[YQ-Wang]

What type of PR is this?

/kind bug

What this PR does / why we need it:

TAS summarizes grouped PodSet capacity using the smallest leader-placement penalty among the available domains. During downward assignment, however, the leader domain was selected from the existing worker-capacity order without checking that its penalty fit within the available worker slack. This could consume more capacity than the summary accounted for, leave worker pods unassigned, and return an empty topology assignment even though a feasible placement existed.

This change preserves that summarized invariant during domain selection and hierarchy descent. It prioritizes a leader-capable domain only when its capacity penalty fits within the available slack, applies the same constraint to BestFit and LeastFreeCapacity selection, and uses one consistent partial-consumption path for sliced and non-sliced assignments.

The regression scenario has three hosts with worker capacities 2, 4, and 3, one 1-CPU leader, and four 2-CPU workers. The leader is placed on the 3-capacity host, allowing the workers to be assigned 1/2/1 across the hosts.

AI assistance was used to help implement and review this change. The author reviewed and tested the resulting code and takes responsibility for it.

Which issue(s) this PR fixes:

Fixes #7446

Special notes for your reviewer:

Validation completed:

• gofmt check on all changed Go files
• git diff --check HEAD
• go test -race ./pkg/cache/scheduler -run 'TestFindTopologyAssignments|TestFindLevelWithFitDomainsPreservesFeasibleGroupedPlacements|TestUpdateCountsToMinimumGenericPreservesFeasibleGroupedPlacements' -count=1
• go test ./pkg/cache/... ./pkg/scheduler/...
• go vet ./pkg/cache/... ./pkg/scheduler/...
• Local kind A/B validation on Kubernetes v1.36.1 with isolated 2/4/3-capacity worker nodes: the clean base repeatedly failed with remainingCount: 1, while the patched controller admitted the workload in one scheduling cycle and reserved all nine CPU slots. Restoring the patched image after the base reproduction admitted the workload again.

make verify could not be run locally because this macOS environment does not have GNU sed; the directly relevant formatting, unit, race, and vet checks above all passed.

Does this PR introduce a user-facing change?

TAS: Fixed a bug that could prevent admission of otherwise feasible LeaderWorkerSet workloads when the selected leader domain reduced the capacity available to workers. For example, a 1-CPU leader and four 2-CPU workers can now be placed across 2-, 4-, and 3-CPU hosts in one rack by assigning the leader to the 3-CPU host, leaving capacity for all four workers.

Summary by CodeRabbit

• Bug Fixes
  • Improved grouped workload placement so leader and worker workloads respect available capacity across topology domains.
  • Corrected placement decisions for preferred, unconstrained, and slice-based topology configurations.
  • Improved handling of leader resource usage when selecting domains and calculating remaining capacity.
  • Ensured workloads are distributed across appropriate domains while preserving placement constraints.

[netlify]

✅ Deploy Preview for kubernetes-sigs-kueue canceled.

Name	Link
🔨 Latest commit	0d5f099
🔍 Latest deploy log	https://app.netlify.com/projects/kubernetes-sigs-kueue/deploys/6a6324f5289f7400086d7e61

[coderabbitai]

📝 Walkthrough

Walkthrough

TAS grouped PodSet placement now incorporates leader capacity during domain selection, topology descent, and capacity consumption. Five table-driven scenarios validate leader and worker assignments across hostname, slice, preferred, mixed-profile, and below-slice constraints.

Changes

TAS grouped placement

Layer / File(s)	Summary
Leader-aware domain selection pkg/cache/scheduler/tas_flavor_snapshot.go	Best-fit and least-free-capacity selection filter domains by leader capacity and use leader-aware slice capacity calculations.
Leader-prioritized capacity consumption pkg/cache/scheduler/tas_flavor_snapshot.go	Domain descent prioritizes suitable leader domains, while leader and worker capacity counters are consumed consistently.
Grouped placement validation pkg/cache/scheduler/tas_cache_test.go	Five grouped-podset scenarios validate capacity preservation and expected assignments across topology and TAS profile variations.

