目录
| 服务器版本 | docker软件版本 | Kubernetes(k8s)集群版本 | CPU架构 |
|---|---|---|---|
| CentOS Linux release 7.4.1708 (Core) | Docker version 20.10.12 | v1.21.9 | x86_64 |
Kubernetes集群架构：k8scloude1作为master节点，k8scloude2，k8scloude3作为worker节点
| 服务器 | 操作系统版本 | CPU架构 | 进程 | 功能描述 |
|---|---|---|---|---|
| k8scloude1/192.168.110.130 | CentOS Linux release 7.4.1708 (Core) | x86_64 | docker，kube-apiserver，etcd，kube-scheduler，kube-controller-manager，kubelet，kube-proxy，coredns，calico | k8s master节点 |
| k8scloude2/192.168.110.129 | CentOS Linux release 7.4.1708 (Core) | x86_64 | docker，kubelet，kube-proxy，calico | k8s worker节点 |
| k8scloude3/192.168.110.128 | CentOS Linux release 7.4.1708 (Core) | x86_64 | docker，kubelet，kube-proxy，calico | k8s worker节点 |
二.前言
本文介绍cordon节点，drain驱逐节点，delete 节点，在对k8s集群节点执行维护（例如内核升级、硬件维护等）时候会用到。
cordon节点，drain驱逐节点，delete 节点的前提是已经有一套可以正常运行的Kubernetes集群，关于Kubernetes(k8s)集群的安装部署，可以查看博客《Centos7 安装部署Kubernetes(k8s)集群》https://www.cnblogs.com/renshengdezheli/p/16686769.html
三.cordon节点
3.1 cordon节点概览
cordon 节点会使其停止调度，会将node状态调为SchedulingDisabled，之后再创建新pod，新pod不会被调度到该节点，原有的pod不会受到影响，仍正常对外提供服务。
3.2 cordon节点
创建目录存放yaml文件
[root@k8scloude1 ~]# mkdir deploy
[root@k8scloude1 ~]# cd deploy/
使用--dry-run生成deploy配置文件
[root@k8scloude1 deploy]# kubectl create deploy nginx --image=nginx --dry-run=client -o yaml >nginx.yaml
[root@k8scloude1 deploy]# cat nginx.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
creationTimestamp: null
labels:
app: nginx
name: nginx
spec:
replicas: 1
selector:
matchLabels:
app: nginx
strategy: {}
template:
metadata:
creationTimestamp: null
labels:
app: nginx
spec:
containers:
- image: nginx
name: nginx
resources: {}
status: {}
修改deploy配置文件，replicas: 5表示副本数为 5，deploy将创建5个pod
[root@k8scloude1 deploy]# vim nginx.yaml
#修改配置文件：
# replicas: 5 副本数修改为5
#terminationGracePeriodSeconds: 0 宽限期修改为0
# imagePullPolicy: IfNotPresent 镜像下载策略为存在镜像就不下载
[root@k8scloude1 deploy]# cat nginx.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
creationTimestamp: null
labels:
app: nginx
name: nginx
spec:
replicas: 5
selector:
matchLabels:
app: nginx
strategy: {}
template:
metadata:
creationTimestamp: null
labels:
app: nginx
spec:
terminationGracePeriodSeconds: 0
containers:
- image: nginx
name: nginx
imagePullPolicy: IfNotPresent
resources: {}
status: {}
创建deploy和使用pod yaml文件创建pod
[root@k8scloude1 deploy]# cat pod.yaml
apiVersion: v1
kind: Pod
