mirror of https://github.com/easzlab/kubeasz.git
215 lines
10 KiB
Markdown
215 lines
10 KiB
Markdown
## 1、前言
|
||
在当下微服务架构盛行的时代，用户希望应用程序时时刻刻都是可用，为了满足不断变化的新业务，需要不断升级更新应用程序，有时可能需要频繁的发布版本。实现"零停机"、“零感知”的持续集成(Continuous Integration)和持续交付/部署(Continuous Delivery)应用程序，一直都是软件升级换代不得不面对的一个难题和痛点，也是一种追求的理想方式，也是DevOps诞生的目的。
|
||
## 2、滚动发布
|
||
把一次完整的发布过程，合理地分成多个批次，每次发布一个批次，**成功后**，再发布下一个批次，最终完成所有批次的发布。在整个滚动过程期间，保证始终有可用的副本在运行，从而平滑的发布新版本，实现**零停机(without an outage)**、用户**零感知**，是一种非常主流的发布方式。由于其自动化程度比较高，通常需要复杂的发布工具支撑，而k8s可以完美的胜任这个任务。
|
||
## 3、k8s滚动更新机制
|
||
**k8s创建副本应用程序的最佳方法就是部署(Deployment)，部署自动创建副本集(ReplicaSet)，副本集可以精确地控制每次替换的Pod数量，从而可以很好的实现滚动更新**。具体来说，k8s每次使用一个新的副本控制器(replication controller)来替换已存在的副本控制器，从而始终使用一个新的Pod模板来替换旧的pod模板。
|
||
>大致步骤如下：
|
||
>1. 创建一个新的replication controller。
|
||
>2. 增加或减少pod副本数量，直到满足当前批次期望的数量。
|
||
>3. 删除旧的replication controller。
|
||
|
||
## 4、演示
|
||
>使用kubectl更新一个已部署的应用程序，并模拟回滚。为了方便分析，将应用程序的pod副本数量设置为10。
|
||
``` bash
|
||
$ kubectl run busy --image=busybox:1.28.4 sleep 36000000 --replicas=10
|
||
```
|
||
### 4.1. 发布微服务
|
||
- 当前服务状态查看
|
||
``` bash
|
||
# 查看部署列表
|
||
root@kube-aio:~# kubectl get deploy busy
|
||
NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE
|
||
busy 10 10 10 10 5m
|
||
|
||
# 查看正在运行的pod
|
||
root@kube-aio:~# kubectl get pod | grep busy
|
||
busy-794c95f5d7-56b6w 1/1 Running 0 5m
|
||
busy-794c95f5d7-8ddjr 1/1 Running 0 5m
|
||
busy-794c95f5d7-8zm8r 1/1 Running 0 5m
|
||
busy-794c95f5d7-9hjhp 1/1 Running 0 5m
|
||
busy-794c95f5d7-df2r2 1/1 Running 0 5m
|
||
busy-794c95f5d7-fsn94 1/1 Running 0 5m
|
||
busy-794c95f5d7-k4w8r 1/1 Running 0 5m
|
||
busy-794c95f5d7-lsmgb 1/1 Running 0 5m
|
||
busy-794c95f5d7-rg8kw 1/1 Running 0 5m
|
||
busy-794c95f5d7-xpxxt 1/1 Running 0 5m
|
||
|
||
# 通过pod描述，查看应用程序的当前映像版本
|
||
root@kube-aio:~# kubectl describe pod busy-794c95f5d7-56b6w |grep Image
|
||
Image: busybox:1.28.4
|
||
Image ID: docker-pullable://busybox@sha256:141c253bc4c3fd0a201d32dc1f493bcf3fff003b6df416dea4f41046e0f37d47
|
||
```
|
||
- 升级镜像版本到1.29
|
||
- 为了更清晰看到更新过程，可另开一个窗口使用`$ watch kubectl get deployment busy`实时查看变化
|
||
``` bash
|
||
$ kubectl set image deployments/busy busy=busybox:1.29
