Kubernetes(k8s)的Pod控制器Horizontal Pod Autoscaler(HPA)详细讲解

1. 概述

kubectl scale命令实现Pod的扩缩容需要人工操作。kubernetes可以通过监测Pod的使用情况，实现Pod数量的自动调整，这就是HPA控制器

HPA可以获取每个Pod的利用率，然后和HPA中定义的指标进行对比，同时计算出需要伸缩的具体值，最后实现Pod的数量的调整

2. kubernetes-sigs/metrics-server的介绍和安装

可以用来收集K8s集群中的资源使用情况。不要将metrics-server用于常规的监控方案，可以直接使用kebelet的/metrics/resource

特点：

• 每15秒进行一次数据采集
• 占用资源少，每个Node占用1 mili core的CPU和2 MB内存
• 目前只能对CPU和内存进行监控

2.1 metrics-server安装

1. 下载yaml文件。为replicas: 2的高可用服务

[root@k8s-master ~]# wget https://github.com/kubernetes-sigs/metrics-server/releases/latest/download/high-availability.yaml

当前使用的是metrics-server 0.6.1版本，支持Kubernetes 1.23.2

2. 修改yaml文件配置

[root@k8s-master ~]# cat high-availability.yaml
......省略部分......
      - args:
        - --cert-dir=/tmp
        - --secure-port=4443
        # - --kubelet-preferred-address-types=InternalIP,ExternalIP,Hostname
        # 注释上面那行，添加下面两行
        - --kubelet-preferred-address-types=InternalIP,Hostname,InternalDNS,ExternalDNS,ExternalIP    # node address类型的优先权设置
        - --kubelet-insecure-tls    # 取消安全验证
        - --kubelet-use-node-status-port
        - --metric-resolution=15s
        # image: k8s.gcr.io/metrics-server/metrics-server:v0.6.1
        # 注释上面那行，添加下面这行
        image: registry.cn-hangzhou.aliyuncs.com/google_containers/metrics-server:v0.6.1
        imagePullPolicy: IfNotPresent
        livenessProbe:
          failureThreshold: 3
......省略部分......
[root@k8s-master ~]# 

3. 部署应用

[root@k8s-master ~]# kubectl apply -f high-availability.yaml 
serviceaccount/metrics-server created
clusterrole.rbac.authorization.k8s.io/system:aggregated-metrics-reader created
clusterrole.rbac.authorization.k8s.io/system:metrics-server created
rolebinding.rbac.authorization.k8s.io/metrics-server-auth-reader created
clusterrolebinding.rbac.authorization.k8s.io/metrics-server:system:auth-delegator created
clusterrolebinding.rbac.authorization.k8s.io/system:metrics-server created
service/metrics-server created
deployment.apps/metrics-server created
Warning: policy/v1beta1 PodDisruptionBudget is deprecated in v1.21+, unavailable in v1.25+; use policy/v1 PodDisruptionBudget
poddisruptionbudget.policy/metrics-server created
apiservice.apiregistration.k8s.io/v1beta1.metrics.k8s.io created
[root@k8s-master ~]# 

2.2 查看资源使用情况

[root@k8s-master ~]# kubectl top node
NAME         CPU(cores)   CPU%   MEMORY(bytes)   MEMORY%   
k8s-master   142m         7%     1154Mi          67%       
k8s-node1    85m          4%     851Mi           49%       
k8s-node2    57m          2%     567Mi           32%       
[root@k8s-master ~]# 
[root@k8s-master ~]# kubectl top pod -n kube-system
NAME                                       CPU(cores)   MEMORY(bytes)   
calico-kube-controllers-57d95cb479-5zppz   3m           27Mi            
calico-node-2m8xb                          32m          123Mi           
calico-node-jnll4                          32m          117Mi           
calico-node-v6zcv                          31m          96Mi            
coredns-7f74c56694-snzmv                   2m           13Mi            
coredns-7f74c56694-whh84                   1m           17Mi            
etcd-k8s-master                            13m          62Mi            
kube-apiserver-k8s-master                  39m          437Mi           
kube-controller-manager-k8s-master         12m          84Mi            
kube-proxy-9gc7d                           1m           17Mi            
kube-proxy-f9w7h                           1m           30Mi            
kube-proxy-s8rwk                           1m           26Mi            
kube-scheduler-k8s-master                  3m           34Mi            
metrics-server-59755df759-n4lm7            3m           13Mi            
metrics-server-59755df759-zv86z            3m           15Mi            
[root@k8s-master ~]# 

