Avoid skipping triage jobs

[olupton]

Avoid skipping triage jobs in case of failure. See here in the docs; without this then I believe the failed status of the earlier run-jobs.outcome job step caused the triage job to be skipped.

[terrykong]

Thanks for tackling this @olupton ! Can you confirm if this would address this: #306 ?

Two questions:

1. Do you know why it failed in the first place and why needs.publish-completion.outputs.STATUS != 'success' was insufficient?
2. Would needs.<job_id>.result work?

[olupton]

Can you confirm if this would address this: #306 ?

I'm not sure about that. It's not obvious to me why the triage job ran in https://github.com/NVIDIA/JAX-Toolbox/actions/runs/6516962688, I'm not optimistic that this change will have fixed that issue.

1. Do you know why it failed in the first place and why needs.publish-completion.outputs.STATUS != 'success' was insufficient?

Based on https://docs.github.com/en/actions/learn-github-actions/expressions#failure-with-conditions:

You can include extra conditions for a step to run after a failure, but you must still include failure() to override the default status check of success() that is automatically applied to if conditions that don't contain a status check function.

I understand there was an implicit success() in the if clause of the triage job, so the (intentional) failure of run-jobs.outcome (because Slurm jobs failed) stopped the triage job from running.

2. Would needs.<job_id>.result work?

I guess yes if we made the triage job explicitly depend (needs) on run-jobs (which is the job whose overall status should be red in the case that I understand is supposed to trigger the triage job). Not sure what the advantage would be.

I launched https://github.com/NVIDIA/JAX-Toolbox/actions/runs/6545350101 manually using the latest version of this PR.

[olupton]

Outdated.