controller: drain: extract node taint and pod drain primitives

[tkatila]

Upcoming GPU recovery controller needs similar functionality that exists in the GPU FW update controller now. Move drain functionality to its own file so it can be used from both controllers.

This also moves the controller to use Eviction API instead of Pod Delete.

[Copilot]

Pull request overview

Refactors the GPUFirmwareUpdate controller’s node-drain logic into reusable primitives (taint management, GPU pod detection, and eviction) in preparation for an upcoming GPU recovery controller, and switches draining from direct Pod deletion to the Kubernetes Eviction API.

Changes:

• Extracted node tainting, pod selection (GPU-only and full-drain helpers), and eviction helpers into internal/controller/drain.go.
• Updated GPUFirmwareUpdate controller to use the shared drain primitives and retry evictions during drain polling.
• Added unit tests for the new drain primitives and granted RBAC permission for pods/eviction.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
internal/controller/gpufirmwareupdate_controller.go	Replaces inline taint/pod-drain logic with shared drain helpers and switches to eviction API usage.
internal/controller/gpufirmwareupdate_controller_test.go	Adjusts tests for new drain behavior and adds an assertion that non-GPU pods are not evicted.
internal/controller/drain.go	Introduces reusable taint, pod selection, and eviction primitives (GPU-only and full-node drain building blocks).
internal/controller/drain_test.go	Adds focused unit tests for drain primitives using a fake client and interceptors.
config/rbac/role.yaml	Adds create permission for pods/eviction.
charts/gpu-base-operator/templates/role.yaml	Mirrors RBAC pods/eviction permission in the Helm chart.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.