Kubernetes--应用滚动升级

1.预备知识

滚动升级(rolling update)是每一次只更新一小部分的副本，成功后再继续更新更多的副本，最终把所有副本更新。

好处：不用停机，实现平滑的升级。

---

2.Deployment的滚动升级

2.1 Deployment与Replica Set与Pod的关系

如下图所示(网上找的)，

一个Deployment管理多个Replica Set，一个Replica Set 管理多个Pod
正因为这样，Deployment能够实现回滚，每一次部署新的Deployment时，都会生成一个新的Replica Set，并且会保留下来，所以以后想回退版本的话，就能直接利用旧的Replica Set回滚回到先前版本。

2.2 开始动手

1. 部署一个需要3个Pod的Deployment，镜像为httpd:2.2.31,yam文件如下

apiVersion: apps/v1beta1
kind: Deployment
metadata:
  name: httpd
spec: 
  replicas: 3
  minReadySeconds: 5
  strategy:
    type: RollingUpdate
    rollingUpdate:
      maxSurge: 1
      maxUnavailable: 1
  template:
    metadate:
      labels:
        run: httpd
    spec:
      containers:
      - name: httpd
        image: httpd:2.2.31
        ports:
        - containerPort: 80

2.执行部署并查看

[root@ceph1 sector7]# kubectl get deployment -o wide
NAME      DESIRED   CURRENT   UP-TO-DATE   AVAILABLE   AGE       CONTAINERS   IMAGES         SELECTOR
httpd     3         3         3            1           13s       httpd        httpd:2.2.31   run=httpd
[root@ceph1 sector7]# kubectl get deployment -o wide
NAME      DESIRED   CURRENT   UP-TO-DATE   AVAILABLE   AGE       CONTAINERS   IMAGES         SELECTOR
httpd     3         3         3            3           1m        httpd        httpd:2.2.31   run=httpd
[root@ceph1 sector7]# kubectl get pods -o wide
NAME                    READY     STATUS    RESTARTS   AGE       IP            NODE
httpd-9658687dd-7q7fw   1/1       Running   0          1m        10.244.0.36   ceph1
httpd-9658687dd-f9r9t   1/1       Running   0          1m        10.244.2.11   localhost.localdomain
httpd-9658687dd-ql46w   1/1       Running   0          1m        10.244.1.36   ceph3
[root@ceph1 sector7]# kubectl get rs
NAME              DESIRED   CURRENT   READY     AGE
httpd-9658687dd   3         3         3         2m

通过Deployment，Replica Set，Pod的名字也可以看出彼此的等级关系：(最后的是随机字符串)
Deployment：httpd
Replica Set：httpd-9658687dd
Pod：httpd-9658687dd-7q7fw

注意当前httpd的版本为：2.2.31，下面开始升级
3. 升级
将httpd.yaml文件中的image修改为：httpd:2.2.32
再执行一次部署kubectl apply -f httpd.yaml
再次查看：

[root@ceph1 sector7]# kubectl get deployment -o wide
NAME      DESIRED   CURRENT   UP-TO-DATE   AVAILABLE   AGE       CONTAINERS   IMAGES         SELECTOR
httpd     3         4         2            2           13m       httpd        httpd:2.2.32   run=httpd
[root@ceph1 sector7]# kubectl get rs
NAME               DESIRED   CURRENT   READY     AGE
httpd-76c8bd9f65   2         2         0         9s
httpd-9658687dd    2         2         2         13m
[root@ceph1 sector7]# kubectl get pods -o wide
NAME                     READY     STATUS              RESTARTS   AGE       IP            NODE
httpd-76c8bd9f65-kvzdh   0/1       ContainerCreating   0          2s        <none>        ceph1
httpd-76c8bd9f65-vfblg   1/1       Running             0          14s       10.244.1.37   ceph3
httpd-76c8bd9f65-wfqkg   1/1       Running             0          15s       10.244.2.12   localhost.localdomain
httpd-9658687dd-f9r9t    0/1       Terminating         0          13m       10.244.2.11   localhost.localdomain
[root@ceph1 sector7]# kubectl get pods -o wide
NAME                     READY     STATUS    RESTARTS   AGE       IP            NODE
httpd-76c8bd9f65-kvzdh   1/1       Running   0          16s       10.244.0.37   ceph1
httpd-76c8bd9f65-vfblg   1/1       Running   0          28s       10.244.1.37   ceph3
httpd-76c8bd9f65-wfqkg   1/1       Running   0          29s       10.244.2.12   localhost.localdomain
[root@ceph1 sector7]# kubectl get deployment -o wide
NAME      DESIRED   CURRENT   UP-TO-DATE   AVAILABLE   AGE       CONTAINERS   IMAGES         SELECTOR
httpd     3         3         3            3           13m       httpd        httpd:2.2.32   run=httpd
[root@ceph1 sector7]# kubectl get rs
NAME               DESIRED   CURRENT   READY     AGE
httpd-76c8bd9f65   3         3         3         4m
httpd-9658687dd    0         0         0         17m

