Kubernetes的Rolling Update实战
一.背景
随着Kubernetes的进一步深入使用,我们越来越体会到它给我们的工作带来的高效与便利。Rolling Update是Kubernetes系统中的一个强大的功能,能够为我们的运维工作带来极大的便利。
二.步骤
2.1 部署最初始版本Deployment。
初始Deployment的YAML如下:
apiVersion: apps/v1beta1kind: Deploymentmetadata: name: deployment-rolloutspec: replicas: 4 template: metadata: labels: test: httpd spec: containers: - name: httpd-rollout-test image: httpd:2.2.31 ports: - containerPort: 80
执行命令:
[root@k8s-m ~]# kubectl get deployment -o wideNAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE CONTAINERS IMAGES SELECTORdeployment-rollout 4 4 4 4 107s httpd-rollout-test httpd:2.2.31 test=httpd[root@k8s-m ~]# kubectl get replicaset -o wideNAME DESIRED CURRENT READY AGE CONTAINERS IMAGES SELECTORdeployment-rollout-5fb9c69c5c 4 4 4 116s httpd-rollout-test httpd:2.2.31 pod-template-hash=5fb9c69c5c,test=httpd
2.2 将初始版本升级。
升级到2.4.31,YAML如下:
apiVersion: apps/v1beta1kind: Deploymentmetadata: name: deployment-rolloutspec: revisionHistoryLimit: 10 replicas: 4 template: metadata: labels: test: httpd spec: containers: - name: httpd-rollout-test image: httpd:2.4.33 ports: - containerPort: 80
执行命令:
[root@k8s-m ~]# kubectl apply -f Httpd-Deployment-rollout-v1.yamldeployment.apps/deployment-rollout configured[root@k8s-m ~]# kubectl get deployment -o wideNAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE CONTAINERS IMAGES SELECTORdeployment-rollout 4 5 2 3 3m57s httpd-rollout-test httpd:2.4.33 test=httpd[root@k8s-m ~]# kubectl get replicaset -o wideNAME DESIRED CURRENT READY AGE CONTAINERS IMAGES SELECTORdeployment-rollout-54766f574f 2 2 0 93s httpd-rollout-test httpd:2.4.33 pod-template-hash=54766f574f,test=httpddeployment-rollout-5fb9c69c5c 3 3 3 4m12s httpd-rollout-test httpd:2.2.31 pod-template-hash=5fb9c69c5c,test=httpd[root@k8s-m ~]# kubectl get pod -o wideNAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODEdeployment-rollout-54766f574f-668pc 0/1 ContainerCreating 0 114s k8s-n1 deployment-rollout-54766f574f-ssz67 0/1 ContainerCreating 0 115s k8s-n2 deployment-rollout-5fb9c69c5c-m5dvs 1/1 Running 0 4m34s 10.244.2.47 k8s-n2 deployment-rollout-5fb9c69c5c-p9grr 1/1 Running 0 4m34s 10.244.1.113 k8s-n1 deployment-rollout-5fb9c69c5c-tzm7w 1/1 Running 0 4m34s 10.244.2.48 k8s-n2
一会儿后再观察一下:
[root@k8s-m ~]# kubectl get replicaset -o wideNAME DESIRED CURRENT READY AGE CONTAINERS IMAGES SELECTORdeployment-rollout-54766f574f 4 4 4 4m24s httpd-rollout-test httpd:2.4.33 pod-template-hash=54766f574f,test=httpddeployment-rollout-5fb9c69c5c 0 0 0 7m3s httpd-rollout-test httpd:2.2.31 pod-template-hash=5fb9c69c5c,test=httpd[root@k8s-m ~]# kubectl get deployment -o wideNAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE CONTAINERS IMAGES SELECTORdeployment-rollout 4 4 4 4 7m18s httpd-rollout-test httpd:2.4.33 test=httpd[root@k8s-m ~]# kubectl get pod -o wideNAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODEdeployment-rollout-54766f574f-668pc 1/1 Running 0 4m48s 10.244.1.114 k8s-n1 deployment-rollout-54766f574f-h7gf9 1/1 Running 0 99s 10.244.2.50 k8s-n2 deployment-rollout-54766f574f-ssz67 1/1 Running 0 4m49s 10.244.2.49 k8s-n2 deployment-rollout-54766f574f-vklvr 1/1 Running 0 41s 10.244.1.115 k8s-n1