metadata:
creationTimestamp: null
labels:
run: pod1
name: pod1
spec:
terminationGracePeriodSeconds: 0
containers:
- image: nginx
imagePullPolicy: IfNotPresent
name: n1
resources: {}
dnsPolicy: ClusterFirst
restartPolicy: Always
status: {}
[root@k8scloude1 deploy]# kubectl apply -f pod.yaml
pod/pod1 created
[root@k8scloude1 deploy]# kubectl apply -f nginx.yaml
deployment.apps/nginx created
查看pod，可以看到deploy生成5个pod（nginx-6cf858f6cf-XXXXXXX），还有一个pod1。
[root@k8scloude1 deploy]# kubectl get pods -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
nginx-6cf858f6cf-fwhmh 1/1 Running 0 52s 10.244.251.217 k8scloude3 <none> <none>
nginx-6cf858f6cf-hr6bn 1/1 Running 0 52s 10.244.251.218 k8scloude3 <none> <none>
nginx-6cf858f6cf-j2ccs 1/1 Running 0 52s 10.244.112.161 k8scloude2 <none> <none>
nginx-6cf858f6cf-l7n4w 1/1 Running 0 52s 10.244.112.162 k8scloude2 <none> <none>
nginx-6cf858f6cf-t6qxq 1/1 Running 0 52s 10.244.112.163 k8scloude2 <none> <none>
pod1 1/1 Running 0 60s 10.244.251.216 k8scloude3 <none> <none>
假设某天要对k8scloude2进行维护测试，不希望k8scloude2节点上被分配新的pod，可以对某个节点执行cordon之后，此节点就不会再调度新的pod了
cordon k8scloude2节点，k8scloude2节点变为SchedulingDisabled状态
[root@k8scloude1 deploy]# kubectl cordon k8scloude2
node/k8scloude2 cordoned
[root@k8scloude1 deploy]# kubectl get nodes
NAME STATUS ROLES AGE VERSION
k8scloude1 Ready control-plane,master 8d v1.21.0
k8scloude2 Ready,SchedulingDisabled <none> 7d23h v1.21.0
k8scloude3 Ready <none> 7d23h v1.21.0
kubectl scale deploy命令使nginx deploy的副本数扩展为10个
[root@k8scloude1 deploy]# kubectl scale deploy nginx --replicas=10
deployment.apps/nginx scaled
查看pod，可以发现新生成的pod都被调度到到k8scloude3上，某个节点被cordon之后，新的pod将不被调度到该节点，原先的pod不变。
[root@k8scloude1 deploy]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
nginx-6cf858f6cf-7fdnr 1/1 Running 0 4s 10.244.251.221 k8scloude3 <none> <none>
nginx-6cf858f6cf-fwhmh 1/1 Running 0 9m9s 10.244.251.217 k8scloude3 <none> <none>
nginx-6cf858f6cf-g92ls 1/1 Running 0 4s 10.244.251.219 k8scloude3 <none> <none>
nginx-6cf858f6cf-hr6bn 1/1 Running 0 9m9s 10.244.251.218 k8scloude3 <none> <none>
nginx-6cf858f6cf-j2ccs 1/1 Running 0 9m9s 10.244.112.161 k8scloude2 <none> <none>
nginx-6cf858f6cf-l7n4w 1/1 Running 0 9m9s 10.244.112.162 k8scloude2 <none> <none>
nginx-6cf858f6cf-lsvsg 1/1 Running 0 4s 10.244.251.223 k8scloude3 <none> <none>
nginx-6cf858f6cf-mpwjl 1/1 Running 0 4s 10.244.251.222 k8scloude3 <none> <none>
