HPA pod说明

apiVersion: autoscaling/v2beta2
kind: HorizontalPodAutoscaler
metadata:
  name: php-apache
  namespace: default
spec:
  # HPA的伸缩对象描述，HPA会动态修改该对象的pod数量
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: php-apache
  # HPA的最小pod数量和最大pod数量
  minReplicas: 1
  maxReplicas: 10
  # 监控的指标数组，支持多种类型的指标共存
  metrics:
  # Object类型的指标
  - type: Object
    object:
      metric:
        # 指标名称
        name: requests-per-second
      # 监控指标的对象描述，指标数据来源于该对象
      describedObject:
        apiVersion: networking.k8s.io/v1beta1
        kind: Ingress
        name: main-route
      # Value类型的目标值，Object类型的指标只支持Value和AverageValue类型的目标值
      target:
        type: Value
        value: 10k
  # Resource类型的指标
  - type: Resource
    resource:
      name: cpu
      # Utilization类型的目标值，Resource类型的指标只支持Utilization和AverageValue类型的目标值
      target:
        type: Utilization
        averageUtilization: 50
  # Pods类型的指标
  - type: Pods
    pods:
      metric:
        name: packets-per-second
      # AverageValue类型的目标值，Pods指标类型下只支持AverageValue类型的目标值
      target:
        type: AverageValue
        averageValue: 1k
  # External类型的指标
  - type: External
    external:
      metric:
        name: queue_messages_ready
        # 该字段与第三方的指标标签相关联，（此处官方文档有问题，正确的写法如下）
        selector:
          matchLabels:
            env: "stage"
            app: "myapp"
      # External指标类型下只支持Value和AverageValue类型的目标值
      target:
        type: AverageValue
        averageValue: 30

HPA 简单例子：

apiVersion: autoscaling/v1
kind: HorizontalPodAutoscaler
metadata:
  name: nsfw-caffe  # 名称
  namespace: test #k8s命名空间
spec:
  maxReplicas: 100  # 最大副本数
  minReplicas: 3   # 最小副本数
  scaleTargetRef:   
    apiVersion: apps/v1
    kind: Deployment  
    name: nsfw-caffe   # 监控名为nsfw-caffe的Deployment
  targetCPUUtilizationPercentage: 50  # cpu 阈值

get hpa:

kubectl -n test get hpa

查看原始deployment的resource有没有设置cpu的限制如果没有则加上：

kubectl -n test  set resources deployment/nsfw-caffe --limits=cpu=100m