[quality] pkg/k8s GPU health subsystem has zero test coverage — 5 pure helpers untested

[clubanderson]

Finding

The GPU health and discovery subsystem in pkg/k8s contains ~32KB of untested code across 4 files:

File	Size	Description
client_gpu_discovery.go	10,864 B	Multi-vendor GPU node discovery (NVIDIA, AMD, Intel, Google TPU, IBM AIU)
client_gpu_health.go	9,210 B	GPU node health monitoring with 7 health checks
client_gpu_types.go	6,718 B	Type definitions for GPU health structures
client_gpu_nvidia.go	5,564 B	NVIDIA operator status inspection

Pure helper functions with zero coverage:

1. checkOperatorPod() (client_gpu_health.go) — inspects pod status for GPU operator DaemonSet pods. Checks for CrashLoopBackOff, non-Running state, and pod-not-found scenarios.
2. isStuckPod() (client_gpu_health.go) — determines if a pod is stuck via 3 conditions: ContainerStatusUnknown, terminating >5min, pending >10min.
3. deriveGPUNodeStatus() (client_gpu_health.go) — derives overall health (healthy/degraded/unhealthy) from check results with critical vs non-critical classification.
4. unstructuredNestedMap() (client_gpu_nvidia.go) — traverses nested unstructured K8s objects.
5. unstructuredNestedSlice() (client_gpu_nvidia.go) — traverses nested unstructured K8s objects for slices.

Why this matters:

1. The GPU health endpoint drives the console UI's GPU dashboard — bugs in deriveGPUNodeStatus or isStuckPod silently misclassify node health
2. checkOperatorPod handles edge cases (CrashLoopBackOff, missing pods) that determine whether operators see alerts
3. unstructuredNestedMap/Slice parse arbitrary CRD structures — nil panics here crash the handler
4. These are all pure functions that can be unit-tested with constructed inputs (no K8s client mocks needed)

Recommendation

1. Add table-driven tests for isStuckPod covering all 3 stuck conditions + happy path
2. Add tests for deriveGPUNodeStatus covering healthy/degraded/unhealthy transitions
3. Add tests for checkOperatorPod covering Running, CrashLoopBackOff, Pending, not-found
4. Add tests for unstructuredNestedMap/unstructuredNestedSlice covering nil, missing keys, valid paths

Priority

• Impact: high (GPU dashboard correctness, silent health misclassification)
• Effort: low (all pure functions, no mocks needed)

---

Filed by quality agent (ACMM L4/L6 — full mode)

[github-actions]

🐝 Hi @clubanderson! I'm kubestellar-hive[bot], an automation bot for this repo.

Trusted users — org members and contributors with write access — can mention @kubestellar-hive in a comment to trigger repo automation.
On issues, that mention queues an automated fix attempt. On pull requests, it records extra context for existing automation.
This is not an interactive Q&A bot, so mentions should be treated as requests for automation rather than a conversation.

Automation may take a moment to start, and follow-up happens through workflow activity rather than chat replies.

[clubanderson]

Quality agent review: PR #19556 now covers the GPU health helpers (deriveGPUNodeStatus, isStuckPod, checkOperatorPod, unstructuredNestedMap/Slice, IsGPUResourceName, SumGPURequested) with 33 tests. Combined with #19553's kagent helper tests, these two PRs address the most impactful zero-coverage packages in the Go backend.

The pkg/k8s directory still has additional untested files (client_core_resources.go at 10.8KB, client_events.go at 5.3KB, client_storage.go at 4.3KB) — those contain K8s client methods that require more involved mocking and are lower priority since they're thin wrappers around the K8s API.

[github-actions]

⚠️ Post-merge Playwright verification failed after PR #19567 merged. The fix shipped, but Playwright flagged an issue — see test report. A regression issue will be filed for follow-up.