HPA 缩放在Spring启动期间触发。HPA 中的当前 CPU 指标很高，尽管节点级别的 CPU 利用率较低

Question

我们在 GKE 上有一个启用了自动缩放 （HPA） 的Spring启动应用程序。在启动期间，HPA 会启动并开始缩放 Pod，即使没有流量也是如此。“kubectl 得到 hpa”的结果显示当前 CPU 平均利用率很高，而节点和 POD 的 CPU 利用率非常低。在纵向扩展期间，行为是相同的，并且最终会创建多个 Pod，从而导致 Node 扩展。

应用部署Yaml：

apiVersion: apps/v1
kind: Deployment
metadata:
  name: myapp-api
  labels:
    app: myapp
spec:
  replicas: 3
  selector:
    matchLabels:
      app: myapp
  template:
    metadata:
      labels:
        app: myapp
    spec:
      serviceAccount: myapp-ksa
      containers:
      - name: myapp
        image: gcr.io/project/myapp:126
        env:
        - name: DB_USER
          valueFrom:
            secretKeyRef:
              name: my-db-credentials
              key: username
        - name: DB_PASS
          valueFrom:
            secretKeyRef:
              name: my-db-credentials
              key: password
        - name: DB_NAME
          valueFrom:
            secretKeyRef:
              name: my-db-credentials
              key: database
        - name: INSTANCE_CONNECTION
          valueFrom:
            configMapKeyRef:
              name: connectionname
              key: connectionname
        resources:
          requests:
            cpu: "200m"
            memory: "256Mi"
        ports:
        - containerPort: 8080
        livenessProbe:
          httpGet:
            path: /actuator/health
            port: 8080
          initialDelaySeconds: 90
          periodSeconds: 5
        readinessProbe:
          httpGet:
            path: /actuator/health
            port: 8080
          initialDelaySeconds: 60
          periodSeconds: 10
      - name: cloudsql-proxy
        image: gcr.io/cloudsql-docker/gce-proxy:1.17
        env:
          - name: INSTANCE_CONNECTION
            valueFrom:
              configMapKeyRef:
                name: connectionname
                key: connectionname
        command: ["/cloud_sql_proxy",
                  "-ip_address_types=PRIVATE",
                  "-instances=$(INSTANCE_CONNECTION)=tcp:5432"]
        securityContext:
          runAsNonRoot: true
          runAsUser: 2
          allowPrivilegeEscalation: false
        resources:
          requests:
            memory: "128Mi"
            cpu:    "100m"

用于HPA的Yaml：

apiVersion: autoscaling/v2beta2
kind: HorizontalPodAutoscaler
metadata:
  name: myapp-api
  labels:
    app: myapp
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: myapp-api
  minReplicas: 1
  maxReplicas: 10
  metrics:
  - type: Resource
    resource:
      name: cpu
      target:
        type: Utilization
        averageUtilization: 80
  - type: Resource
    resource:
      name: memory
      target: 
        type: Utilization
        averageUtilization: 80

各种命令的结果：

$ kubectl get pods

$ kubectl get hpa

$ kubectl top nodes

$ kubectl top pods --all-namespaces

Answer 1

由于问题已经在评论部分得到解决，我决定提供一个社区Wiki答案，只是为了更好地了解其他社区成员。

问题中的kubectl get hpa命令显示导致Pods缩放的高当前内存利用率。

从kubectl get hpa命令读取TARGETScolumn可能会令人困惑：
注意：33%的值适用于内存或cpu？

$ kubectl get hpa app-1
NAME    REFERENCE          TARGETS           MINPODS   MAXPODS   REPLICAS   AGE
app-1   Deployment/app-1   33%/80%, 4%/80%   2         5         5          5m37s

我推荐使用< code>kubectl描述hpa

$ kubectl describe hpa app-1 | grep -A 2 "Metrics:"
Metrics:                                                  ( current / target )
  resource memory on pods  (as a percentage of request):  33% (3506176) / 80%
  resource cpu on pods  (as a percentage of request):     4% (0) / 80%