3. 准备Deployment和Service 3.1 准备Deployment

新建pod-controller.yaml，内容如下。然后运行deployment

[root@k8s-master ~]# cat pod-controller.yaml 
apiVersion: apps/v1
kind: Deployment
metadata:
  name: pod-controller
  namespace: dev
  labels:
    controller: deploy 
spec:
  replicas: 1
  selector:
    matchLabels:
      app: nginx-pod
  template:
    metadata:
      labels:
        app: nginx-pod
    spec:
      containers:
        - name: nginx
          image: nginx:latest
          ports:
            - name: nginx-port
              containerPort: 80
              protocol: TCP
          resources: # 要设置每个容器最低使用的CPU，不然测试不出效果
            requests:
              cpu: "0.1"
[root@k8s-master ~]# 
[root@k8s-master ~]# kubectl apply -f pod-controller.yaml 
deployment.apps/pod-controller created
[root@k8s-master ~]# 

3.2 准备Service

[root@k8s-master ~]# kubectl expose deployment pod-controller --name=nginx-service --type=NodePort --port=80 --target-port=80 -n dev
service/nginx-service exposed
[root@k8s-master ~]# 
[root@k8s-master ~]# kubectl get svc -n dev
NAME            TYPE       CLUSTER-IP     EXTERNAL-IP   PORT(S)        AGE
nginx-service   NodePort   10.96.78.249           80:32469/TCP   19s
[root@k8s-master ~]#

访问：http://任意集群IP:32469

4. 部署HPA

新建pod-hpa.yaml，内容如下。然后运行HPA

[root@k8s-master ~]# cat pod-hpa.yaml
apiVersion: autoscaling/v1
kind: HorizontalPodAutoscaler
metadata:                                 
  name: pod-hpa                             # hpa的名称
  namespace: dev                            # hpa所属的命名空间
  labels:
    controller: hpa                         # 给hpa打标签
spec:
  minReplicas: 1                            # 缩容最小Pod数量
  maxReplicas: 10                           # 扩容最大Pod数量
  targetCPUUtilizationPercentage: 3         # 为了便于测试，这里设置CPU使用率警戒线为3%
  scaleTargetRef:                           # 指定要控制的Deployment的信息
    apiVersion: apps/v1
    kind: Deployment
    name: pod-controller
[root@k8s-master ~]# 
[root@k8s-master ~]# kubectl apply -f pod-hpa.yaml 
horizontalpodautoscaler.autoscaling/pod-hpa created
[root@k8s-master ~]# 

查看HPA

[root@k8s-master ~]# kubectl get hpa -n dev
NAME      REFERENCE                   TARGETS        MINPODS   MAXPODS   REPLICAS   AGE
pod-hpa   Deployment/pod-controller   0%/3%          1         10        1          82s
[root@k8s-master ~]#

5. 测试

使用Postman或者Jmeter等压力测试工具，不停的请求http://任意集群IP:32469

[root@k8s-master ~]# kubectl get deploy pod-controller -n dev -o wide
NAME             READY   UP-TO-DATE   AVAILABLE   AGE   CONTAINERS   IMAGES         SELECTOR
pod-controller   10/10   10           10          41m   nginx        nginx:latest   app=nginx-pod
[root@k8s-master ~]#

可以看到已经启动了10个Pod了。等没有请求访问nginx服务。最后只剩一个Pod了