Step conditions using !cancelled() need explict success check

[genehack]

Current Behavior

Slack context

.github/.github/workflows/pathogen-repo-ci.yaml.in

Lines 267 to 275 in 351da09

	- name: Run phylogenetic
	if: hashFiles('phylogenetic/Snakefile') && hashFiles('phylogenetic/build-configs/ci/config.yaml') && !cancelled()
	id: phylogenetic
	run: nextstrain build phylogenetic --configfile build-configs/ci/config.yaml
	- name: Run nextclade
	if: hashFiles('nextclade/Snakefile') && hashFiles('nextclade/build-configs/ci/config.yaml') && !cancelled()
	id: nextclade
	run: nextstrain build nextclade --configfile build-configs/ci/config.yaml

Expected behavior

The Conda build step failing should abort the rest of the workflow

Possible solution

• Add an explicit success() check
• ??

[tsibley]

Background: The commit which switched to using !cancelled() instead of continue-on-error: true (which didn't have this bug).

Adding a success() check alone won't DTRT, as we intentionally wanted the phylogenetic step to run even if the ingest step failed and the nextclade step to run even if the ingest or phylogenetic steps failed. Conceptually, we want to check that every step before the ingest step was successful.

Possible solutions:

1. Calculate status of steps before ingest explicitly (in a new step) using something like contains(steps.*.conclusion, 'failure'), setting it as the new step output, and then condition on that in ingest, phylogenetic, and nextclade (in addition to continuing to condition on !cancelled()).

2. Move each of ingest, phylogenetic, and nextclade into their own jobs, where they're naturally independent of each other, instead of doing a lot of work to make them independent steps in the same job.

[victorlin]

2. Move each of ingest, phylogenetic, and nextclade into their own jobs, where they're naturally independent of each other, instead of doing a lot of work to make them independent steps in the same job.

+1