apiVersion: http.keda.sh/v1alpha1
kind: HTTPScaledObject
metadata:
  name: clip-http-scaler
  namespace: bj1-dcloud
spec:
  hosts:
     - annomidware.d-robotics.cc
  pathPrefixes:
    - /classification/clip/v1/predict
    - /classification/clip/v1/embedding/
  
  # 目标 Deployment 或 StatefulSet
  scaleTargetRef:
    name: clip
    kind: Deployment
    apiVersion: apps/v1
    service: clip-svc  # 对应 HTTPRoute 中的后端服务
    port: 80

  # 副本数配置
  replicas:
    min: 0  # 支持缩容到 0（空闲时节省资源）
    max: 1  # 最大副本数（根据实际需求调整）

  # 扩缩容策略
  scalingMetric:
    # requestRate:
    #   granularity: 10s
    #   targetValue: 2  # 每秒 10 个请求触发扩容
    #   window: 1m  # 1 分钟时间窗口
    concurrency:
      targetValue: 6
  # 扩缩容行为控制
  scaledownPeriod: 300