Estimated code review effort: 4 (Complex) | ~45 minutes

Sequence Diagram(s)

sequenceDiagram
  participant findTopologyAssignment
  participant findBestFitDomainBy
  participant prioritizeLeaderDomain
  participant consumeWithLeadersGeneric
  findTopologyAssignment->>findBestFitDomainBy: select domain with sufficient leader capacity
  findTopologyAssignment->>prioritizeLeaderDomain: order fitting domains for leader placement
  prioritizeLeaderDomain->>consumeWithLeadersGeneric: consume leader-adjusted capacity

Loading

Possibly related PRs

• kubernetes-sigs/kueue#12436: Both adjust TAS leader/worker capacity accounting and grouped placement tests.
• kubernetes-sigs/kueue#13027: Both modify TAS domain selection behavior and TestFindTopologyAssignments expectations.

Suggested labels: area/tas

Suggested reviewers: amy, mbobrovskyi, tenzen-y, kshalot

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Linked Issues check	✅ Passed	The code and tests address #7446 by making leader-aware domain selection preserve worker capacity during grouped PodSet descent.
Out of Scope Changes check	✅ Passed	Changes stay focused on TAS placement logic and regression tests for grouped leader/worker assignment.
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly and concisely describes the PR's main change: preserving capacity during grouped leader placement.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[kubernetes-prow]

Hi @YQ-Wang. Thanks for your PR.

I'm waiting for a kubernetes-sigs member to verify that this patch is reasonable to test. If it is, they should reply with /ok-to-test on its own line. Until that is done, I will not automatically test new commits in this PR, but the usual testing commands by org members will still work.

Regular contributors should join the org to skip this step.

Once the patch is verified, the new status will be reflected by the ok-to-test label.

I understand the commands that are listed here.

Details

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-sigs/prow repository.

[tenzen-y]

/ok-to-test

[YQ-Wang]

/retest

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@pkg/cache/scheduler/tas_flavor_snapshot.go`:
- Around line 1362-1386: The fallback in findBestFitDomainBy must not select a
domain with domain.leaderState below leaderCount when no candidate meets the
state requirement. Preserve the already-selected leader-feasible domain from the
caller or otherwise pass it into this selection path, and return it instead of
candidates[0] whenever the affinity-tier candidates exclude all leader-feasible
domains.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 79aeb670-33e7-45a5-b0fb-5748d8916a0c

📥 Commits

Reviewing files that changed from the base of the PR and between a7afef6 and 1e6f436.

📒 Files selected for processing (3)

• pkg/cache/scheduler/tas_cache_test.go
• pkg/cache/scheduler/tas_flavor_snapshot.go
• pkg/cache/scheduler/tas_flavor_snapshot_grouped_placement_test.go

[mimowo]

cc @pajakd ptal

[mimowo]

/area tas

[mimowo]

/release-note-edit

TAS: Fixed a bug that could prevent admission of otherwise feasible LeaderWorkerSet workloads when the selected leader domain reduced the capacity available to workers. For example, a 1-CPU leader and four 2-CPU workers can now be placed across 2-, 4-, and 3-CPU hosts in one rack by assigning the leader to the 3-CPU host, leaving capacity for all four workers.

[mimowo]

/lgtm
/approve
/cherrypick release-0.19
/cherrypick release-0.18
Thank you for fixing this epic TAS bug for PodGroups 👍

[k8s-infra-cherrypick-robot]

@mimowo: once the present PR merges, I will cherry-pick it on top of release-0.18, release-0.19 in new PRs and assign them to you.

Details

In response to this:

/lgtm
/approve
/cherrypick release-0.19
/cherrypick release-0.18
Thank you for fixing this epic TAS bug for PodGroups 👍

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-sigs/prow repository.

[kubernetes-prow]

LGTM label has been added.

Details

Git tree hash: 8590618300c96866ce84dcc5e811a1be880bec54

[kubernetes-prow]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: mimowo, YQ-Wang

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

The pull request process is described here

Details

Needs approval from an approver in each of these files:

• OWNERS [mimowo]

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

[k8s-infra-cherrypick-robot]

@mimowo: new pull request created: #13766

Details

In response to this:

/lgtm
/approve
/cherrypick release-0.19
/cherrypick release-0.18
Thank you for fixing this epic TAS bug for PodGroups 👍

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-sigs/prow repository.

[k8s-infra-cherrypick-robot]

@mimowo: new pull request created: #13767

Details

In response to this:

/lgtm
/approve
/cherrypick release-0.19
/cherrypick release-0.18
Thank you for fixing this epic TAS bug for PodGroups 👍

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-sigs/prow repository.