feat: harden deletion with timeout, namespace cleanup, and force-remo…

[vrdc-sap]

…ve on stuck uninstall

Description

Changes proposed in this pull request:

• harden deletion with timeout, namespace cleanup, and force-remove on stuck uninstall

Related issue(s)
#56

[hyperspace-insights]

Summary

The following content is AI-generated and provides a summary of the pull request:

⚠️ Warnings:

• Could not get issue kyma-project/gpu#123. Status: 404 - UnknownObjectException

---

Harden GPU Operator Deletion with Timeout, Namespace Cleanup, and Force-Remove on Stuck Uninstall

New Feature

🛡️ Hardens the GPU Operator deletion flow by introducing a configurable uninstall timeout, explicit gpu-operator namespace cleanup after Helm uninstall, and a force-remove fallback for the finalizer when Helm gets stuck.

Changes

• internal/helm/interface.go: Updated the Installer interface — Uninstall now accepts a time.Duration timeout parameter.
• internal/helm/installer.go: Updated Uninstall to enable Wait: true and pass the timeout to Helm's uninstall action, so the controller waits for pods to terminate before proceeding.
• internal/controller/gpu_controller.go:
  • Added a deleteTimeout constant (3 minutes) passed to Uninstall.
  • On timeout (context.DeadlineExceeded), the finalizer is force-removed so the CR is not stuck forever; a warning is logged prompting manual namespace cleanup.
  • Added deleteNamespace() to explicitly delete the gpu-operator namespace (with foreground propagation) after a successful Helm uninstall, since Helm does not clean up namespaces on its own.
  • Extracted removeFinalizer() as a shared helper used by both the success and timeout paths.
  • Added isTimeoutError() helper to detect context.DeadlineExceeded.
• internal/controller/gpu_controller_test.go: Expanded deletion tests into a BeforeEach-scoped suite with four new cases:
  • Non-timeout Helm failure keeps the finalizer and returns an error.
  • Helm timeout force-removes the finalizer without blocking the CR.
  • Successful uninstall triggers namespace deletion.
  • Missing namespace is ignored gracefully (idempotent).
• config/rbac/role.yaml: Added delete verb to the namespaces resource permissions to allow the controller to clean up the namespace.

---

• 🔄 Regenerate and Update Summary
• ✏️ Insert as PR Description (deletes this comment)
• 🗑️ Delete comment

PR Bot Information

Version: 1.21.0

• GithubContextProvider: fix: remove namespace field from cluster-scoped Gpu sample CR #43
• GithubContextProvider: feat: add helm sdk installer with install, upgrade, and uninstall #33
• Output Template: Default Template
• LLM: anthropic--claude-4.6-sonnet
• Event Trigger: pull_request.opened
• File Content Strategy: Full file content
• Correlation ID: a7b14c00-3239-4bba-8ce8-f8e10304b63f
• Summary Prompt: Default Prompt

[hyperspace-insights]

⚠️ Warnings:

• Could not get issue kyma-project/gpu#123. Status: 404 - UnknownObjectException

---

The PR hardens the deletion path with a Helm uninstall timeout, force-finalizer removal on stuck uninstalls, and explicit namespace cleanup. The logic is mostly sound, but there are a few substantive issues: the timeout path skips namespace cleanup (contradicting the PR's goal), removeFinalizer doesn't guard against NotFound on re-fetch, isTimeoutError can false-positive on reconciler-context cancellation, the Helm context.Context parameter is silently discarded with no cancellation path during the blocking wait, and the namespace-deletion test is missing an assertion that Uninstall was actually invoked.

PR Bot Information

Version: 1.21.0

• Agent Instructions:
  • CLAUDE.md
• LLM: anthropic--claude-4.6-sonnet
• Event Trigger: pull_request.opened
• File Content Strategy: Full file content
• Correlation ID: a7b14c00-3239-4bba-8ce8-f8e10304b63f