nginx-6cf858f6cf-s8x6b 1/1 Running 0 4s 10.244.251.220 k8scloude3 <none> <none>
nginx-6cf858f6cf-t6qxq 1/1 Running 0 9m9s 10.244.112.163 k8scloude2 <none> <none>
pod1 1/1 Running 0 9m17s 10.244.251.216 k8scloude3 <none> <none>
来个极端的例子，先把deploy的副本数变为0，再变为10，此时所有的pod都运行在k8scloude3节点了。
[root@k8scloude1 deploy]# kubectl scale deploy nginx --replicas=0
deployment.apps/nginx scaled
[root@k8scloude1 deploy]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
pod1 1/1 Running 0 10m 10.244.251.216 k8scloude3 <none> <none>
[root@k8scloude1 deploy]# kubectl scale deploy nginx --replicas=10
deployment.apps/nginx scaled
[root@k8scloude1 deploy]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
nginx-6cf858f6cf-5cx9s 1/1 Running 0 8s 10.244.251.231 k8scloude3 <none> <none>
nginx-6cf858f6cf-6cblj 1/1 Running 0 8s 10.244.251.228 k8scloude3 <none> <none>
nginx-6cf858f6cf-827cz 1/1 Running 0 8s 10.244.251.233 k8scloude3 <none> <none>
nginx-6cf858f6cf-b989n 1/1 Running 0 8s 10.244.251.229 k8scloude3 <none> <none>
nginx-6cf858f6cf-kwxhn 1/1 Running 0 8s 10.244.251.224 k8scloude3 <none> <none>
nginx-6cf858f6cf-ljjxz 1/1 Running 0 8s 10.244.251.225 k8scloude3 <none> <none>
nginx-6cf858f6cf-ltrpr 1/1 Running 0 8s 10.244.251.227 k8scloude3 <none> <none>
nginx-6cf858f6cf-lwf7g 1/1 Running 0 8s 10.244.251.230 k8scloude3 <none> <none>
nginx-6cf858f6cf-xw84l 1/1 Running 0 8s 10.244.251.226 k8scloude3 <none> <none>
nginx-6cf858f6cf-zpwhq 1/1 Running 0 8s 10.244.251.232 k8scloude3 <none> <none>
pod1 1/1 Running 0 11m 10.244.251.216 k8scloude3 <none> <none>
3.3 uncordon节点
要让节点恢复调度pod，uncordon即可。
uncordon k8scloude2节点，k8scloude2节点状态变为Ready，恢复调度。
#需要uncordon
[root@k8scloude1 deploy]# kubectl uncordon k8scloude2
node/k8scloude2 uncordoned
[root@k8scloude1 deploy]# kubectl get nodes
NAME STATUS ROLES AGE VERSION
k8scloude1 Ready control-plane,master 8d v1.21.0
k8scloude2 Ready <none> 8d v1.21.0
k8scloude3 Ready <none> 8d v1.21.0
四.drain节点
4.1 drain节点概览
在对节点执行维护（例如内核升级、硬件维护等）之前， 可以使用 kubectl drain 从节点安全地逐出所有 Pods。 安全的驱逐过程允许 Pod 的容器 体面地终止， 并确保满足指定的 PodDisruptionBudgets，PodDisruptionBudget 是一个对象，用于定义可能对一组 Pod 造成的最大干扰。。
说明： 默认情况下， kubectl drain 将忽略节点上不能杀死的特定系统 Pod； 'drain' 驱逐或删除除镜像 pod 之外的所有 pod（不能通过 API 服务器删除）。如果有 daemon set-managed pods，drain 不会在没有 --ignore-daemonsets 的情况下继续进行，并且无论如何它都不会删除任何 daemon set-managed pods，因为这些 pods 将立即被 daemon set 控制器替换，它会忽略不可调度的标记。如果有任何 pod 既不是镜像 pod，也不是由复制控制器、副本集、守护程序集、有状态集或作业管理的，那么除非您使用 --force，否则 drain 不会删除任何 pod。如果一个或多个 pod 的管理资源丢失， --force 也将允许继续删除。
