Can I force a job to be re-run under "Re-run failed" even if it succeeded?

[spalladino]

Select Topic Area

Question

Body

On our CI workflow, we have a setup steps that primes an ephemeral environment that our tests need to ping. This environment is configured to shutdown automatically after a few minutes of inactivity, to save costs.

The problem we have is that, when choosing to "Re-run failed" on a failed workflow, this setup step is not re-run, since it was successful on the original run. But without re-running this setup, we cannot re-run the failed tests. Is there a way to specify in the workflow yml that a given step must always be run, even during a "re-run failed" if it was successful in the latest attempt?

[mickeygousset]

I think there are two different questions here:

1. You can choose to "Rerun all Jobs" not just failed jobs
2. If can include an if statement in a step, such as if: always(), which will make the step always run, even if steps previous to it have failed. This is great for being able to log test results somewhere after tests have failed.

[spalladino]

Thanks for the reply @mickeygousset, but I think I didn't make my original question clear. What we're looking for is, when re-running failed jobs only, whether it's possible to have a specific job to be re-run even if it succeeded originally. To give an example, let's say I have the following workflow: a set of parallel test jobs that depend on a setup job.

graph TD;
setup --> test1;
setup --> test2;
setup --> test3;
setup --> test4;

Let's say in a given run all jobs succeeded except for test2. I would want "re-run failed jobs" to run both setup and test2, even though setup had succeeded.

Note that hitting "rerun all jobs" wouldn't do what we want, since it'd re-run test1, test3, and test4, which we don't need to re-run. And marking a step as if: always() only works for later jobs in the dependency tree, not previous ones.

[mickeygousset]

Ah, ok. yeah, there is no good way to do that OOTB for what you have here.

That is an interesting question. I'm going to take that back to my team internally and see if anyone has any thoughts

[joshjohanning]

I agree, nothing really out of the box. Re-running the failed job would only re-run test2 and any jobs that depend on test2.

For something out of the box, you could add in some of setup's logic into the test2 job that includes what the setup job is doing and only conditionally run it.

And most re-use, you could create the steps of the setup job as a composite action so that all jobs share the same logic without having to duplicate a lot.

- name: setup
  if: ${{ github.run_attempt > 1 }}
  uses: ./setup-action

# test2 stuff...

There is also a debug variable you can tap into for this if you don't want to use github.run_attempt. When you re-run failed jobs, you could check the box to enable debug logging on the next job run. The debug logging isn't helpful in your case, but the fact that there's a variable that you can use potentially is:

- name: setup
  if: steps.debug.outputs.RUNNER_DEBUG == 1
  uses: ./setup-action

# test2 stuff...

[spalladino]

Got it, thanks for the response!

[spalladino]

Is there a way to turn this question into a suggestion for a new feature for github actions, so there's a more out-of-the-box way to accomplish it? Is there a feature-request-inbox?

[mickeygousset]

Its feedback that will get taken back to the Actions team. I think it is actually a valid use case. Might be a bit tricky to implement, but its worth turning into a feature request.

[spalladino]

Thank you!