I'm attempting to get autoscaling set up using custom metrics.

Currently, I have:

• Prometheus Operator running in the monitoring namespace.
• K8S Prometheus Adapter running in the custom-metrics namespace, exposed through a service named api.
• A deployment running in the my-namespace namespace.

I added a HorizontalPodAutoscaler in the custom-metrics namespace, targeting my deployment.

The YAML looks like:

apiVersion: autoscaling/v1
kind: HorizontalPodAutoscaler
metadata:
  name: my-deployment-hpa
  namespace: custom-metrics
spec:
  maxReplicas: 10
  minReplicas: 1
  scaleTargetRef:
    apiVersion: apps/v1beta2
    kind: Deployment
    name: my-deployment
  metrics:
  - type: Object
    object:
      metricName: queue_length
      targetValue: 1000
      target:
        apiVersion: extensions/v1beta1
        kind: Service
        name: api

When I describe the HorizontalPodAutoscaler via kubectl describe hpa my-deployment-hpa -n custom-metrics, I see AbleToScale: False because FailedGetScale the HPA controller was unable to get the target's current scale: deployments/scale.extensions "my-deployment" not found.

Does Kubernetes expect that the custom metrics API, the scale target and the HorizontalPodAutoscaler all exist in the same namespace?