Pod Restarts Every 15 Minutes on GKE with ScaledObject 'metrics-api'

[RomainGIRAUDEAU]

Hey,
I'm running KEDA v2.17 on Google Kubernetes Engine (GKE) Autopilot, and I’ve observed that my deployment scales down to 1 replica after about 15 minutes, even though my API consistently returns a static value of 4.

I'm using the metrics-api trigger. The API is a dummy endpoint that always returns this payload: { "replicaCount": 4 }

I expected the deployment to stay at 4 replicas indefinitely, but it consistently scales down after a while. There are no errors in keda-operator or keda-operator-metrics-apiserver, and the API is healthy (no crashes, OOMs, etc.).

ScaledObject Configuration

apiVersion: keda.sh/v1alpha1
kind: ScaledObject
metadata:
  name: pod-scaler
  namespace: default
spec:
  scaleTargetRef:
    name: minimal-spot-server
  pollingInterval: 10
  cooldownPeriod: 60
  minReplicaCount: 0
  maxReplicaCount: 8
  triggers:
  - type: metrics-api
    metadata:
      targetValue: "1.0"
      url: http://backend.default.svc.cluster.local:80/metrics_test
      valueLocation: "replicaCount"

Why is the deployment scaling down to 1 after 15 minutes, despite the metrics-api returning a constant value of 4? Is there something I'm missing about how KEDA interprets or persists external metrics?

Thanks for your assistance!