kubectl drain 的成功返回，表明所有的 Pods（除了上一段中描述的被排除的那些）， 已经被安全地逐出（考虑到期望的终止宽限期和你定义的 PodDisruptionBudget）。 然后就可以安全地关闭节点， 比如关闭物理机器的电源，如果它运行在云平台上，则删除它的虚拟机。
4.2 drain 节点
查看node状态和pod
[root@k8scloude1 deploy]# kubectl get nodes
NAME STATUS ROLES AGE VERSION
k8scloude1 Ready control-plane,master 8d v1.21.0
k8scloude2 Ready <none> 8d v1.21.0
k8scloude3 Ready <none> 8d v1.21.0
[root@k8scloude1 deploy]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
nginx-6cf858f6cf-58wnd 1/1 Running 0 65s 10.244.112.167 k8scloude2 <none> <none>
nginx-6cf858f6cf-5rrk4 1/1 Running 0 65s 10.244.112.164 k8scloude2 <none> <none>
nginx-6cf858f6cf-86wxr 1/1 Running 0 65s 10.244.251.237 k8scloude3 <none> <none>
nginx-6cf858f6cf-89wj9 1/1 Running 0 65s 10.244.112.168 k8scloude2 <none> <none>
nginx-6cf858f6cf-9njrj 1/1 Running 0 65s 10.244.251.236 k8scloude3 <none> <none>
nginx-6cf858f6cf-hchtb 1/1 Running 0 65s 10.244.251.234 k8scloude3 <none> <none>
nginx-6cf858f6cf-mb2ft 1/1 Running 0 65s 10.244.112.166 k8scloude2 <none> <none>
nginx-6cf858f6cf-nq6zv 1/1 Running 0 65s 10.244.112.169 k8scloude2 <none> <none>
nginx-6cf858f6cf-pl7ww 1/1 Running 0 65s 10.244.251.235 k8scloude3 <none> <none>
nginx-6cf858f6cf-sf2w6 1/1 Running 0 65s 10.244.112.165 k8scloude2 <none> <none>
pod1 1/1 Running 0 36m 10.244.251.216 k8scloude3 <none> <none>
drain驱逐节点：drain=cordon+evicted
drain k8scloude2节点，--delete-emptydir-data删除数据，--ignore-daemonsets忽略daemonsets
[root@k8scloude1 deploy]# kubectl drain k8scloude2
node/k8scloude2 cordoned
error: unable to drain node "k8scloude2", aborting command...
There are pending nodes to be drained:
k8scloude2
cannot delete Pods with local storage (use --delete-emptydir-data to override): kube-system/metrics-server-bcfb98c76-k5dmj
cannot delete DaemonSet-managed Pods (use --ignore-daemonsets to ignore): kube-system/calico-node-nsbfs, kube-system/kube-proxy-lpj8z
[root@k8scloude1 deploy]# kubectl get node
NAME STATUS ROLES AGE VERSION
k8scloude1 Ready control-plane,master 8d v1.21.0
k8scloude2 Ready,SchedulingDisabled <none> 8d v1.21.0
k8scloude3 Ready <none> 8d v1.21.0
[root@k8scloude1 deploy]# kubectl drain k8scloude2 --ignore-daemonsets
node/k8scloude2 already cordoned
error: unable to drain node "k8scloude2", aborting command...
