apiVersion: keda.sh/v1alpha1
kind: ScaledObject
metadata:
  name: dify-worker
  namespace: dify
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: worker
  minReplicaCount: 1
  maxReplicaCount: 3
  pollingInterval: 15
  cooldownPeriod: 60
  triggers:
    - type: prometheus
      metadata:
        serverAddress: http://prometheus.observability.svc.cluster.local:9090
        metricName: vllm_num_requests_waiting
        query: 'sum(vllm:num_requests_waiting{model_name="qwen3.6-27b"})'
        threshold: "10"
        activationThreshold: "10"