在此过程中,我们会发现,ReplicaSet deployment-rollout-5fb9c69c5c逐步被deployment-rollout-54766f574f取代,创建新版本的Pod之后,原始版本的Pod被终止。这是一个平滑的升级过程。Kubernetes提供了两个参数maxSurge和maxUnavailable来精细控制Pod的替换数量。
2.3 版本的回滚。
默认情况下,Kubernetes只会保留最近几个revision。在上面的升级过程中,通过revisionHistoryLimit设置了revision的数量。我们通过命令来看一下:
[root@k8s-m ~]# kubectl rollout history deployment deployment-rolloutdeployment.extensions/deployment-rolloutREVISION CHANGE-CAUSE1 2
继续升级。新版的YAML如下:
apiVersion: apps/v1beta1kind: Deploymentmetadata: name: deployment-rolloutspec: revisionHistoryLimit: 10 replicas: 4 template: metadata: labels: test: httpd spec: containers: - name: httpd-rollout-test image: httpd:2.4.34 ports: - containerPort: 80
执行升级并记录升级日志。
[root@k8s-m ~]# kubectl apply -f Httpd-Deployment-rollout-v2.yaml --recorddeployment.apps/deployment-rollout configured[root@k8s-m ~]# kubectl rollout history deployment deployment-rolloutdeployment.extensions/deployment-rolloutREVISION CHANGE-CAUSE1 2 3 kubectl apply --filename=Httpd-Deployment-rollout-v2.yaml --record=true[root@k8s-m ~]# kubectl get deployment deployment-rollout -o wideNAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE CONTAINERS IMAGES SELECTORdeployment-rollout 4 4 4 4 31m httpd-rollout-test httpd:2.4.34 test=httpd[root@k8s-m ~]# kubectl get replicaset -o wideNAME DESIRED CURRENT READY AGE CONTAINERS IMAGES SELECTORdeployment-rollout-54766f574f 0 0 0 29m httpd-rollout-test httpd:2.4.33 pod-template-hash=54766f574f,test=httpddeployment-rollout-54fc66bb6 4 4 4 6m12s httpd-rollout-test httpd:2.4.34 pod-template-hash=54fc66bb6,test=httpddeployment-rollout-5fb9c69c5c 0 0 0 32m httpd-rollout-test httpd:2.2.31 pod-template-hash=5fb9c69c5c,test=httpd[root@k8s-m ~]# kubectl get pod -o wideNAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODEdeployment-rollout-54fc66bb6-2zvhk 1/1 Running 0 6m26s 10.244.1.116 k8s-n1 deployment-rollout-54fc66bb6-7rsj6 1/1 Running 0 4m16s 10.244.1.117 k8s-n1 deployment-rollout-54fc66bb6-nlx4h 1/1 Running 0 6m26s 10.244.2.51 k8s-n2 deployment-rollout-54fc66bb6-xxnrg 1/1 Running 0 6m24s 10.244.2.52 k8s-n2
修改YAML文件中镜像的版本继续升级,观察升级结果:
[root@k8s-m ~]# kubectl rollout history deployment deployment-rolloutdeployment.extensions/deployment-rolloutREVISION CHANGE-CAUSE1 2 3 kubectl apply --filename=Httpd-Deployment-rollout-v2.yaml --record=true4 kubectl apply --filename=Httpd-Deployment-rollout-v3.yaml --record=true[root@k8s-m ~]# kubectl get replicaset -o wideNAME DESIRED CURRENT READY AGE CONTAINERS IMAGES SELECTORdeployment-rollout-54766f574f 0 0 0 33m httpd-rollout-test httpd:2.4.33 pod-template-hash=54766f574f,test=httpddeployment-rollout-54fc66bb6 0 0 0 10m httpd-rollout-test httpd:2.4.34 pod-template-hash=54fc66bb6,test=httpddeployment-rollout-5b99bbfbbc 4 4 4 