[xianlubird]

Summary

Complete the two missing single-node vLLM Component scenarios in the operator upgrade matrix: aggregated and disaggregated. Production admission and DGD/DCD controllers must produce the complete expected Deployment resource graphs and propagate intentional edits to the requested worker generation.

Key Changes

Core logic: 0 lines. This PR adds fixture coverage only.

• Add latest/vllm/{aggregated,disaggregated}/single/component inputs and full output goldens, with runtime images fixed at 1.6.0. Use vLLM's supported --model flag, and configure NixlConnector with producer/consumer roles for the PD workers.
• Reuse the shared topology, resource-graph and target-generation worker-edit assertions already on main; no new helpers or normalization.
• Replace the README's fixed coverage count with merge-order-independent tracking. This PR alone advances canonical CREATE coverage from 23/27 to 25/27.

Validation

Validated on current main plus these fixtures using an exclusive CPU Kind v1.36.1 cluster:

Behavior	Evidence
Aggregated single-node lowering	Complete golden matched: 2 DCD-owned Deployments, correct Component provider and no Grove/LWS/DS fallback; intentional worker edit reached the target generation.
Disaggregated single-node lowering	Complete golden matched: frontend/prefill/decode DCDs and Deployments; intentional edits reached both worker components at the target generation.
Workload containment	Historical snapshot restoration and Component quota barriers passed; serving Pods and ReplicaSets remained blocked.
Matrix boundary	This branch correctly reports only the two TensorRT-LLM single-node cases missing. Combining the two independent branches passes all 27 canonical entries.

Review follow-up corrects the three --model-path arguments to vLLM's --model and supplies explicit NixlConnector producer/consumer configuration for PD (the released 1.6 wrapper requires KV transfer configuration for prefill). Regenerated only these two new latest output goldens from production-controller resource graphs. The semantic changes are the corrected args in DGD intent, DCD templates, and Deployment templates. Worker generation identities remain exactly bound to the reviewed DGD specification; no hash wildcard or new normalization is introduced.

Replayed the corrected aggregated and disaggregated CPU cases through the original admission/controller suite, including the target-generation worker-edit checks for the aggregated worker and both PD roles. Upgrade unit regressions, golden matcher tests, and all 19 fixture-tooling tests passed. Historical inputs, provenance, RECONCILED_VERSION, and all pre-existing fixture goldens remain unchanged.

This validates CPU admission/controller/API resource graphs and upgrade intent. It does not run serving Pods, model inference, or real PD KV transfer.

Issues and Ongoing Work

• Closes test(operator): add the vLLM Component upgrade matrix #17001. Part of DEP: Golden lowering and release-bound operator upgrade fixtures #16650; completing the canonical CREATE matrix does not close the DEP or the missing-origin historical follow-up.
• test(operator): cover DisaggregatedSet and multinode LWS upgrade fixtures #17013 already landed multinode Component and DisaggregatedSet coverage for all three engines.
• test(operator): cover SGLang single-node disaggregated upgrades #17398 already landed SGLang single-node PD coverage and the shared generation checks.
• test(operator): cover TensorRT-LLM single-node Component upgrades #17007 adds the other backend's two single-node cases. Both PRs now target current main independently; neither depends on the other. The former test(operator): cover SGLang single-node disaggregated upgrades #17005-based stack has been removed.

Pull Request Checklist

• I used Dynamo's /pr-checklist skill.

Summary by CodeRabbit

• Tests
  • Added upgrade-test coverage for single-component vLLM deployments in aggregated and disaggregated configurations, including expected frontend, worker, and Kubernetes resource states.
• Documentation
  • Clarified that release-matrix completeness and the missing historical-origin case remain tracked, while the release-matrix gate continues to be enforced.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[dynamo-ops]

/ok to test 3857a62

[devin-ai-integration]

✅ Dynamo PR CI passed — run 40798789461 (attempt 1) on a64ef82dfa

Gate checks: ✅ backend-status-check · ✅ deploy-status-check · ✅ dynamo-status-check

_{Posted automatically by Devin for run 40798789461. Updated on every full-CI run of this PR.}

[dynamo-ops]

/ok to test 4fb85ae

[coderabbitai]

No actionable comments were generated in the recent review.

