Cleanup and update releng/test_config.yaml

[xmudrii]

releng/test_config.yaml contains definitions for kubernetes/kubernetes tests (mainly periodics). We use those definitions to generate jobs for release branches when cutting a new release branch.

Over the time, this file has become unmaintained which is a huge risk for having jobs that are not properly testing release branches.

Here are some problems that I observed; this is a non exhaustive list, in other words, there might be other issues that I didn't observe:

• .jobs
  • ci-kubernetes-e2e-gce-cos-.+?-default is inconsistent
    • Different release branches has different args (to confirm if intentional)
      • Only ci-kubernetes-e2e-gce-cos-k8sbeta-default has --env=ENABLE_CACHE_MUTATION_DETECTOR=true arg
      • Only ci-kubernetes-e2e-gce-cos-k8sstable1-default has --env=ENABLE_POD_SECURITY_POLICY=true
      • Other ci-kubernetes-e2e-gce-cos-.+?-default jobs have no (additional args)
    • k8sbeta and k8sstable1 jobs are missing testgridNumFailuresToAlert parameter
  • We're missing betaapis tests for k8sbeta, k8sstable1, and k8sstable2 releases
    • Verify if args for betaapis tests are still valid
• .images
  • cos1, cos2, ubuntu1, and ubuntu2 images are not present in jobs names and therefore are likely unused (check if it is safe to remove these images)
• .testSuites
  • flaky, soak, stackdriver, updown, and nosat suites are not present in jobs names and therefore are likely unused (check if it is safe to remove these images)
• .nodeImages
  • cos1, cos2, ubuntu1, and ubuntu2 images are not present in jobs names and therefore are likely unused (check if it is safe to remove these images)
• .nodeTestSuites
  • gkespec and serial are likely unused

Once these issues are addressed, we should verify if arguments are up to date for remaining test options.

cc @kubernetes/release-engineering @dims

[xmudrii]

/sig release
/area release-eng

[xmudrii]

/priority important-soon

[dims]

/milestone v1.31

[k8s-ci-robot]

@dims: The provided milestone is not valid for this repository. Milestones in this repository: [someday, v1.24, v1.25]

Use /milestone clear to clear the milestone.

In response to this:

/milestone v1.31

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.

[k8s-triage-robot]

The Kubernetes project currently lacks enough contributors to adequately respond to all issues.

This bot triages un-triaged issues according to the following rules:

• After 90d of inactivity, lifecycle/stale is applied
• After 30d of inactivity since lifecycle/stale was applied, lifecycle/rotten is applied
• After 30d of inactivity since lifecycle/rotten was applied, the issue is closed

You can:

• Mark this issue as fresh with /remove-lifecycle stale
• Close this issue with /close
• Offer to help out with Issue Triage

Please send feedback to sig-contributor-experience at kubernetes/community.

/lifecycle stale

[xmudrii]

/remove-lifecycle stale

[xmudrii]

We're running into this issue again ahead of the 1.31 jobs cut. These are the high priority issues at the moment:

• ci-kubernetes-e2e-gce-cos-k8sbeta-default (1.30) has --env=ENABLE_CACHE_MUTATION_DETECTOR=true, do we need to keep this for 1.31 as well?
• ci-kubernetes-e2e-gce-cos-k8sbeta-alphafeatures (1.30) uses a custom testSuite called alphafeatures-ccm-eventedpleg, do we need to keep it for 1.31 as well?
• ci-kubernetes-e2e-gce-cos-k8sstable1-default (1.29) has --env=ENABLE_POD_SECURITY_POLICY=true, however, this env is present only for 1.29. Do we need it for any other version or is it specific to only 1.29?
• We have betaapis tests only for 1.26 and 1.27. Are we getting rid of the betaapis tests or do we need those tests for all other release branches?

/priority critical-urgent
cc @dims @mehabhalodiya

[dims]

ci-kubernetes-e2e-gce-cos-k8sbeta-default (1.30) has --env=ENABLE_CACHE_MUTATION_DETECTOR=true, do we need to keep this for 1.31 as well?

Yes, we should keep it. helps catch the scenario where anything mutates a shared informer cache (see description here)

ci-kubernetes-e2e-gce-cos-k8sbeta-alphafeatures (1.30) uses a custom testSuite called alphafeatures-ccm-eventedpleg, do we need to keep it for 1.31 as well?

Yes, please. will help give us signal on an alpha feature https://github.com/kubernetes/kubernetes/blob/9413cf204ac92711cc8aff472b1ed11ba79760ac/pkg/features/kube_features.go#L261-L267

ci-kubernetes-e2e-gce-cos-k8sstable1-default (1.29) has --env=ENABLE_POD_SECURITY_POLICY=true, however, this env is present only for 1.29. Do we need it for any other version or is it specific to only 1.29?

We can drop it, does not do much AFAICT looking through references - https://cs.k8s.io/?q=ENABLE_POD_SECURITY_POLICY%7CEnablePodSecurityPolicy&i=nope&files=&excludeFiles=vendor&repos=

[BenTheElder]

Can we please just use config fork annotations on normal prowjobs and stop using this other thing entirely?

I don't think SIG release should be defining these jobs, just forking them when new branches are cut, and I really doubt anyone else in the project is looking at this file. It's hard to even understand what these config fields are doing.

cc @liggitt @tallclair re: PSP

yes we want to keep the cache mutation detector enabled.

[liggitt]

ENABLE_POD_SECURITY_POLICY is irrelevant in 1.25+ and can be dropped

[BenTheElder]

documented and widely used:

https://github.com/kubernetes/test-infra/tree/master/releng/config-forker#supported-annotations

Versus ~no docs for generate_tests and the out of band config.

If the config fork annotations are not powerful enough for some reason, let's fix that.

[xmudrii]

Can we please just use config fork annotations on normal prowjobs and stop using this other thing entirely?

Long term, we can look into it. But short term and specially for this upcoming release, definitely no. We're happy to look into it, but this will require some time and we want to get the release branch up and running ASAP.

[BenTheElder]

I'm pretty sure https://kubernetes.slack.com/archives/CN0K3TE2C/p1722833073336809 is related

The problem is that job maintainers are going to PR the job definitions and expect auto-forking, and are not aware that this out of band config needs changing as well, since it is only used once per release branch.

I think we really need to switch for 1.32 or we'll keep having regressions.

cc @pohly @aojea, @kubernetes/release-engineering

[xmudrii]

/assign
Let met look into this and if we can quickly drop this