artifacts: annotation for logs when something else fails

[stevekuznetsov]

In general, if the e2e tests container fails, we will often want to get the logs from the installer as well, even if the installer container did not fail. We should thread this through as an annotation like ci-operator.openshift.io/wait-for-container-artifacts and slap the setup container into there.

/cc @vrutkovs @smarterclayton

[vrutkovs]

Discussed this on CI meeting today - the plan is to store the output of installer, gzip it and let ci-operator to decide if it needs to be kept or not. +1 from my side on this, I can prepare the template updates once this is decided

[stevekuznetsov]

xref https://jira.coreos.com/browse/DPTP-109

[smarterclayton]

The design as implemented of this made general debugging and triage of e2e issues worse. It's much harder now to find the failure point in logs, and success logs are much worse. It replaced quick scans that might take 2-3 seconds and turned them into 30-40 second actions.

The original goal was to keep build logs focused on triage. I'd like to suggest having the annotation cause the logs to be stored as artifacts but not displayed in output, and I think that would accomplish Vadim's original request while preserving the key goal of keeping build logs tight.

[droslean]

fixed on #206
/close

[openshift-ci-robot]

@droslean: Closing this issue.

Details

In response to this:

fixed on #206
/close

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.