apiVersion: keda.sh/v1alpha1
kind: ScaledObject
metadata:
  name: openclaw-brain-a100-scaling
  namespace: customer1
spec:
  scaleTargetRef:
    name: openclaw-brain-vllm
    kind: Deployment
  service:
    name: openclaw-brain-service
    port: 8000
  minReplicaCount: 0
  maxReplicaCount: 1
  cooldownPeriod: 900  # 15 minutes idle before scale-down
  triggers:
    - type: http
      metadata:
        path: "/v1/models"
        host: "openclaw-brain-service.customer1.svc.cluster.local"
        requestCount: "1"  # Scale up on 1+ request (wake script handles cold start)