詳細聊聊k8s deployment的滾動更新（一）

wilson排球發表於2019-01-30
K8S

一、知識準備

● 本文詳細探索deployment在滾動更新時候的行為

二、環境準備

元件 版本
OS Ubuntu 18.04.1 LTS
docker 18.06.0-ce

三、準備映象

首先準備2個不同版本的映象，用於測試（已經在阿里雲上建立好2個不同版本的nginx映象）

docker pull registry.cn-beijing.aliyuncs.com/mrvolleyball/nginx:v1
docker pull registry.cn-beijing.aliyuncs.com/mrvolleyball/nginx:v2
root@k8s-master:~# docker run -d --rm -p 10080:80 nginx:v1
e88097841c5feef92e4285a2448b943934ade5d86412946bc8d86e262f80a050
root@k8s-master:~# curl http://127.0.0.1:10080
----------
version: v1
hostname: f5189a5d3ad3

四、deployment、replicaset、pod之間的關係

                           +------------+
                           | deployment |
                           +-----+------+
                                 |
                                 |
                                 |
                                 |
       +--------------------------------------------------+
       |                         |                        |
       |                         |                        |
       |                         |                        |
       |                         |                        |
       |                         |                        |
       |                         |                        |
+------v------+           +------v------+          +------v------+
|replicaset:v1|           |replicaset:v2|          |replicaset:v3|
+-------------+           +------+------+          +-------------+
                                 |
                                 |
                        +--------+---------+
                        |                  |
                        |                  |
                    +---v---+          +---v---+
                    |pod:v2 |          |pod:v2 |
                    +-------+          +-------+

● deployment排程replicaset，pod由replicaset排程
● deployment管理多個replicaset版本，可用於回滾
● replicaset控制pod的行為，包括新增pod、刪除pod

我們首先準備一個yaml檔案用於測試：

root@k8s-master:~# more roll_update.yaml
apiVersion: extensions/v1beta1
kind: Deployment
metadata:
  name: image-deployment
spec:
  replicas: 1
  template:
    metadata:
      labels:
        app: image-update
    spec:
      containers:
      - name: nginx
        image: registry.cn-beijing.aliyuncs.com/mrvolleyball/nginx:v1
        imagePullPolicy: Always

簡單驗證一下：

root@k8s-master:~# kubectl apply -f roll_update.yaml
deployment.extensions "update-deployment" created
root@k8s-master:~# kubectl get deploy
NAME                DESIRED   CURRENT   UP-TO-DATE   AVAILABLE   AGE
update-deployment   3         3         3            3           54s
root@k8s-master:~# kubectl get rs
NAME                           DESIRED   CURRENT   READY     AGE
update-deployment-7db77f7cc6   3         3         3         56s
root@k8s-master:~# kubectl get pod
NAME                                 READY     STATUS    RESTARTS   AGE
update-deployment-7db77f7cc6-7j49g   1/1       Running   0          1m
update-deployment-7db77f7cc6-b75wn   1/1       Running   0          1m
update-deployment-7db77f7cc6-cfnt5   1/1       Running   0          1m

deployment、replicaset、pod都已經正常啟動，下面分析一下他們的行為：

deployment

root@k8s-master:~# kubectl describe deploy update-deployment
Name:                   update-deployment
Namespace:              default
...
Replicas:               3 desired | 3 updated | 3 total | 3 available | 0 unavailable
StrategyType:           RollingUpdate
MinReadySeconds:        0
RollingUpdateStrategy:  1 max unavailable, 1 max surge
...
NewReplicaSet:   update-deployment-7db77f7cc6 (3/3 replicas created)
Events:
  Type    Reason             Age   From                   Message
  ----    ------             ----  ----                   -------
  Normal  ScalingReplicaSet  1m    deployment-controller  Scaled up replica set update-deployment-7db77f7cc6 to 3

● deployment建立了一個replicaset，叫做update-deployment-7db77f7cc6（7db77f7cc6是replicaset的template hash值）
● 根據配置檔案的要求，replicaset的副本數為3

replicaset

root@k8s-master:~# kubectl describe rs update-deployment-7db77f7cc6
Name:           update-deployment-7db77f7cc6
Namespace:      default
...
Controlled By:  Deployment/update-deployment
Replicas:       3 current / 3 desired
Pods Status:    3 Running / 0 Waiting / 0 Succeeded / 0 Failed
...
Events:
  Type    Reason            Age   From                   Message
  ----    ------            ----  ----                   -------
  Normal  SuccessfulCreate  3m    replicaset-controller  Created pod: update-deployment-7db77f7cc6-7j49g
  Normal  SuccessfulCreate  3m    replicaset-controller  Created pod: update-deployment-7db77f7cc6-b75wn
  Normal  SuccessfulCreate  3m    replicaset-controller  Created pod: update-deployment-7db77f7cc6-cfnt5

● replicaset建立了3個pod

pod

root@k8s-master:~# kubectl describe pod update-deployment-7db77f7cc6-7j49g
Name:           update-deployment-7db77f7cc6-7j49g
Namespace:      default
...
Status:         Running
IP:             10.10.169.140
Controlled By:  ReplicaSet/update-deployment-7db77f7cc6
...
Events:
  Type    Reason                 Age   From                Message
  ----    ------                 ----  ----                -------
  Normal  Scheduled              9m    default-scheduler   Successfully assigned update-deployment-7db77f7cc6-7j49g to k8s-node2
  Normal  SuccessfulMountVolume  9m    kubelet, k8s-node2  MountVolume.SetUp succeeded for volume "default-token-v9nkm"
  Normal  Pulling                9m    kubelet, k8s-node2  pulling image "registry.cn-beijing.aliyuncs.com/mrvolleyball/nginx:v1"
  Normal  Pulled                 9m    kubelet, k8s-node2  Successfully pulled image "registry.cn-beijing.aliyuncs.com/mrvolleyball/nginx:v1"
  Normal  Created                9m    kubelet, k8s-node2  Created container
  Normal  Started                9m    kubelet, k8s-node2  Started container