There are pending nodes to be drained:
k8scloude2
error: cannot delete Pods with local storage (use --delete-emptydir-data to override): kube-system/metrics-server-bcfb98c76-k5dmj
[root@k8scloude1 deploy]# kubectl drain k8scloude2 --ignore-daemonsets --force --delete-emptydir-data
node/k8scloude2 already cordoned
WARNING: ignoring DaemonSet-managed Pods: kube-system/calico-node-nsbfs, kube-system/kube-proxy-lpj8z
evicting pod pod/nginx-6cf858f6cf-sf2w6
evicting pod pod/nginx-6cf858f6cf-5rrk4
evicting pod kube-system/metrics-server-bcfb98c76-k5dmj
evicting pod pod/nginx-6cf858f6cf-58wnd
evicting pod pod/nginx-6cf858f6cf-mb2ft
evicting pod pod/nginx-6cf858f6cf-89wj9
evicting pod pod/nginx-6cf858f6cf-nq6zv
pod/nginx-6cf858f6cf-5rrk4 evicted
pod/nginx-6cf858f6cf-mb2ft evicted
pod/nginx-6cf858f6cf-sf2w6 evicted
pod/nginx-6cf858f6cf-58wnd evicted
pod/nginx-6cf858f6cf-nq6zv evicted
pod/nginx-6cf858f6cf-89wj9 evicted
pod/metrics-server-bcfb98c76-k5dmj evicted
node/k8scloude2 evicted
查看pod，k8scloude2节点被drain之后，pod都调度到了k8scloude3节点。
节点被drain驱逐的本质就是删除节点上的pod，k8scloude2节点被drain驱逐之后，k8scloude2上运行的pod会被删除。
deploy是一个控制器，会监控pod的副本数，当k8scloude2上的pod被驱逐之后，副本数少于10，于是在可调度的节点创建pod,补足副本数。
单独的pod不具备再生性，删除之后就真删除了，如果k8scloude3被驱逐，则pod pod1会被删除，其他可调度节点也不会再生一个pod1。
[root@k8scloude1 deploy]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
nginx-6cf858f6cf-7gh4z 1/1 Running 0 84s 10.244.251.240 k8scloude3 <none> <none>
nginx-6cf858f6cf-7lmfd 1/1 Running 0 85s 10.244.251.238 k8scloude3 <none> <none>
nginx-6cf858f6cf-86wxr 1/1 Running 0 6m14s 10.244.251.237 k8scloude3 <none> <none>
nginx-6cf858f6cf-9bn2b 1/1 Running 0 85s 10.244.251.243 k8scloude3 <none> <none>
nginx-6cf858f6cf-9njrj 1/1 Running 0 6m14s 10.244.251.236 k8scloude3 <none> <none>
nginx-6cf858f6cf-bqk2w 1/1 Running 0 84s 10.244.251.241 k8scloude3 <none> <none>
nginx-6cf858f6cf-hchtb 1/1 Running 0 6m14s 10.244.251.234 k8scloude3 <none> <none>
nginx-6cf858f6cf-hjddp 1/1 Running 0 84s 10.244.251.244 k8scloude3 <none> <none>
nginx-6cf858f6cf-pl7ww 1/1 Running 0 6m14s 10.244.251.235 k8scloude3 <none> <none>
nginx-6cf858f6cf-sgxfg 1/1 Running 0 84s 10.244.251.242 k8scloude3 <none> <none>
pod1 1/1 Running 0 41m 10.244.251.216 k8scloude3 <none> <none>
查看node节点状态
[root@k8scloude1 deploy]# kubectl get nodes
NAME STATUS ROLES AGE VERSION
k8scloude1 Ready control-plane,master 8d v1.21.0
k8scloude2 Ready,SchedulingDisabled <none> 8d v1.21.0
k8scloude3 Ready <none> 8d v1.21.0
4.3 uncordon节点
要取消drain某个节点，直接uncordon即可，没有undrain操作。
[root@k8scloude1 deploy]# kubectl undrain k8scloude2
Error: unknown command "undrain" for "kubectl"
Did you mean this?
drain
Run 'kubectl --help' for usage.
