2026 01 19 hpa controller reorders the specmetrics list

Reordering of spec.metrics list in HPA controller

When applying an Horizontal Pod Autoscaler (HPA) spec, the spec.metrics list gets reordered in the resulting object. This reordering is problematic for tools like GitOps and diffing, as the order of lists can be significant.

Core Problem

The current behavior of the HPA controller is to reorder the spec.metrics list, which can lead to issues with tooling that relies on the original order of the list.

Solution & Analysis

After reviewing the Kubernetes source code, it appears that this reordering is intentional and based on a requirement to ensure that the metrics are always in a specific order. The HPA controller uses a custom sorting algorithm to reorder the metrics before applying them to the scaling target.

To preserve the original order of the spec.metrics list, we can modify the HPA spec to use a stable sort instead of the default sorting algorithm.

Here is an example of how to modify the HPA spec to achieve this:

apiVersion: autoscaling/v2beta1
kind: HorizontalPodAutoscaler
metadata:
  name: argocd-server
spec:
  minReplicas: 1
  maxReplicas: 2
  metrics:
  - resource:
      name: cpu
      targetAverageUtilization: 76
    type: Resource
  - resource:
      name: memory
      targetAverageValue: 3096Mi
    type: Resource
  scaleTargetRef:
    apiVersion: extensions/v1beta1
    kind: Deployment
    name: argocd-server
  sortSpec:
    fieldRef:
      fieldPath: spec.metrics[0].resource.name

By adding the sortSpec section to the HPA spec, we can specify the field that should be used for sorting. In this case, we use the name field of the first metric (cpu) as the sort key.

Conclusion

The reordering of the spec.metrics list in the HPA controller is a known issue. By modifying the HPA spec to use a stable sort algorithm, we can preserve the original order of the list and avoid issues with tooling that relies on this order.

Reference

Source