● pod被replicaset建立之後，開始分配到worker節點、拉取映象、啟動容器等一系列操作
● 所以pod的命名方式是：update-deployment-7db77f7cc6-7j49g（deployment名字-replicaset模板hash名字-pod模板hash名字）

不禁有同學要問，為什麼搞這麼複雜，啟動一個pod需要動用這麼多元件呢？下面用一個場景說明為啥需要這麼多元件：

映象版本更新

● 當映象版本有更新時（三種方法都可以實現，參考前一篇文章：更新k8s映象版本的三種方式），既要保證服務可用，又要保證線上更新，流程應該是：
  1、先增加一個pod，映象版本為新版本
  2、pod可用之後，刪除一個老版本pod
  3、迴圈第1、2步，直到老版本pod全部刪除，新版本的pod全部可用
● 上述的這個過程就是replicaset的作用，它根據需求，自動的增加新版本pod，然後刪除老版本pod，直到老版本pod全部刪除，新版本的pod全部可用
● 如果此時版本需要回退，那replicaset需要把剛才的步驟逆向更新一遍，實現版本回退
● deployment的作用就是管理replicaset。deployment會儲存各個版本的replicaset，一旦需要進行版本回滾，deployment會立即回滾replicaset的版本，從而控制pod狀態

下面測試一下：

使用patch命令更新映象版本，並且使用pause命令來觀察：

root@k8s-master:~# kubectl patch deployment update-deployment 
--patch `{"spec": {"template": {"spec": {"containers": [{"name": "nginx","image":"registry.cn-beijing.aliyuncs.com/mrvolleyball/nginx:v2"}]}}}}` 
&& kubectl rollout pause deployment update-deployment
deployment.extensions "update-deployment" patched
deployment.apps "update-deployment" paused

此時pod狀態：

root@k8s-master:~# kubectl get pod -owide
NAME                                 READY     STATUS        RESTARTS   AGE       IP              NODE
update-deployment-7db77f7cc6-7j49g   1/1       Running       0          1h        10.10.169.140   k8s-node2
update-deployment-7db77f7cc6-b75wn   1/1       Running       0          1h        10.10.235.211   k8s-master
update-deployment-7db77f7cc6-cfnt5   1/1       Terminating   0          1h        10.10.36.126    k8s-node1
update-deployment-7fb7b4b557-6987x   1/1       Running       0          7s        10.10.36.127    k8s-node1
update-deployment-7fb7b4b557-dxdqb   1/1       Running       0          10s       10.10.169.139   k8s-node2

新增了2個pod，而刪除了1個老版本的pod

此時replicaset狀態：

root@k8s-master:~# kubectl get rs -owide
NAME                           DESIRED   CURRENT   READY     AGE       CONTAINERS   IMAGES                                                   SELECTOR
update-deployment-7db77f7cc6   2         2         2         1h        nginx        registry.cn-beijing.aliyuncs.com/mrvolleyball/nginx:v1   app=roll-update,pod-template-hash=3863393772
update-deployment-7fb7b4b557   2         2         2         4m        nginx        registry.cn-beijing.aliyuncs.com/mrvolleyball/nginx:v2   app=roll-update,pod-template-hash=3963606113

有一個新版本的replicaset建立了出來，並且需求的pod數量為2，而原來的replicaset需求的pod數量從3降為2

檢視replicaset版本：

root@k8s-master:~# kubectl rollout history deploy update-deployment
deployments "update-deployment"
REVISION  CHANGE-CAUSE
1         <none>
2         update version to v2

新增了一個版本2

由於使用pause命令，更新過程到此會卡主，我們讓更新的過程繼續下去：

root@k8s-master:~# kubectl rollout resume deployment update-deployment
deployment.apps "update-deployment" resumed

檢視狀態：

root@k8s-master:~# kubectl get pod
NAME                                 READY     STATUS    RESTARTS   AGE
update-deployment-7fb7b4b557-6987x   1/1       Running   0          15m
update-deployment-7fb7b4b557-dxdqb   1/1       Running   0          15m
update-deployment-7fb7b4b557-wg5c8   1/1       Running   0          1m
root@k8s-master:~# kubectl get rs -owide
NAME                           DESIRED   CURRENT   READY     AGE       CONTAINERS   IMAGES                                                   SELECTOR
update-deployment-7db77f7cc6   0         0         0         1h        nginx        registry.cn-beijing.aliyuncs.com/mrvolleyball/nginx:v1   app=roll-update,pod-template-hash=3863393772
update-deployment-7fb7b4b557   3         3         3         14m       nginx        registry.cn-beijing.aliyuncs.com/mrvolleyball/nginx:v2   app=roll-update,pod-template-hash=3963606113

v1版本的replicaset已經沒有pod，但是歷史記錄還是保留的，可以通過deployment排程快速回滾

五、小結

● 本文介紹了deployment滾動更新時，deployment、replicaset、pod的細節以及建立過程
● 介紹了deployment版本管理的方式
● 下一小節將會介紹在滾動更新過程中最大可用、liveness以及readiness等

至此，本文結束
在下才疏學淺，有撒湯漏水的，請各位不吝賜教…

相關文章