that we should stop the build as soon as we know that there is a problem.

[desruisseaux]

Maybe a nice compromise would be: let finish the jobs that are already running, but do not start the jobs that are waiting in the queue. The rational is that if some running jobs are close to finish successfully, it would be a waste (even from carbon footprint perspective) to cancel them and restart them from the beginning when we select "Re-run failed jobs". However, I doubt that GitHub Action provides this option…