Recent review info

Run configuration

• Source mode: Auto
• Repository: ai-dynamo/dynamo
• Branch: main
• File: .coderabbit.yaml
• Commit: 833405b
• Layers: Repository YAML
• Review profile: CHILL
• Plan: Enterprise
• Run ID: fcbab2a0-c52e-490d-981d-d501764c35a3

Commits

Reviewing files that changed from the base of the PR and between 833405b and 4fb85ae.

Files selected for processing: 5

• deploy/operator/internal/controller/testdata/upgrade/README.md
• deploy/operator/internal/controller/testdata/upgrade/latest/vllm/aggregated/single/component/input.yaml
• deploy/operator/internal/controller/testdata/upgrade/latest/vllm/aggregated/single/component/output.yaml
• deploy/operator/internal/controller/testdata/upgrade/latest/vllm/disaggregated/single/component/input.yaml
• deploy/operator/internal/controller/testdata/upgrade/latest/vllm/disaggregated/single/component/output.yaml

Included review availability: This review used your included allowance. Your plan provides up to 12 included reviews per hour; 11 remain after this review.

Walkthrough

The upgrade test data now includes aggregated and disaggregated single-node vLLM Component input and expected output manifests. The README updates its description of matrix completeness and the missing-origin historical case.

Changes

vLLM Component upgrade fixtures

Layer / File(s)	Summary
Aggregated single-node fixture deploy/operator/internal/controller/testdata/upgrade/latest/vllm/aggregated/single/component/*, deploy/operator/internal/controller/testdata/upgrade/README.md	Adds an input manifest for one frontend and two vLLM workers. The expected output captures Deployments, component and graph resources, discovery permissions, Services, and a ServiceAccount. The README updates the matrix status.
Disaggregated single-node fixture deploy/operator/internal/controller/testdata/upgrade/latest/vllm/disaggregated/single/component/*	Adds an input manifest for one frontend and two each of prefill and decode workers. The expected output captures their Deployments, component and graph resources, discovery permissions, Services, and a ServiceAccount.

Priority: Low

Estimated code review effort: 3 (Moderate) | ~20 minutes

Merge Risk: Minimal · up to 4fb85

No specific issue in these single-node upgrade fixtures currently blocks merging; complete the normal validation checks before merging.

Pre-merge checks: 4

Check name	Status	Explanation
Title check		The title "test(operator): cover vLLM single-node Component upgrades" matches the main change: fixture coverage for the aggregated and disaggregated single-node vLLM Component upgrade scenarios. It is…
Linked Issues check		Issue #17001 requires four vLLM Component fixtures. The head contains aggregated and disaggregated multinode Component fixtures that were present at the merge base, plus the two new single-node Compon…
Out of Scope Changes check		The changes stay within issue #17001. The four new fixture files add the two requested single-node cases. The README change updates matrix coverage tracking for these fixtures. No production controlle…
Description check		The description provides a clear overview under “Summary,” detailed key changes, validation evidence and limits, ongoing work, issue references, and a completed checklist. It does not use the template…

Full details: Title check

Explanation

The title "test(operator): cover vLLM single-node Component upgrades" matches the main change: fixture coverage for the aggregated and disaggregated single-node vLLM Component upgrade scenarios. It is concise and specific, and it follows a conventional commit style. The PR adds only test fixtures (inputs and output goldens) and a README update, with no core logic changes, which the title accurately reflects.

Full details: Linked Issues check

Explanation

Issue #17001 requires four vLLM Component fixtures. The head contains aggregated and disaggregated multinode Component fixtures that were present at the merge base, plus the two new single-node Component fixtures. The new input and output goldens capture the DGD, DCD, and immediate Deployment resource graphs. The documented shared topology, intentional-edit, and workload-containment checks cover the required test behavior.

Full details: Out of Scope Changes check

Explanation

The changes stay within issue #17001. The four new fixture files add the two requested single-node cases. The README change updates matrix coverage tracking for these fixtures. No production controller logic or unrelated resource changes are included.

Full details: Description check

Explanation

The description provides a clear overview under “Summary,” detailed key changes, validation evidence and limits, ongoing work, issue references, and a completed checklist. It does not use the template’s exact “Overview” heading, but it contains the required information.

Finishing Touches: 💡 1

🛠️ Fix failing CI checks: 💡

• Commit to this branch
• Create a new PR

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

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

[sttts]

One fixture-validity finding: use the supported vLLM model flag in both new single-node scenarios. Details inline.

[dynamo-ops]

/ok to test a64ef82