We have found fail-fast has been getting in the way more than it's been helping recently. For example:
Sometimes only a particular set of versions or OSes are broken, and we want to let the other checks complete so we can know the full set.
We see quite a large degree of parallelism when running the checks, so most of the jobs have nearly finished anyway when they are cancelled by fail-fast.
fail-fast deals better with transient failures.
Merge / deployment checklist
[ ] Confirm this change is backwards compatible with existing workflows.
[ ] Confirm the readme has been updated if necessary.
[ ] Confirm the changelog has been updated if necessary.
We have found
fail-fasthas been getting in the way more than it's been helping recently. For example:fail-fast.fail-fastdeals better with transient failures.Merge / deployment checklist