|
||
```
|
||
### 4.2. 验证发布
|
||
``` bash
|
||
# 检查rollout状态
|
||
root@kube-aio:~# kubectl rollout status deployments/busy
|
||
deployment "busy" successfully rolled out
|
||
|
||
# 检查pod详情
|
||
root@kube-aio:~# kubectl describe pod busy-665cdb7b-44jnt |grep Image
|
||
Image: busybox:1.29
|
||
Image ID: docker-pullable://busybox@sha256:cb63aa0641a885f54de20f61d152187419e8f6b159ed11a251a09d115fdff9bd
|
||
```
|
||
从上面可以看到，镜像已经升级到1.29版本
|
||
### 4.3. 回滚发布
|
||
``` bash
|
||
# 回滚发布
|
||
root@kube-aio:~# kubectl rollout undo deployments/busy
|
||
deployment.apps "busy"
|
||
|
||
# 回滚完成
|
||
root@kube-aio:~# kubectl rollout status deployments/busy
|
||
deployment "busy" successfully rolled out
|
||
|
||
# 镜像又回退到1.28.4 版本
|
||
root@kube-aio:~# kubectl describe pod busy-794c95f5d7-4x9bn |grep Image
|
||
Image: busybox:1.28.4
|
||
Image ID: docker-pullable://busybox@sha256:141c253bc4c3fd0a201d32dc1f493bcf3fff003b6df416dea4f41046e0f37d47
|
||
```
|
||
|
||
到目前为止，整个滚动发布工作就圆满完成了！！！
|
||
**那么如果我们想回滚到指定版本呢？答案是k8s完美支持，并且还可以通过资源文件进行配置保留的历史版次量**。由于篇幅有限，感兴趣的朋友，可以自己下去实战，回滚命令如下：
|
||
```javascript
|
||
kubectl rollout undo deployment/busy --to-revision=<版次>
|
||
```
|
||
## 5、原理
|
||
k8s精确地控制着整个发布过程，分批次有序地进行着滚动更新，直到把所有旧的副本全部更新到新版本。实际上，k8s是通过两个参数来精确地控制着每次滚动的pod数量：
|
||
|
||
>* **`maxSurge` 滚动更新过程中运行操作期望副本数的最大pod数，可以为绝对数值(eg：5)，但不能为0；也可以为百分数(eg：10%)。**
|
||
>* **`maxUnavailable` 滚动更新过程中不可用的最大pod数，可以为绝对数值(eg：5)，但不能为0；也可以为百分数(eg：10%)。**
|
||
|
||
如果未指定这两个可选参数，则k8s会使用默认配置：
|
||
``` bash
|
||
root@kube-aio:~# kubectl get deploy busy -o yaml
|
||
apiVersion: apps/v1
|
||
kind: Deployment
|
||
metadata:
|
||
annotations:
|
||
deployment.kubernetes.io/revision: "3"
|
||
creationTimestamp: 2018-08-19T02:42:56Z
|
||
generation: 3
|
||
labels:
|
||
run: busy
|
||
name: busy
|
||
namespace: default
|
||
resourceVersion: "199461"
|
||
uid: 93fde307-a359-11e8-a93b-525400c61543
|
||
spec:
|
||
progressDeadlineSeconds: 600
|
||
replicas: 10
|
||
revisionHistoryLimit: 10
|
||
selector:
|
||
matchLabels:
|
||
run: busy
|
||
strategy:
|
||
rollingUpdate:
|
||
maxSurge: 1 # 滚动更新中最多超过预期值的 pod数
|
||
maxUnavailable: 1 # 滚动更新中最多不可用的 pod数
|
||
type: RollingUpdate
|
||
...
