Provide parallel_one_host_only via workers config file

[b10n1k]

https://progress.opensuse.org/issues/158146

[github-actions]

Great PR! Please pay attention to the following items before merging:

Files matching docs/*.asciidoc:

• Consider generating documentation locally to verify it is rendered correctly using tools/generate-docs

This is an automatically generated QA checklist based on modified files.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 98.41%. Comparing base (aa7ac4b) to head (dc4b30a).
Report is 2 commits behind head on master.

❗ Current head dc4b30a differs from pull request most recent head 0e31752

Please upload reports for the commit 0e31752 to get more accurate results.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #5695      +/-   ##
==========================================
- Coverage   98.42%   98.41%   -0.01%     
==========================================
  Files         394      394              
  Lines       38516    38447      -69     
==========================================
- Hits        37908    37837      -71     
- Misses        608      610       +2     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[b10n1k]

All the unnecessary changes have been reverted. I couldnt extend the tests further but the existing coverage should be satisfied for now.

[Martchus]

Looks good but I'd move out the two documentation changes just for now.

[Martchus]

I still think this lacks some additions to the scheduler logic. I'll come up with a test case next week to show that.

[Martchus]

I pushed a few more commits to add tests for the scheduler logic and extend it. It was not completely trivial because the evaluation of the setting now needs to be done in a much more dynamic way. I haven't done any manual tests yet.

[Martchus]

With my changes I find it good enough to be merged. I am keeping the "not-ready" label until we/I did some manual testing.

[Martchus]

I now tested this manually with a cluster of 2 jobs and all workers with PARALLEL_ONE_HOST_ONLY=1 as we'd use it in production. None of the jobs had the setting.

• With all workers on different hosts nothing is scheduled as expected (as we need at least 2 workers with a common host name).
• As soon as I replaced one worker so that there are now two workers with a common host jobs were scheduled on those two slots as expected.
• When I replaced one more worker so that there are now two hosts with two slots each in the pool I was able to occupy all worker slots and clusters were still not scheduled across multiple hosts.

So I guess that's good enough.

[Martchus]

Really annoying that codecov uploads occaisonally fail:

debug - 2024-07-02 09:24:29,772 -- Upload result --- {"result": "RequestResult(error=RequestError(code='HTTP Error 401', params={}, description='{\"detail\":\"Invalid token header. No credentials provided.\"}'), warnings=[], status_code=401, text='{\"detail\":\"Invalid token header. No credentials provided.\"}')"}
error - 2024-07-02 09:24:29,772 -- Upload failed: {"detail":"Invalid token header. No credentials provided."}

Since this isn't treated as an error (the CircleCI codecov step passes) I'll have to re-run all CircleCI tests.

[Martchus]

I retried yesterday and also today and Codecov is reliably broken. That's also the case on other PRs like #5723.

Strangely, OBS doesn't have the same problem. I checked the upload steps of a few PRs on https://github.com/openSUSE/open-build-service/pull. Maybe we should compare our setup with the one from OBS.

For the time being I checked https://output.circle-artifacts.com/output/job/5a4e33c3-6409-401d-b492-b79ba22e4c05/artifacts/0/cover_db/coverage.html manually and it looks good. So I'm merging this PR now.