用戶希望應用程序始終可用，開發人員應該每天多次部署新版本的應用程序。在Kubernetes中，這是通過滾動更新完成的。滾動更新允許通過使用新的實例逐步更新Pods實例來實現部署的更新，從而實現零停機。新的Pod將在具有可用資源的節點上進行調度。
在上一個模塊中，我們將應用程序擴展為運行多個實例。這是執行更新而不影響應用程序可用性的要求。默認情況下，更新期間可用的最大Pod數和可以創建的新Pod的最大數量是一個。兩個選項都可以配置為數字或百分比（Pod）。在Kubernetes中，更新是版本化的，任何部署更新都可以恢復為以前的（穩定）版本。
版本記錄：每次對Deployment的操作，都能保存下來，給予后續可能的回滾使用。
1. 查看正在運行的Pods
[root@master ~]# kubectl get pods -o wide NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES kubernetes-bootcamp-7f56b875dc-6zbjt 1/1 Running 0 5h31m 10.244.0.15 master <none> <none> kubernetes-bootcamp-7f56b875dc-cp7cq 1/1 Running 0 19m 10.244.0.16 master <none> <none> kubernetes-bootcamp-7f56b875dc-n4wkn 1/1 Running 0 19m 10.244.0.17 master <none> <none> kubernetes-bootcamp-7f56b875dc-pmvck 1/1 Running 0 19m 10.244.0.18 master <none> <none>
2. 使用新的鏡像更新程序
[root@master ~]# kubectl set image deployments/kubernetes-bootcamp kubernetes-bootcamp=jocatalin/kubernetes-bootcamp:v2 deployment.extensions/kubernetes-bootcamp image updated
3. 確認更新完成
[root@master ~]# kubectl rollout status deployment/kubernetes-bootcamp deployment "kubernetes-bootcamp" successfully rolled out
4. 查看Pod的變化
可以看見Pod的IP都變成新的了
[root@master ~]# kubectl get pods -l run=kubernetes-bootcamp -o wide NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES kubernetes-bootcamp-5bf4d5689b-5f68t 1/1 Running 0 3m44s 10.244.0.20 master <none> <none> kubernetes-bootcamp-5bf4d5689b-c4zgm 1/1 Running 0 3m41s 10.244.0.22 master <none> <none> kubernetes-bootcamp-5bf4d5689b-c99hl 1/1 Running 0 3m45s 10.244.0.19 master <none> <none> kubernetes-bootcamp-5bf4d5689b-rpff2 1/1 Running 0 3m41s 10.244.0.21 master <none> <none>
5. 回滾應用
第一種情況：
我們這里假設更新錯了一個不存在的鏡像
kubectl set image deployments/kubernetes-bootcamp kubernetes-bootcamp=gcr.io/google-samples/kubernetes-bootcamp:v10 deployment.extensions/kubernetes-bootcamp image updated
查看Pods
[root@master ~]# kubectl get pods -l run=kubernetes-bootcamp -o wide NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES kubernetes-bootcamp-597cfc5b76-jtmrr 0/1 ImagePullBackOff 0 51s 10.244.0.23 master <none> <none> kubernetes-bootcamp-597cfc5b76-pslgq 0/1 ImagePullBackOff 0 51s 10.244.0.24 master <none> <none> kubernetes-bootcamp-5bf4d5689b-5f68t 1/1 Running 0 23m 10.244.0.20 master <none> <none> kubernetes-bootcamp-5bf4d5689b-c99hl 1/1 Running 0 23m 10.244.0.19 master <none> <none> kubernetes-bootcamp-5bf4d5689b-rpff2 1/1 Running 0 23m 10.244.0.21 master <none> <none>
查看部署
[root@master ~]# kubectl get deployments -l run=kubernetes-bootcamp NAME READY UP-TO-DATE AVAILABLE AGE kubernetes-bootcamp 3/4 2 3 5h58m
回滾
[root@master ~]# kubectl rollout undo deployments/kubernetes-bootcamp deployment.extensions/kubernetes-bootcamp rolled back
查看pods
[root@master ~]# kubectl get pods -l run=kubernetes-bootcamp -o wide NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES kubernetes-bootcamp-5bf4d5689b-5f68t 1/1 Running 0 29m 10.244.0.20 master <none> <none> kubernetes-bootcamp-5bf4d5689b-c99hl 1/1 Running 0 29m 10.244.0.19 master <none> <none> kubernetes-bootcamp-5bf4d5689b-jrrh9 1/1 Running 0 70s 10.244.0.25 master <none> <none> kubernetes-bootcamp-5bf4d5689b-rpff2 1/1 Running 0 28m 10.244.0.21 master <none> <none>
可以看出兩個錯誤的Pods已經不見了，原來的10.244.0.22的pods被新的Pods（10.244.0.25）取代。
第二種情況，我們全部更新完成，是否還可以回滾
[root@master ~]# curl http://192.168.0.10:31238 Hello Kubernetes bootcamp! | Running on: kubernetes-bootcamp-5bf4d5689b-h4wsw | v=2 [root@master ~]# kubectl set image deployments/kubernetes-bootcamp kubernetes-bootcamp=jocatalin/kubernetes-bootcamp:v1 deployment.extensions/kubernetes-bootcamp image updated [root@master ~]# curl http://192.168.0.10:31238 Hello Kubernetes bootcamp! | Running on: kubernetes-bootcamp-7f56b875dc-kvxn5 | v=1 [root@master ~]# kubectl rollout undo deployments/kubernetes-bootcamp deployment.extensions/kubernetes-bootcamp rolled back [root@master ~]# curl http://192.168.0.10:31238 Hello Kubernetes bootcamp! | Running on: kubernetes-bootcamp-5bf4d5689b-rfhqx | v=2
證明也是可以回滾前一個版本的。