chart: api-server Deployment always includes spec.replicas when HPA is enabled

[thejoeejoee]

Official Helm Chart version

1.19.0 (latest released)

Apache Airflow version

3.1

Kubernetes Version

1.35

Helm Chart configuration

apiServer:
  replicas: 1
  hpa:
    enabled: true
    minReplicaCount: 1
    maxReplicaCount: 5
    metrics:
      - type: Resource
        resource:
          name: cpu
          target:
            type: Utilization
            averageUtilization: 80

Docker Image customizations

None — using the official image.

What happened

When apiServer.hpa.enabled: true, the Deployment manifest still contains spec.replicas: 1.
HPA modifies this field at runtime to scale the api-server. GitOps tooling continuously detects
a diff between the desired state in Git (replicas: 1) and the live cluster state (replicas
managed by HPA), causing a permanent out-of-sync condition that cannot be resolved without
disabling HPA or ignoring the field via a custom diff strategy.

What you think should happen instead

When HPA is enabled, spec.replicas should be omitted from the Deployment entirely.
This allows HPA to be the sole owner of the replica count and eliminates the spurious diff.

How to reproduce

1. Deploy the chart with apiServer.hpa.enabled: true and apiServer.replicas: 1 (defaults).
2. Observe the live Deployment — HPA immediately takes ownership and mutates spec.replicas.
3. Any GitOps reconciliation loop will report a permanent diff on spec.replicas.

Set apiServer.replicas: ~ in values.yaml as a workaround — the field is then omitted
from the rendered Deployment and HPA can manage replicas without conflict.

Anything else

The same pattern (kindIs "invalid" guard on spec.replicas) is the standard fix used by
other Helm charts to support HPA without GitOps drift.

Are you willing to submit PR?

• Yes I am willing to submit a PR!

Code of Conduct

• I agree to follow this project's Code of Conduct