每天一点基础K8S--K8S中的调度策略--节点亲和性、反亲和性

K8S中的调度策略--节点亲和性、pod亲和性

上面实验了nodeName和nodeSelector，其中，nodeName是通过节点的名称进行区分，在一些特定场景下还是很有用的，如果将节点调度到某一高性能节点。但是nodeName还是显得有点过于严苛。nodeSelector则是通过node的标签进行选择。同样的，通过node的label进行pod资源调度的还有nodeAffinity、podAffinity、podantiAffinity

亲和性语法
kubectl explain pod.spec.affinity

requiredDuringSchedulingIgnoredDuringExecution： 调度器只有在规则被满足的时候才能执行调度。此功能类似于 nodeSelector， 但其语法表达能力更强。

preferredDuringSchedulingIgnoredDuringExecution： 调度器会尝试寻找满足对应规则的节点。如果找不到匹配的节点，调度器仍然会调度该 Pod。


IgnoredDuringExecution 意味着如果节点标签在 Kubernetes 调度 Pod 后发生了变更，Pod 仍将继续运行

节点亲和性nodeAffinity--强亲和性

requiredDuringSchedulingIgnoredDuringExecution

[root@master-worker-node-1 pod]# cat test-node-affinity.yaml 
apiVersion: v1
kind: Pod
metadata:
  name: test-node-affinity-1
  labels:
    function: test-node-affinity
spec:
  containers:
  - name: test-node-affinity-1
    image: busybox:latest
    imagePullPolicy: IfNotPresent
    command: ['/bin/sh','-c','sleep 12345']
  affinity:
    nodeAffinity:
      requiredDuringSchedulingIgnoredDuringExecution:
        nodeSelectorTerms:
        - matchExpressions:
          - key: test-node-affinity
            operator: In
            values: 
            - target-node

创建pod
[root@master-worker-node-1 pod]# kubectl apply -f test-node-affinity.yaml 
pod/test-node-affinity-1 created

因为没有node满足这个label，故pod无法运行
[root@master-worker-node-1 pod]# kubectl get nodes --show-labels | grep target

[root@master-worker-node-1 pod]# kubectl get pods -w 
NAME                   READY   STATUS    RESTARTS   AGE
test-node-affinity-1   0/1     Pending   0          3m50s

pod处于pending状态是因为无法调度
[root@master-worker-node-1 pod]# kubectl describe pods test-node-affinity-1  |  tail -4
Events:
  Type     Reason            Age    From               Message
  ----     ------            ----   ----               -------
  Warning  FailedScheduling  4m42s  default-scheduler  0/4 nodes are available: 2 node(s) had untolerated taint {node-role.kubernetes.io/control-plane: }, 4 node(s) didn't match Pod's node affinity/selector. preemption: 0/4 nodes are available: 4 Preemption is not helpful for scheduling.

给某一节点添加label
[root@master-worker-node-1 pod]# kubectl label nodes only-worker-node-3 test-node-affinity=target-node
node/only-worker-node-3 labeled


一会发现后，pod运行正常
[root@master-worker-node-1 pod]# kubectl get pods -w 
NAME                   READY   STATUS    RESTARTS   AGE
test-node-affinity-1   0/1     Pending   0          3m50s
test-node-affinity-1   0/1     Pending   0          6m51s
test-node-affinity-1   0/1     ContainerCreating   0          6m51s
test-node-affinity-1   0/1     ContainerCreating   0          6m52s
test-node-affinity-1   1/1     Running             0          6m53s

节点亲和性nodeAffinity--弱亲和性

preferredDuringSchedulingIgnoredDuringExecution

先删除环境中所有node的label
[root@master-worker-node-1 pod]# kubectl label nodes only-worker-node-3 test-node-affinity-
node/only-worker-node-3 unlabeled

[root@master-worker-node-1 pod]# cat test-node-affinity-2.yaml 
apiVersion: v1
kind: Pod
metadata:
  name: test-node-affinity-2
  labels:
    function: test-node-affinity
spec:
  containers:
  - name: test-node-affinity-2
    image: busybox:latest
    imagePullPolicy: IfNotPresent
    command: ['/bin/sh','-c','sleep 12345']
  affinity:
    nodeAffinity:
      preferredDuringSchedulingIgnoredDuringExecution:
      - weight: 40
        preference:
          matchExpressions:
          - key: test-node-affinity
            operator: In
            values:
            - target-node-1

创建pod
[root@master-worker-node-1 pod]# kubectl apply -f test-node-affinity-2.yaml 
pod/test-node-affinity-2 created

虽然没有任何一个node满足pod prefer的label要求，但是pod仍然可以正常运行
[root@master-worker-node-1 pod]# kubectl get pods -o wide
NAME                   READY   STATUS    RESTARTS   AGE   IP            NODE                 NOMINATED NODE   READINESS GATES
test-node-affinity-2   1/1     Running   0          26s   10.244.54.8   only-worker-node-4   <none>           <none>

弱亲和性里面的weight字段需要有两个选项才有意思

[root@master-worker-node-1 pod]# cat test-node-affinity-3.yaml 
apiVersion: v1
kind: Pod
metadata:
  name: test-node-affinity-3
  labels:
    function: test-node-affinity
spec:
  containers:
  - name: test-node-affinity-3
    image: busybox:latest
    imagePullPolicy: IfNotPresent
    command: ['/bin/sh','-c','sleep 12345']
  affinity:
    nodeAffinity:
      preferredDuringSchedulingIgnoredDuringExecution:
      - weight: 40    # 有两个可选项的时候，weight才用意义。
        preference:
          matchExpressions:
          - key: test-node-affinity
            operator: In
            values:
            - target-node-1
      - weight: 80
        preference:
          matchExpressions:
          - key: test-node-affinity
            operator: In
            values:
            - target-node-2

给node添加对应的label
[root@master-worker-node-1 pod]# kubectl label nodes only-worker-node-3 test-node-affinity=target-node-1
node/only-worker-node-3 labeled
[root@master-worker-node-1 pod]# kubectl label nodes only-worker-node-4 test-node-affinity=target-node-2
node/only-worker-node-4 labeled

因为target-node-2的weight的权重更大，按理应该会被调度到only-worker-node-4上

[root@master-worker-node-1 pod]# kubectl apply -f test-node-affinity-3.yaml 
pod/test-node-affinity-3 created
[root@master-worker-node-1 pod]# kubectl get pods -o wide
NAME                   READY   STATUS    RESTARTS   AGE   IP            NODE                 NOMINATED NODE   READINESS GATES
test-node-affinity-2   1/1     Running   0          18m   10.244.54.8   only-worker-node-4   <none>           <none>
test-node-affinity-3   1/1     Running   0          12s   10.244.54.9   only-worker-node-4   <none>           <none>

节点反亲和性

节点反亲和性没有专门的语法，可以通过operator中的
NotIn、DoesNotExist和node污点实现。

小结

node节点亲和性分为强亲和性（required）和弱亲和性（preferred）

required是只有满足了才会被调度，perferred是尽力满足，无法满足也能调度

node反亲和性可通过NotIn、DoesNotExist和node污点实现