Kubernetes VPA：targetref 选择器问题 + 最少资源

Question

我有 2 个问题： - 我的 Vertical pod autoscaler 不遵循我的最小资源策略：

Spec:
  Resource Policy:
    Container Policies:
      Min Allowed:
        Cpu:     50m        <==== mini allowed for CPU
        Memory:  75Mi
      Mode:      auto
  Target Ref:
    API Version:  extensions/v1beta1
    Kind:         Deployment
    Name:         hello-world
  Update Policy:
    Update Mode:  Auto
Status:
  Conditions:
    Last Transition Time:  2019-03-19T19:11:36Z
    Status:                True
    Type:                  RecommendationProvided
  Recommendation:
    Container Recommendations:
      Container Name:  hello-world
      Lower Bound:
        Cpu:     25m
        Memory:  262144k
      Target:
        Cpu:     25m       <==== actual CPU configured by the VPA
        Memory:  262144k

我将 VPA 配置为使用 targetref 使用新型标签选择器，但在推荐系统日志中它说我使用的是旧版：
```
Error while fetching legacy selector. Reason: v1beta1 selector not found
```

这是我的部署配置：

---
apiVersion: extensions/v1beta1
kind: Deployment
metadata:
  name: hello-world
  namespace: hello-world
  labels:
    name: hello-world
spec:
  selector:
    matchLabels:
      name: hello-world
  replicas: 2
  template:
    metadata:
      labels:
        name: hello-world
    spec:
      securityContext:
        fsGroup: 101
      containers:
        - name: hello-world
          image: xxx/hello-world:latest
          imagePullPolicy: Always
          ports:
            - containerPort: 3000
              protocol: TCP
          resources:
            limits:
              cpu: 500m
              memory: 500Mi
            requests:
              cpu: 100m
              memory: 150Mi
          volumeMounts:
          - mountPath: /u/app/www/images
            name: nfs-volume
      volumes:
      - name: nfs-volume
        persistentVolumeClaim:
          claimName: hello-world

这是我的 VPA 配置：

---
apiVersion: "autoscaling.k8s.io/v1beta2"
kind: VerticalPodAutoscaler
metadata:
  name: hello-world
  namespace: hello-world
spec:
  targetRef:
    apiVersion: "extensions/v1beta1"
    kind: Deployment
    name: hello-world
  resourcePolicy:
    containerPolicies:
    - minAllowed:
        cpu: 50m
        memory: 75Mi
      mode: auto
  updatePolicy:
    updateMode: "Auto"

我是运行 kubernetes v1.13.2，VPA v0.4，这是他的配置：

---
apiVersion: extensions/v1beta1
kind: Deployment
metadata:
  name: vpa-recommender
  namespace: kube-system
spec:
  replicas: 1
  template:
    metadata:
      labels:
        app: vpa-recommender
    spec:
      serviceAccountName: vpa-recommender
      containers:
      - name: recommender
        image: k8s.gcr.io/vpa-recommender:0.4.0
        imagePullPolicy: Always
        resources:
          limits:
            cpu: 200m
            memory: 1000Mi
          requests:
            cpu: 50m
            memory: 500Mi
        ports:
        - containerPort: 8080
        command:
        - ./recommender
        - --alsologtostderr=false
        - --logtostderr=false
        - --prometheus-address=http://prometheus-service.monitoring:9090/
        - --prometheus-cadvisor-job-name=cadvisor
        - --v=10

谢谢

Answer 1

我认为您没有使用旧的 fetcher。

这是一个code:

legacySelector, fetchLegacyErr := feeder.legacySelectorFetcher.Fetch(vpa)
if fetchLegacyErr != nil {
    glog.Errorf("Error while fetching legacy selector. Reason: %+v", fetchLegacyErr)
}
selector, fetchErr := feeder.selectorFetcher.Fetch(vpa)
if fetchErr != nil {
    glog.Errorf("Cannot get target selector from VPA's targetRef. Reason: %+v", fetchErr)
}

Autoscaler 只是尝试先获取旧版选择器，然后再使用新的。

关于资源限制。

这是源代码中的一个 comment（PodResourcePolicy 是规范中的一个 "resourcePolicy" 块）：

PodResourcePolicy controls how autoscaler computes the recommended resources for containers belonging to the pod. There can be at most one entry for every named container and optionally a single wildcard entry with containerName = '*', which handles all containers that don't have individual policies.

我认为，您还应该在规范中设置 ContainerName，因为您需要一个 pod 范围的策略：

apiVersion: "autoscaling.k8s.io/v1beta2"
kind: VerticalPodAutoscaler
metadata:
  name: hello-world
  namespace: hello-world
spec:
  targetRef:
    apiVersion: "extensions/v1beta1"
    kind: Deployment
    name: hello-world
  resourcePolicy:
    containerPolicies:
    - minAllowed:
        cpu: 50m
        memory: 75Mi
      mode: auto
      containerName: "*" # Added line
  updatePolicy:
    updateMode: "Auto"

Answer 2

出现以下错误：

Error while fetching legacy selector. Reason: v1beta1 selector not found"

已在 VPA v0.7 中修复。请参阅 this commit 以供参考。

Kubernetes VPA：targetref 选择器问题 + 最少资源

Kubernetes VPA : issue with targetref selector + minimal resources

autoscaling

kubernetes