generated from bing/readnotes
94 lines
1.9 KiB
Markdown
94 lines
1.9 KiB
Markdown
# Kubernetes集群升级
|
||
|
||
只能从一个次版本升级到下一个次版本，或者在次版本相同时升级补丁版本。 也就是说，升级时不可以跳过次版本。 例如，你只能从 1.y 升级到 1.y+1，而不能从 1.y 升级到 1.y+2。
|
||
|
||
## 升级control plane
|
||
|
||
必须先升级第一个controlplane
|
||
|
||
升级`kubeadm`
|
||
|
||
```bash
|
||
export VERSION="1.20.1"
|
||
yum install -y kubeadm-${VERSION}-0 --disableexcludes=kubernetes
|
||
```
|
||
|
||
腾空controlplane
|
||
|
||
```bash
|
||
export NODE="k8smaster1"
|
||
kubectl drain ${NODE} --ignore-daemonsets
|
||
```
|
||
|
||
查看是否可以升级以及可升级的版本
|
||
|
||
```bash
|
||
kubeadm upgrade plan
|
||
```
|
||
|
||
升级到指定版本
|
||
|
||
```bash
|
||
kubeadm upgrade apply v${VERSION}
|
||
```
|
||
|
||
取消控制面节点的保护
|
||
|
||
```bash
|
||
kubectl uncordon ${NODE}
|
||
```
|
||
|
||
升级其他controlplane, 参考[升级node](##升级node)
|
||
|
||
## 升级node
|
||
|
||
升级之前，请确保control plane已经升级，建议一次只升级1个或部分节点，避免影响服务。
|
||
|
||
升级`kubeadm`，在node执行
|
||
|
||
```bash
|
||
export VERSION="1.20.1-0"
|
||
yum install -y kubeadm-${VERSION} --disableexcludes=kubernetes
|
||
```
|
||
|
||
将节点标记为不可调度并逐出工作负载，为维护做好准备。在control执行
|
||
|
||
```bash
|
||
export NODE="k8snode1"
|
||
kubectl drain ${NODE} --ignore-daemonsets
|
||
```
|
||
|
||
输出类似下文
|
||
|
||
```bash
|
||
node/k8snode1 already cordoned
|
||
WARNING: ignoring DaemonSet-managed Pods: kube-system/calico-node-jcpf2, kube-system/kube-proxy-rsskt
|
||
evicting pod kube-system/calico-kube-controllers-744cfdf676-g86dz
|
||
pod/calico-kube-controllers-744cfdf676-g86dz evicted
|
||
node/k8snode1 evicted
|
||
```
|
||
|
||
升级kubelet配置，在node执行
|
||
|
||
```bash
|
||
kubeadm upgrade node
|
||
```
|
||
|
||
升级kubelet服务，在node执行
|
||
|
||
```bash
|
||
yum install -y kubelet-${VERSION} kubectl-${VERSION} --disableexcludes=kubernetes
|
||
```
|
||
|
||
重启服务
|
||
|
||
```bash
|
||
systemctl daemon-reload
|
||
systemctl restart kubelet
|
||
```
|
||
|
||
将节点标记为可调度，让节点重新上线
|
||
|
||
```bash
|
||
kubectl uncordon ${NODE}
|
||
``` |