uncordon k8scloude2节点，节点恢复调度
[root@k8scloude1 deploy]# kubectl uncordon k8scloude2
node/k8scloude2 uncordoned
[root@k8scloude1 deploy]# kubectl get nodes
NAME STATUS ROLES AGE VERSION
k8scloude1 Ready control-plane,master 8d v1.21.0
k8scloude2 Ready <none> 8d v1.21.0
k8scloude3 Ready <none> 8d v1.21.0
把deploy副本数变为0，再变为10，再观察pod分布
[root@k8scloude1 deploy]# kubectl scale deploy nginx --replicas=0
deployment.apps/nginx scaled
[root@k8scloude1 deploy]# kubectl get pods -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
pod1 1/1 Running 0 52m 10.244.251.216 k8scloude3 <none> <none>
[root@k8scloude1 deploy]# kubectl scale deploy nginx --replicas=10
deployment.apps/nginx scaled
k8scloude2节点恢复可调度pod状态
[root@k8scloude1 deploy]# kubectl get pods -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
nginx-6cf858f6cf-4sqj8 1/1 Running 0 6s 10.244.112.172 k8scloude2 <none> <none>
nginx-6cf858f6cf-cjqxv 1/1 Running 0 6s 10.244.112.176 k8scloude2 <none> <none>
nginx-6cf858f6cf-fk69r 1/1 Running 0 6s 10.244.112.175 k8scloude2 <none> <none>
nginx-6cf858f6cf-ghznd 1/1 Running 0 6s 10.244.112.173 k8scloude2 <none> <none>
nginx-6cf858f6cf-hnxzs 1/1 Running 0 6s 10.244.251.246 k8scloude3 <none> <none>
nginx-6cf858f6cf-hshnm 1/1 Running 0 6s 10.244.112.171 k8scloude2 <none> <none>
nginx-6cf858f6cf-jb5sh 1/1 Running 0 6s 10.244.112.170 k8scloude2 <none> <none>
nginx-6cf858f6cf-l9xlm 1/1 Running 0 6s 10.244.112.174 k8scloude2 <none> <none>
nginx-6cf858f6cf-pgjlb 1/1 Running 0 6s 10.244.251.247 k8scloude3 <none> <none>
nginx-6cf858f6cf-rlnh6 1/1 Running 0 6s 10.244.251.245 k8scloude3 <none> <none>
pod1 1/1 Running 0 52m 10.244.251.216 k8scloude3 <none> <none>
删除deploy，删除pod。
[root@k8scloude1 deploy]# kubectl delete -f nginx.yaml
deployment.apps "nginx" deleted
[root@k8scloude1 deploy]# kubectl delete pod pod1 --force
warning: Immediate deletion does not wait for confirmation that the running resource has been terminated. The resource may continue to run on the cluster indefinitely.
pod "pod1" force deleted
[root@k8scloude1 deploy]# kubectl get pods -o wide
No resources found in pod namespace.
五.delete 节点
5.1 delete节点概览
delete 删除节点就直接把一个节点就k8s集群中删除了，delete 节点之前需要先drain 节点。
关于delete节点以及重装节点的详细内容，请查看博客《模拟重装Kubernetes(k8s)集群：删除k8s集群然后重装》https://www.cnblogs.com/renshengdezheli/p/16686997.html
5.2 delete节点
kubectl drain 安全驱逐节点上面所有的 pod，--ignore-daemonsets往往需要指定的，这是因为deamonset会忽略SchedulingDisabled标签（使用kubectl drain时会自动给节点打上不可调度SchedulingDisabled标签）,因此deamonset控制器控制的pod被删除后，可能马上又在此节点上启动起来,这样就会成为死循环。因此这里忽略daemonset。
[root@k8scloude1 ~]# kubectl drain k8scloude3 --ignore-daemonsets
node/k8scloude3 cordoned
WARNING: ignoring DaemonSet-managed Pods: kube-system/calico-node-wmz4r, kube-system/kube-proxy-84gcx
evicting pod kube-system/calico-kube-controllers-6b9fbfff44-rl2mh
pod/calico-kube-controllers-6b9fbfff44-rl2mh evicted
node/k8scloude3 evicted
k8scloude3变为SchedulingDisabled
[root@k8scloude1 ~]# kubectl get nodes
NAME STATUS ROLES AGE VERSION
k8scloude1 Ready control-plane,master 64m v1.21.0
k8scloude2 Ready <none> 56m v1.21.0
k8scloude3 Ready,SchedulingDisabled <none> 56m v1.21.0
删除节点k8scloude3
[root@k8scloude1 ~]# kubectl delete nodes k8scloude3
node "k8scloude3" deleted
[root@k8scloude1 ~]# kubectl get nodes
NAME STATUS ROLES AGE VERSION
k8scloude1 Ready control-plane,master 65m v1.21.0
k8scloude2 Ready <none> 57m v1.21.0