KEDA Autoscaler with Latency and CPU Utilization

fastapi-scaledobject.yaml

apiVersion: keda.sh/v1alpha1
kind: ScaledObject
metadata:
  name: fastapi-latency-autoscaler
  namespace: default
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: model  # update to your actual deployment name
  minReplicaCount: 1
  maxReplicaCount: 5
  pollingInterval: 30  # seconds
  cooldownPeriod: 300  # seconds before scaling down
  triggers:
    - type: prometheus
      metadata:
        serverAddress: http://prom-kube-prometheus-stack-prometheus.monitoring.svc:9090
        metricName: fastapi_latency_p95
        query: |
          histogram_quantile(0.95, sum(rate(http_request_duration_seconds_bucket[1m])) by (le))
        threshold: "0.5"
    - type: cpu
      metricType: Utilization # Allowed types are 'Utilization' or 'AverageValue'
      metadata:
        value: "50"