更新完成后，httpd的镜像变为了httpd:2.2.32了。

4. 滚动升级过程
   通过查看log，kubectl describe deployment httpd

Events:
  Type    Reason             Age   From                   Message
  ----    ------             ----  ----                   -------
  Normal  ScalingReplicaSet  21m   deployment-controller  Scaled up replica set httpd-9658687dd to 3
  Normal  ScalingReplicaSet  8m    deployment-controller  Scaled up replica set httpd-76c8bd9f65 to 1
  Normal  ScalingReplicaSet  8m    deployment-controller  Scaled down replica set httpd-9658687dd to 2
  Normal  ScalingReplicaSet  8m    deployment-controller  Scaled up replica set httpd-76c8bd9f65 to 2
  Normal  ScalingReplicaSet  8m    deployment-controller  Scaled down replica set httpd-9658687dd to 1
  Normal  ScalingReplicaSet  8m    deployment-controller  Scaled up replica set httpd-76c8bd9f65 to 3
  Normal  ScalingReplicaSet  8m    deployment-controller  Scaled down replica set httpd-9658687dd to 0

从Message中可以看出，两个Replica Set是逐步更新Pod的，
httpd-9658687dd是最开始的，有3个Pod，httpd-76c8bd9f65是新生成，有0个Pod，依次
httpd-76c8bd9f65 up 为1，httpd-9658687dd down 为2
httpd-76c8bd9f65 up 为2，httpd-9658687dd down 为1
httpd-76c8bd9f65 up 为3，httpd-9658687dd down 为0
当然，滚动升级每次更新的Pod数量是可以指定的，通过两个参数maxSurge和maxUnavailable控制。

• maxSurge
  升级过程中，最多可以比原先设定所多出的Pod数量，可以是具体的数值，也可以是百分比(百分比的话向上取整，默认为25%，比如10个，那就是10*25%=2.5，向上取整为3)
• maxUnavailable
  最多有多少个Pod处于不可用的状态，可以是具体的数值，也可以是百分比(百分比是向下取整，默认为25%)

所以，maxSurge的值越大，初始创建的Pod就越多，maxUnavailable的值越大，初始销毁的旧Pod就越多

5. 升级相关的命令操作

• 查询升级状态
  kubectl rollout status deployment <deployment_name>
• 暂停滚动升级
  kubectl rollout pause deployment <deployment_name>
• 恢复滚动升级
  kubectl rollout resume deployment <deployment_name>

---

3.回退版本

kubectl apply在每次更新应用的时候，都会记录下当前的配置，保存为一个版本revision，默认情况kubernetes只会保留最近的几个revision，但可以在Deployment的配置文件中指定保存的revision的数量，通过revisionHistoryLimit属性设置。

3.1 验证

将上面的httpd.yaml文件复制三份，分别命名为httpd1.yaml，httpd2.yaml，httpd3.yaml,对应镜像修改为httpd:2.4.16,httpd:2.4.17,httpd:2.4.18.

1. 执行部署

[root@ceph1 sector7]# kubectl apply -f httpd1.yaml --record
deployment.apps "httpd" created
[root@ceph1 sector7]# kubectl get deployment -o wide   
NAME      DESIRED   CURRENT   UP-TO-DATE   AVAILABLE   AGE       CONTAINERS   IMAGES         SELECTOR
httpd     3         3         3            3           29s       httpd        httpd:2.4.16   run=httpd
[root@ceph1 sector7]# kubectl apply -f httpd2.yaml  --record
deployment.apps "httpd" configured
[root@ceph1 sector7]# kubectl get deployment -o wide
NAME      DESIRED   CURRENT   UP-TO-DATE   AVAILABLE   AGE       CONTAINERS   IMAGES         SELECTOR
httpd     3         3         3            3           1m        httpd        httpd:2.4.17   run=httpd
[root@ceph1 sector7]# kubectl apply -f httpd3.yaml 
deployment.apps "httpd" configured
[root@ceph1 sector7]# kubectl get deployment -o wide
NAME      DESIRED   CURRENT   UP-TO-DATE   AVAILABLE   AGE       CONTAINERS   IMAGES         SELECTOR
httpd     3         3         3            3           5m        httpd        httpd:2.4.18   run=httpd

由上面信息可知，这一次版本从2.4.16升级到2.4.17再升级到2.4.18，总共有三次操作，而且这一次执行kubectl apply时候加上了**–record**.

–record的作用是将当前命令记录到revision记录中，如果没加上（比如上面更新为httpd3,.yaml时）在revision记录中会显示为，这样子就无法看出更新的信息。

查看历史版本

[root@ceph1 sector7]# kubectl rollout history deployment httpd
deployments "httpd"
REVISION  CHANGE-CAUSE
1         kubectl apply --filename=httpd1.yaml --record=true
2         kubectl apply --filename=httpd2.yaml --record=true
3         <none>

这里的CHANGE_CAUSE就是加上了–record的结果。REVISION就是版本，如果想回退到revision=1,可以执行命令:
kubectl rollout undo deployment httpd --to-revision=1
如果是想回退到上一个版本，则可以不用指定–to-revision
2. 总结
所以在执行apply时候需要加上–record对于以后回退版本会比较清晰。