|
||
```
|
||
### 5.1. 浅析部署概况
|
||
``` bash
|
||
# 初始状态
|
||
root@kube-aio:~# kubectl get deploy busy
|
||
NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE
|
||
busy 10 10 10 10 1h
|
||
|
||
# 再做一遍回退
|
||
root@kube-aio:~# kubectl rollout undo deploy busy
|
||
deployment.apps "busy"
|
||
|
||
# 更新过程1
|
||
root@kube-aio:~# kubectl get deploy busy
|
||
NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE
|
||
busy 10 11 2 9 1h
|
||
|
||
# 更新过程2
|
||
root@kube-aio:~# kubectl get deploy busy
|
||
NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE
|
||
busy 10 11 4 9 1h
|
||
|
||
# 更新过程3
|
||
root@kube-aio:~# kubectl get deploy busy
|
||
NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE
|
||
busy 10 11 6 9 1h
|
||
|
||
# 更新结束
|
||
root@kube-aio:~# kubectl get deploy busy
|
||
NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE
|
||
busy 10 10 10 10 1h
|
||
```
|
||
>* `DESIRED` 最终期望处于READY状态的副本数
|
||
>* `CURRENT` 当前的副本总数
|
||
>* `UP-TO-DATE` 当前完成更新的副本数
|
||
>* `AVAILABLE` 当前可用的副本数
|
||
|
||
当前的副本总数：10(DESIRED) + 1(maxSurge) = 11，所以CURRENT为11。
|
||
当前可用的副本数：10(DESIRED) - 1(maxUnavailable) = 9，所以AVAILABLE为9。
|
||
|
||
### 5.2. 浅析部署详情
|
||
|
||
``` bash
|
||
root@kube-aio:~# kubectl describe deploy busy
|
||
Name: busy
|
||
Namespace: default
|
||
CreationTimestamp: Sun, 19 Aug 2018 12:27:19 +0800
|
||
Labels: run=busy
|
||
Annotations: deployment.kubernetes.io/revision=2
|
||
Selector: run=busy
|
||
Replicas: 10 desired | 10 updated | 10 total | 10 available | 0 unavailable
|
||
StrategyType: RollingUpdate
|
||
MinReadySeconds: 0
|
||
RollingUpdateStrategy: 1 max unavailable, 1 max surge
|
||
Pod Template:
|
||
Labels: run=busy
|
||
Containers:
|
||
busy:
|
||
Image: busybox:1.29
|
||
Port: <none>
|
||
Host Port: <none>
|
||
Args:
|
||
sleep
|
||
3600000
|
||
Environment: <none>
|
||
Mounts: <none>
|
||
Volumes: <none>
|
||
Conditions:
|
||
Type Status Reason
|
||
---- ------ ------
|
||
Available True MinimumReplicasAvailable
|
||
Progressing True NewReplicaSetAvailable
|
||
OldReplicaSets: <none>
|
||
NewReplicaSet: busy-84cb46955d (10/10 replicas created)
|
||
Events:
|
||
Type Reason Age From Message
|
||
---- ------ ---- ---- -------
|
||
Normal ScalingReplicaSet 1m deployment-controller Scaled up replica set busy-9669c8599 to 10
|
||
Normal ScalingReplicaSet 46s deployment-controller Scaled up replica set busy-84cb46955d to 1
|
||
Normal ScalingReplicaSet 46s deployment-controller Scaled down replica set busy-9669c8599 to 9
|
||
Normal ScalingReplicaSet 46s deployment-controller Scaled up replica set busy-84cb46955d to 2
|
||
Normal ScalingReplicaSet 43s deployment-controller Scaled down replica set busy-9669c8599 to 8
|
||
Normal ScalingReplicaSet 43s deployment-controller Scaled up replica set busy-84cb46955d to 3
|
||
Normal ScalingReplicaSet 43s deployment-controller Scaled down replica set busy-9669c8599 to 7
|
||
Normal ScalingReplicaSet 43s deployment-controller Scaled up replica set busy-84cb46955d to 4
|
||
Normal ScalingReplicaSet 40s deployment-controller Scaled down replica set busy-9669c8599 to 6
|
||
Normal ScalingReplicaSet 28s (x12 over 40s) deployment-controller (combined from similar events): Scaled down replica set busy-9669c8599 to 0
|
||
```
|
||
整个滚动过程是通过控制两个副本集来完成的，新的副本集：busy-84cb46955d；旧的副本集：busy-9669c8599 。
|
||
理想状态下的滚动过程：
|
||
>1. 创建新副本集，并为其分配1个新版本的pod。
|
||
>2. 通知旧副本集，销毁1个旧版本的pod。
|
||
>3. 当旧副本销毁成功后，通知新副本集，再新增1个新版本的pod；当新副本创建成功后，通知旧副本再减少1个pod。
|
||
>只要销毁成功，新副本集就会创造新的pod，一直循环，直到旧的副本集pod数量为0。
|
||
### 5.4 总结
|
||
**`无论理想还是不理想，k8s最终都会使应用程序全部更新到期望状态，都会始终保持最大的副本总数和可用副本总数的不变性！！！`**
|
||
|
||
[阅读原文](http://www.cnblogs.com/justmine/p/8688828.html)
|
||
|