fix scheduler version skew handling for in-place pod resize

[wlp1153468871]

What type of PR is this?

/kind bug

What this PR does / why we need it:

If PodStatus.Resize field is not empty, scheduler uses max(ResourcesAllocated, Requests) even if the feature-gate is disabled.

Which issue(s) this PR fixes:

Fixes #117767

Special notes for your reviewer:

Does this PR introduce a user-facing change?

 NONE

Additional documentation e.g., KEPs (Kubernetes Enhancement Proposals), usage docs, etc.:

- [KEP]: https://github.com/kubernetes/enhancements/tree/master/keps/sig-node/1287-in-place-update-pod-resources#version-skew-strategy

[k8s-ci-robot]

[APPROVALNOTIFIER] This PR is NOT APPROVED

This pull-request has been approved by: wlp1153468871
Once this PR has been reviewed and has the lgtm label, please assign deads2k, sanposhiho for approval. For more information see the Kubernetes Code Review Process.

The full list of commands accepted by this bot can be found here.

Needs approval from an approver in each of these files:

• pkg/api/OWNERS
• pkg/quota/v1/OWNERS
• pkg/scheduler/OWNERS

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[AxeZhan]

Why are we changing these resource size in tests?🤔️

[wlp1153468871]

According to the chapter Version Skew Strategy of the KEP-1287, when kube-scheduler computes the pod resource requests, it should use max(AllocatedResources, Requests) if PodStatus.Resize != "".

In this test case, the PodStatus.Resize is equal to Infeasible and the AllocatedResources is less than Requests, so the expected result is Requests. I'm not sure that this change is reasonable when the PodStatus.Resize is equal to Infeasible. If I misunderstood, please correct me. Thanks.

FYI:

• https://github.com/kubernetes/enhancements/tree/master/keps/sig-node/1287-in-place-update-pod-resources#version-skew-strategy
• https://github.com/kubernetes/kubernetes/blob/master/pkg/apis/core/types.go#L2576

[AxeZhan]

Ah I see, sgtm

[wlp1153468871]

@vinaykul Can you help me review a PR? Thank you

[alexzielenski]

/remove-sig api-machinery

[alexzielenski]

untagged mistakenly

/sig api-machinery
/triage accepted
/cc @deads2k

[wlp1153468871]

@mikedanese @sttts Can you help me review a PR? Thank you

[vinaykul]

@wlp1153468871 Thank you very much for picking this up! Could we please fix both kubelet and scheduler fix in a single PR so that it fully resolves the issue?

We should add an e2e test for this.. maybe we can extend test/e2e/node/pod_resize.go
In order to exercise InPlacePodVerticalScaling feature-gate enabled in apiserver but disabled in kubelet and scheduler, one way might be to clone this CI job into two other CI jobs (maybe with 72 hour cadence) and set this FG to false using --feature-gates flag passed via KUBELET_TEST_ARGS and SCHEDULER_TEST_ARGS.

@bobbypage @BenTheElder Does that above suggestion look reasonable? (Will setting the --feature-gates flag override for kubelet & scheduler? If not, would the inverse work?)

I'm also wondering if an e2e skew testing CI job already exists. It sounds a boilerplate version skew test job for our version skew support matrix is something many alpha and beta features could use, and in general to detect skew regressions. I do see references to --check-version-skew flag in test-infra repo but my git grep shows them all set to false.

[wlp1153468871]

@wlp1153468871 Thank you very much for picking this up! Could we please fix both kubelet and scheduler fix in a single PR so that it fully resolves the issue?

We should add an e2e test for this.. maybe we can extend test/e2e/node/pod_resize.go In order to exercise InPlacePodVerticalScaling feature-gate enabled in apiserver but disabled in kubelet and scheduler, one way might be to clone this CI job into two other CI jobs (maybe with 72 hour cadence) and set this FG to false using --feature-gates flag passed via KUBELET_TEST_ARGS and SCHEDULER_TEST_ARGS.

@bobbypage @BenTheElder Does that above suggestion look reasonable? (Will setting the --feature-gates flag override for kubelet & scheduler? If not, would the inverse work?)

I'm also wondering if an e2e skew testing CI job already exists. It sounds a boilerplate version skew test job for our version skew support matrix is something many alpha and beta features could use, and in general to detect skew regressions. I do see references to --check-version-skew flag in test-infra repo but my git grep shows them all set to false.

Thank you very much for your reply. However, I have noticed that pbialon has already claimed the task related to kubelet. I would be grateful to hear your thoughts on this matter, @bobbypage @BenTheElder

[k8s-ci-robot]

@wlp1153468871: The following test failed, say /retest to rerun all failed tests or /retest-required to rerun all mandatory failed tests:

Test name	Commit	Details	Required	Rerun command
pull-kubernetes-integration	727d8ce	link	true	/test pull-kubernetes-integration

Full PR test history. Your PR dashboard. Please help us cut down on flakes by linking to an open issue when you hit one in your PR.

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. I understand the commands that are listed here.

[vinaykul]

Thank you very much for your reply. However, I have noticed that pbialon has already claimed the task related to kubelet. I would be grateful to hear your thoughts on this matter, @bobbypage @BenTheElder

That's ok, we can have two PRs. The core code change is quite simple in this but the idea is for new contributors to get a broader view of the upgrade scenario considerations and how it can be exercised in CI so it two contributors get to learn, twice as better.

[k8s-ci-robot]

PR needs rebase.

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 active contributors to adequately respond to all issues and PRs.

This bot triages PRs 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 PR is closed

You can:

• Reopen this PR with /reopen
• Mark this PR as fresh with /remove-lifecycle rotten
• Offer to help out with Issue Triage

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

/close

[k8s-ci-robot]

@k8s-triage-robot: Closed this PR.

In response to this:

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

This bot triages PRs 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 PR is closed

You can:

• Reopen this PR with /reopen
• Mark this PR as fresh with /remove-lifecycle rotten
• Offer to help out with Issue Triage

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

/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.