[Bug] Deployment analyzer reports in-progress rollouts as failures #1742

Expected Behavior

A Deployment that is actively progressing (rollout or scaling in progress) should not be reported as a problem. status.readyReplicas trailing spec.replicas is a normal, healthy intermediate state while the Deployment controller is rolling out a new ReplicaSet or scaling up — image pulls and readiness probes take time, and maxSurge/maxUnavailable deliberately allow temporarily fewer ready pods.

Current Behavior

pkg/analyzer/deployment.go reports a failure whenever spec.replicas != status.readyReplicas, without inspecting the Deployment's progress state. During a normal rollout (e.g. spec.replicas: 3, status.replicas: 3, status.readyReplicas: 2), running k8sgpt analyze --filter Deployment emits:

Deployment default/web has 3 replicas but 2 are available with status running

This is a false positive: the third pod is still pulling the image / waiting on its readiness probe, and the Deployment is healthy.

The check also does not consider status.observedGeneration vs metadata.generation. When the controller has not yet reconciled the latest spec (observedGeneration < generation), status is stale relative to spec, and the replica comparison is misleading. No analyzer in pkg/analyzer/ currently consults observedGeneration.

Steps to Reproduce

Reproduces purely at the status level with the fake client already used in pkg/analyzer/deployment_test.go (no live cluster required):

1. Create a Deployment with spec.replicas: 3 and a pod template.
2. Set status.replicas: 3, status.readyReplicas: 2, status.availableReplicas: 2 — the mid-rollout shape.
3. Run the Deployment analyzer (k8sgpt analyze --filter Deployment, or the unit test harness).
4. Observe the failure text above, even though nothing is wrong.

To reproduce on a live cluster: deploy a Deployment with a readiness probe, update its image, and run k8sgpt analyze during the rollout window (seconds to minutes depending on image size and probe time).

Environment

• k8sgpt version: main
• Kubernetes version: n/a (analyzer logic; reproduced with fake client). Rollout semantics are identical across versions.
• AI Backend/Provider: n/a (analyzer path, --explain not required)
• OS/Platform: n/a

Additional Context

Current check:

if deployment.Spec.Replicas != nil && *deployment.Spec.Replicas != deployment.Status.ReadyReplicas {
    // report replica mismatch only
}

Suggested direction for the fix (open to maintainer preference; deliberately does not overlap with the Progressing=False / ProgressDeadlineExceeded reporting from #1739 / #1740, which covers the stuck rollout case):

1. ObservedGeneration guard: skip the replica comparison while status.observedGeneration < metadata.generation (controller hasn't caught up to the latest spec yet).
2. Healthy-progress check: when status.conditions show Progressing=True (with Available=True), treat the replica mismatch as in-progress and do not report; only report when the rollout is genuinely stuck or replicas are unavailable.
3. Add regression tests for: healthy rollout-in-progress, scaling-in-progress, controller-not-yet-synced (observedGeneration < generation), and a genuinely stuck rollout (to keep the existing [Bug] Deployment analyzer misses ProgressDeadlineExceeded when ReadyReplicas still match spec #1739 behavior).

Scope: single file pkg/analyzer/deployment.go + deployment_test.go.

I would like to work on this and can send a PR in that shape if the direction looks right.