QA » openQA Project

We have job_done_hook = env host=openqa.suse.de exclude_group_regex='.*(Development|Public Cloud|Released|Others|Kernel|Virtualization).*' grep_timeout=60 nice ionice -c idle /opt/os-autoinst-scripts/openqa-label-known-issues-and-investigate-hook configured so the hook script runs regardless of the job's result. I'm wondering where we take care not to run into the "Investigate retry job: … failed" assumption for passed/softfailed jobs. Since we have no job_done_hook_enable_… = 1 settings the hook script is actually only running for failed, incomplete or timeout_exceeded results.

Since the job has _TRIGGER_JOB_DONE_HOOK=1 the generic hook script is triggered for this particular job after all (regardless of the result). We apparently don't do any extra checks in openqa-label-known-issues-and-investigate-hook to avoid running into the "Investigate retry job: … failed" assumption so this is what's happening. Supposedly we should have an extra check there. I'm not sure where the _TRIGGER_JOB_DONE_HOOK=1 job settings comes from and why it was added.

mkittler wrote:

I'm not sure where the _TRIGGER_JOB_DONE_HOOK=1 job settings comes from and why it was added.

_TRIGGER_JOB_DONE_HOOK=1 was added by me as part of #98862 for investigate:retry jobs.

We need to run the hook script in order to report when a retry job passed.
For that I also needed to enable job_done_hook, and I guess this is now also called for softfailed. Should I rather configure job_done_hook_passed instead?

It would be good to estimate this with @okurz to clarify whether we can really treat "softfailed" as "passed" here.

In general what users commonly expect is that the investigation jobs tell if the*same* issue happens again. We make the assumption that if a job fails again then likely it's the same issue even though that will not be generally true. IMHO that assumption is still fine for the sake of openqa-investigate. Regarding failed, softfailed I assume we only trigger openqa-investigate in the first place for failed jobs hence we want to know if retry jobs fail. So in my understanding all jobs with "ok-result" should be treated the same

mkittler wrote:

I suppose I will just add a check to skip writing this comments for passed/softfailed jobs.

As discussed in the weekly 2023-03-10 we clarified that we do have the feature to write a comment if a job passes so we should ensure that for any "ok" result we treat it the same. As soft-fail effectively means "known issue" then the reason for job failure can not be the same as the original "new, unreviewed issue".