kubernetes (k8s) CKA认证之第二课：亲和性与 Pod 的调度

手动调度一个 pod

// cat manual-schedule.yaml
apiVersion: v1
kind: Pod
metadata:
  labels:
    run: pod-manual-schedule
  name: pod-manual-schedule
  namespace: default
spec:
  nodeName: "app.centos7.com"
  containers:
  - name: my-pod
    image: nginx:alpine

执行以下命令即可验证。

kubectl create -f manual-schedule.yaml
kubectl get po -owide
kubectl describe po pod-manual-schedule|less

kubectl delete -f manual-schedule.yaml

NodeAffinity 结点的亲和性

kubectl label node app.centos7.com has-eip=yes
kubectl get node --show-labels

vi node-affinity.yaml

apiVersion: v1
kind: Pod
metadata:
  labels:
    run: node-affinity
  name: node-affinity
  namespace: default
spec:
  containers:
  - name: node-affinity
    image: nginx:alpine
    imagePullPolicy: IfNotPresent
  affinity:
    nodeAffinity:
      requiredDuringSchedulingIgnoredDuringExecution:
        nodeSelectorTerms:
        - matchExpressions:
          - key: has-eip
            operator: In
            values:
            - "yes"

运行以下命令加以验证：

kubectl create -f node-affinity.yaml
kubectl get po -owide

结果：

[root@svn ch02]# kubectl get po -owide
NAME            READY   STATUS    RESTARTS   AGE     IP             NODE              NOMINATED NODE   READINESS GATES
node-affinity   1/1     Running   0          5m49s   10.244.2.117   app.centos7.com   <none>           <none>
[root@svn ch02]#

PodAffinity (保留上一步的 node-affinity 处于 Running 状态 ) “豆荚”的亲和性

vi pod-affinity.yaml

apiVersion: v1
kind: Pod
metadata:
  name: pod-affinity
  labels:
    run: pod-affinity
spec:
  containers:
  - name: pod-affinity
    image: nginx:alpine
  affinity:
    podAffinity:
      requiredDuringSchedulingIgnoredDuringExecution:
      - labelSelector:
          matchExpressions:
          - key: run
            operator: In
            values:
            - "node-affinity"
        topologyKey: kubernetes.io/hostname

运行结果:

[root@svn ch02]# kubectl get po --show-labels -owide
NAME            READY   STATUS    RESTARTS   AGE    IP             NODE              NOMINATED NODE   READINESS GATES   LABELS
node-affinity   1/1     Running   0          129m   10.244.2.117   app.centos7.com   <none>           <none>            run=node-affinity
[root@svn ch02]# kubectl create -f pod-affinity.yaml
pod/pod-affinity created
[root@svn ch02]# kubectl get po --show-labels -owide
NAME            READY   STATUS    RESTARTS   AGE    IP             NODE              NOMINATED NODE   READINESS GATES   LABELS
node-affinity   1/1     Running   0          129m   10.244.2.117   app.centos7.com   <none>           <none>            run=node-affinity
pod-affinity    1/1     Running   0          4s     10.244.2.118   app.centos7.com   <none>           <none>            run=pod-affinity
[root@svn ch02]#

pod-anti-affinity 反亲和性

前提：保留上一步的 node-affinity 和 pod-affinity 处于 Running 状态
vi pod-anti-affinity.yaml

apiVersion: v1
kind: Pod
metadata:
  name: pod-anti-affinity
  labels:
    run: pod-anti-affinity
spec:
  containers:
  - name: pod-affinity
    image: nginx:alpine
  affinity:
    podAntiAffinity:
      requiredDuringSchedulingIgnoredDuringExecution:
      - labelSelector:
          matchExpressions:
          - key: run
            operator: In
            values:
            - "pod-affinity"
        topologyKey: kubernetes.io/hostname

由于反亲和性，新生成的 pod-anti-affinity 将会被调度到 app.centos7.com 之外的结点。
运行结果：

[root@svn ch02]# kc create -f pod-anti-affinity.yaml
pod/pod-anti-affinity created
[root@svn ch02]# kc get pod -owide --show-labels
NAME                READY   STATUS    RESTARTS   AGE     IP             NODE              NOMINATED NODE   READINESS GATES   LABELS
node-affinity       1/1     Running   0          153m    10.244.2.117   app.centos7.com   <none>           <none>            run=node-affinity
pod-affinity        1/1     Running   0          24m     10.244.2.118   app.centos7.com   <none>           <none>            run=pod-affinity
pod-anti-affinity   1/1     Running   0          3m18s   10.244.1.9     jks.centos7.com   <none>           <none>            run=pod-anti-affinity
[root@svn ch02]#

Taints 与 Tolerations 污点与容忍

// 为 master 节点配置 taints 禁止普通 pod 调度到 master 上
kubectl taint nodes svn.centos7.com node-role.kubernetes.io/master=:NoSchedule

vi pod-tolerations.yaml

apiVersion: v1
kind: Pod
metadata:
  labels:
    run: pod-tolerations
  name: pod-tolerations
spec:
  containers:
  - name: my-pod
    image: nginx:alpine
  tolerations:
  - key: gpu
    operator: Equal
    value: "yes"
    effect: NoSchedule

给 node jks.centos7.com 增添对应的污点，如此，pod-tolerations 才可被调度在其之上。

kubectl taint node jks.centos7.com gpu=no:NoSchedule 
kubectl taint node jks.centos7.com gpu=yes:NoSchedule --overwrite

运行结果：

[root@svn ch02]# kubectl create -f pod-tolerations.yaml
pod/pod-tolerations created
[root@svn ch02]# kubectl get po -owide
NAME              READY   STATUS    RESTARTS   AGE   IP            NODE              NOMINATED NODE   READINESS GATES
pod-tolerations   1/1     Running   0          9s    10.244.1.18   jks.centos7.com   <none>           <none>
[root@svn ch02]#

References

• kubernetes 对node添加Label并进行选择
• Kubernetes taints 配置