2m33s httpd-rollout-test httpd:2.4.35 pod-template-hash=5b99bbfbbc,test=httpddeployment-rollout-5fb9c69c5c 0 0 0 36m httpd-rollout-test httpd:2.2.31 pod-template-hash=5fb9c69c5c,test=httpd[root@k8s-m ~]# kubectl get pod -o wideNAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODEdeployment-rollout-5b99bbfbbc-7mdg9 1/1 Running 0 65s 10.244.1.119 k8s-n1 deployment-rollout-5b99bbfbbc-9mfbk 1/1 Running 0 2m40s 10.244.1.118 k8s-n1 deployment-rollout-5b99bbfbbc-jn5sb 1/1 Running 0 2m40s 10.244.2.53 k8s-n2 deployment-rollout-5b99bbfbbc-tr6t2 1/1 Running 0 2m38s 10.244.2.54 k8s-n2
httpd的image均已经升级到2.4.35,接下来我们试一下回滚到v2版本:
[root@k8s-m ~]# kubectl rollout undo deployment deployment-rollout --to-revision=3deployment.extensions/deployment-rollout[root@k8s-m ~]# kubectl get deployment -o wideNAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE CONTAINERS IMAGES SELECTORdeployment-rollout 4 4 4 4 42m httpd-rollout-test httpd:2.4.34 test=httpd[root@k8s-m ~]# kubectl get replicaset -o wideNAME DESIRED CURRENT READY AGE CONTAINERS IMAGES SELECTORdeployment-rollout-54766f574f 0 0 0 39m httpd-rollout-test httpd:2.4.33 pod-template-hash=54766f574f,test=httpddeployment-rollout-54fc66bb6 4 4 4 16m httpd-rollout-test httpd:2.4.34 pod-template-hash=54fc66bb6,test=httpddeployment-rollout-5b99bbfbbc 0 0 0 8m35s httpd-rollout-test httpd:2.4.35 pod-template-hash=5b99bbfbbc,test=httpddeployment-rollout-5fb9c69c5c 0 0 0 42m httpd-rollout-test httpd:2.2.31 pod-template-hash=5fb9c69c5c,test=httpd[root@k8s-m ~]# kubectl get pod -o wideNAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODEdeployment-rollout-54fc66bb6-bjskn 1/1 Running 0 68s 10.244.1.121 k8s-n1 deployment-rollout-54fc66bb6-c65th 1/1 Running 0 70s 10.244.2.55 k8s-n2 deployment-rollout-54fc66bb6-cwpbk 1/1 Running 0 70s 10.244.1.120 k8s-n1 deployment-rollout-54fc66bb6-lz9xt 1/1 Running 0 68s 10.244.2.56 k8s-n2 [root@k8s-m ~]# kubectl rollout history deployment deployment-rolloutdeployment.extensions/deployment-rolloutREVISION CHANGE-CAUSE1 2 4 kubectl apply --filename=Httpd-Deployment-rollout-v3.yaml --record=true5 kubectl apply --filename=Httpd-Deployment-rollout-v2.yaml --record=true
每次升级或回滚操作之后REVISION的数字都会增加1。通过kubectl history,我们可以看到每次升级记录,--record记录了每次操作的详细过程。通过kubectl undo执行回滚操作,选项--to-revision指定了回滚的修订版本号。
三.总结
3.1. 再次被Kubernetes的魅力所折服,平常工作中的灰度升级/发布/更新(按批次停止老版本实例,启用新版本实例,新老版本共存,逐步扩大新版本范围,最终把所有用户迁移到新版本上),蓝绿升级/发布/更新(不停止老版本,另外部署一套新版本,新版本通过测试发布后删除老版本),滚动升级/发布/更新(一个或多个服务停止,执行更新,逐步将新版本投入使用,周而复始,最终完成整个集群中所有实例的版本更新)这些概念理解更加透彻。
3.2. 目前有些资料已经不适用于新版本了,不要被网上的资料所误导,一定要立足与官方文档和系统帮助。例如:
Kubernetes 的Deploy相关的命令只有:
Deploy Commands: rollout Manage the rollout of a resource scale 为 Deployment, ReplicaSet, Replication Controller 或者 Job 设置一个新的副本数量 autoscale 自动调整一个 Deployment, ReplicaSet, 或者 ReplicationController 的副本数量
Rollout可用资源类型:
Valid resource types include: * deployments * daemonsets * statefulsets
3.3 Rollout updat也可以通过YAML来实现,后续文章会介绍。
四.相关资料
4.1 本文相关的yaml文件
4.2 官方资料