和前面学习的Affinity类似的，还可以直接在pod的声明中指定一个node去调度。这种方式简单粗暴，我们一起来学习看。
我是T型人小付，一位坚持终身学习的互联网从业者。喜欢我的博客欢迎在csdn上关注我，如果有问题欢迎在底下的评论区交流，谢谢。
实践操作
固定节点因为比较简单粗暴，所以直接上操作。分为两类，一类是指定单个node，另一类是指定node的label。
以下所有yaml文件托管在我的Github仓库
单个node
直接在pod.spec.nodeName字段声明node的名字即可。
通过下面的yaml文件test-nodename.yaml来创建deployment
apiVersion: extensions/v1beta1
kind: Deployment
metadata:
name: test-nodename
spec:
replicas: 3
template:
metadata:
labels:
app: deployment-app
spec:
nodeName: k8s-node1
containers:
- name: mynginx
image: mynginx:v2
ports:
- containerPort: 80所有的pod都在node1
[root@k8s-master nodename]# kubectl apply -f test-nodename.yaml
deployment.extensions/test-nodename created
[root@k8s-master nodename]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
test-nodename-6f8c4686f5-4vx9h 1/1 Running 0 5s 10.244.1.143 k8s-node1 <none> <none>
test-nodename-6f8c4686f5-59mq5 1/1 Running 0 5s 10.244.1.142 k8s-node1 <none> <none>
test-nodename-6f8c4686f5-rzs8c 1/1 Running 0 5s 10.244.1.141 k8s-node1 <none> <none>即使扩容也是
[root@k8s-master nodename]# kubectl scale --replicas=7 deployment/test-nodename
deployment.extensions/test-nodename scaled
[root@k8s-master nodename]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
test-nodename-6f8c4686f5-4vx9h 1/1 Running 0 115s 10.244.1.143 k8s-node1 <none> <none>
test-nodename-6f8c4686f5-59mq5 1/1 Running 0 115s 10.244.1.142 k8s-node1 <none> <none>
test-nodename-6f8c4686f5-82trn 1/1 Running 0 4s 10.244.1.145 k8s-node1 <none> <none>
test-nodename-6f8c4686f5-rzs8c 1/1 Running 0 115s 10.244.1.141 k8s-node1 <none> <none>
test-nodename-6f8c4686f5-sr6q5 1/1 Running 0 4s 10.244.1.146 k8s-node1 <none> <none>
test-nodename-6f8c4686f5-tvx7g 1/1 Running 0 4s 10.244.1.147 k8s-node1 <none> <none>
test-nodename-6f8c4686f5-z7lcf 1/1 Running 0 4s 10.244.1.144 k8s-node1 <none> <none>测试一下如果对node1加上污点会是啥情况
警告：下面的操作可能会很危险，谨慎尝试
kubectl taint nodes k8s-node1 xiaofu=good:NoExecute之后这些pod因为污点而被销毁，然后又因为Deployment而被自动创建，然后又被销毁，于是满屏幕都是Terminating
[root@k8s-master nodename]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
test-nodename-6f8c4686f5-22dr7 0/1 Terminating 0 4m1s <none> k8s-node1 <none> <none>
test-nodename-6f8c4686f5-26548 0/1 Terminating 0 4m18s <none> k8s-node1 <none> <none>
test-nodename-6f8c4686f5-27wsv 0/1 Terminating 0 4m12s <none> k8s-node1 <none> <none>
test-nodename-6f8c4686f5-28vpt 0/1 Terminating 0 4m13s <none> k8s-node1 <none> <none>
test-nodename-6f8c4686f5-29gpl 0/1 Terminating 0 4m28s <none> k8s-node1 <none> <none>
test-nodename-6f8c4686f5-29lnj 0/1 Terminating 0 4m21s <none> k8s-node1 <none> <none>
test-nodename-6f8c4686f5-2b5dc 0/1 Terminating 0 4m17s <none> k8s-node1 <none> <none>
test-nodename-6f8c4686f5-2fqw2 0/1 Terminating 0 4m20s <none> k8s-node1 <none> <none>
test-nodename-6f8c4686f5-2ksnn 0/1 Terminating 0 4m2s <none> k8s-node1 <none> <none>
test-nodename-6f8c4686f5-2r2gs 0/1 Terminating 0 4m20s <none> k8s-node1 <none> <none>
test-nodename-6f8c4686f5-2st4d 0/1 Terminating 0 4m19s <none> k8s-node1 <none> <none>
test-nodename-6f8c4686f5-2xdvr 0/1 Terminating 0 4m27s <none> k8s-node1 <none> <none>
test-nodename-6f8c4686f5-2xphz 0/1 Terminating 0 4m20s <none> k8s-node1 <none> <none>
test-nodename-6f8c4686f5-44tps 0/1 Terminating 0 4m15s <none> k8s-node1 <none> <none>
test-nodename-6f8c4686f5-4gbpp 0/1 Terminating 0 4m22s <none> k8s-node1 <none> <none>
test-nodename-6f8c4686f5-4gr6m 0/1 Terminating 0 4m17s <none> k8s-node1 <none> <none>
test-nodename-6f8c4686f5-4jb6w 0/1 Terminating 0 4m16s <none> k8s-node1 <none> <none>
test-nodename-6f8c4686f5-4lbms 0/1 Terminating 0 4m6s <none> k8s-node1 <none> <none>
test-nodename-6f8c4686f5-4m4fm 0/1 Terminating 0 4m25s <none> k8s-node1 <none> <none>
test-nodename-6f8c4686f5-4mcsk 0/1 Terminating 0 4m17s <none> k8s-node1 <none> <none>
...
...赶紧去掉污点，删除deployment，等着几百个pod自己慢慢销毁吧。
node标签选择器
也可以在字段pod.spec.nodeSelector中指定node要满足的label。
通过下面的yaml文件test-nodeselector.yaml创建deployment
apiVersion: extensions/v1beta1
kind: Deployment
metadata:
name: test-nodeselector
spec:
replicas: 3
template:
metadata:
labels:
app: deployment-app
spec:
nodeSelector:
level: boss
containers:
- name: mynginx
image: mynginx:v2
ports:
- containerPort: 80这里希望node有level:boss的标签，但是没有一个node有这个标签，于是所有pod处于pending状态
[root@k8s-master nodename]# kubectl apply -f test-nodeselector.yaml
deployment.extensions/test-nodeselector created
[root@k8s-master nodename]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
test-nodeselector-b8fd4ccf-2sc6c 0/1 Pending 0 4s <none> <none> <none> <none>
test-nodeselector-b8fd4ccf-8xjtl 0/1 Pending 0 4s <none> <none> <none> <none>
test-nodeselector-b8fd4ccf-mcn8l 0/1 Pending 0 4s <none> <none> <none> <none>给node1打一个label之后成功被调度
[root@k8s-master nodename]# kubectl label node k8s-node1 level=boss
node/k8s-node1 labeled
[root@k8s-master nodename]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
test-nodeselector-b8fd4ccf-2sc6c 1/1 Running 0 87s 10.244.1.175 k8s-node1 <none> <none>
test-nodeselector-b8fd4ccf-8xjtl 1/1 Running 0 87s 10.244.1.176 k8s-node1 <none> <none>
test-nodeselector-b8fd4ccf-mcn8l 1/1 Running 0 87s 10.244.1.174 k8s-node1 <none> <none>删除label
[root@k8s-master nodename]# kubectl label node k8s-node1 level-
node/k8s-node1 labeled总结
这样三种自定义调度pod的方式就都学习完了
- Affinity/AntiAffinity
- Taint/Toleration
- Nodename/Nodeselector
