refactor(operator): extract reusable DCD workload renderer

[julienmancuso]

Summary

Extract DCD workload rendering from DynamoComponentDeploymentReconciler into a concrete, reusable dcdWorkloadRenderer.

Overview

Rendering pod templates and Services is not itself a controller state machine, but it was previously implemented as methods on the DCD reconciler. That forced another workload implementation to either duplicate DCD rendering logic or construct a reconciler solely to call its helpers.

This change introduces a narrow composition boundary:

DCD or composite workload
    -> DCD workload renderer
        -> pod templates and Services

The controller remains responsible for reconciliation, status, watches, finalizers, readiness, and composing the rendered templates into a concrete workload resource.

Details

The package-private dcdWorkloadRenderer receives its dependencies explicitly:

• Kubernetes client.
• Operator configuration.
• Runtime feature configuration.
• Docker secret retriever.

It centralizes the existing rendering behavior for:

• Base pod templates.
• Leader and worker pod templates.
• Multinode template pairs used by LeaderWorkerSet and future composite workloads.
• Component Services.
• Checkpoint-aware pod shaping.
• ServiceAccount selection.
• Legacy worker-selector compatibility.

DynamoComponentDeploymentReconciler delegates workload rendering directly to this renderer and composes the returned templates into its existing Deployment or LeaderWorkerSet resources. No public provider interface is introduced.

The renderer deliberately preserves the existing checkpoint preparation and compatibility lookups. Separating stable-resource preparation and per-reconcile resolved state is deferred to later composition-first work.

Where should the reviewer start?

1. deploy/operator/internal/controller/dynamocomponentdeployment_renderer.go defines the extracted rendering boundary.
2. deploy/operator/internal/controller/dynamocomponentdeployment_controller.go shows the controller delegating rendering and retaining reconciliation ownership.
3. deploy/operator/internal/controller/dynamocomponentdeployment_controller_test.go checks renderer/controller resource equivalence and existing compatibility behavior.

Non-goals

This MR does not:

• Introduce a generic workload-provider interface.
• Change workload-pathway selection.
• Change DGD or DCD status behavior.
• Change rollout behavior.
• Change resource ownership or finalizers.
• Change controller watches.
• Move code into new provider packages.
• Introduce public APIs.
• Make rendering completely side-effect-free.

Compatibility

The refactor preserves:

• Generated pod specifications.
• Leader and worker role labels.
• Component and subcomponent labels.
• Service selectors and names.
• LeaderWorkerSet names and specifications.
• Checkpoint metadata and injection behavior.
• ServiceAccount selection.
• Legacy worker-selector compatibility.
• Existing create, update, delete, status, and readiness behavior.

Validation

docker buildx build --platform linux/arm64 --target linter --progress=plain --build-context snapshot=../snapshot .

docker buildx build --platform linux/arm64 --target tester --progress=plain \
  --build-context snapshot=../snapshot \
  --build-context operator=../operator \
  --build-context operator-chart=../helm/charts/platform/components/operator \
  .

Both Docker targets pass.

Related issues

• DEP: Decouple operator workload pathways from external orchestrators #12035

Summary by CodeRabbit

• Refactor

  • Centralized workload resource rendering for consistent leader, worker, service, and multinode deployment generation.
  • Improved compatibility with existing worker labels and selectors during deployment updates.

• Bug Fixes

  • Ensured generated resources remain consistent across controller and renderer pathways.
  • Preserved checkpoint, discovery, metrics, labeling, and service-account behavior during resource generation.

[datadog-official]

 

⚠️ Warnings

🚦 5 Pipeline jobs failed

PR | TRTLLM DynamoCheckpoint Deploy Test     

🔄 Retry job. This looks flaky and may succeed on retry.

v1alpha1 admission smoke test failed to become reachable: conversion webhook for nvidia.com/v1alpha1, Kind=DynamoGraphDeployment failed: connection refused to the webhook service.

PR | SGLang DynamoCheckpoint Operator Setup     

See error

Failed to create DynamoGraphDeployment due to connection refused on webhook service for v1alpha1 API version.

PR | deploy-operator     

See error

Error syncing: patch host object: update object: Operation cannot be fulfilled on pods 'grove-operator-5c999446c5-fd4k4-x-default-x-ci-30649458250': the object has been modified; please apply your changes to the latest version and try again

View all 5 failed jobs.

ℹ️ Info

🔄 Datadog auto-retried 1 job - 1 passed on retry

Useful? React with 👍 / 👎

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: 8a684fb | Docs | Datadog PR Page | Give us feedback!}

[coderabbitai]

Walkthrough

Changes

Workload rendering centralization

Layer / File(s)	Summary
Reusable workload renderer deploy/operator/internal/controller/dynamocomponentdeployment_renderer.go	Adds shared rendering for multinode pod templates, services, workload labels, checkpoint configuration, service accounts, and legacy worker selector compatibility.
Controller delegation and LWS composition deploy/operator/internal/controller/dynamocomponentdeployment_controller.go	Routes pod, service, label, component-type, and selector generation through the renderer and uses its templates for LeaderWorkerSet resources.
Renderer/controller equivalence test deploy/operator/internal/controller/dynamocomponentdeployment_controller_test.go	Compares directly rendered services and pod templates with resources composed by the controller.

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

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Title check	✅ Passed	The title is concise and accurately summarizes the main change: extracting a reusable DCD workload renderer.
Description check	✅ Passed	The description covers the required sections and adds useful details, review guidance, validation, and a related issue.

---

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

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no bugs or issues to report.