Inter-pod topological affinity and anti-affinity（pod之间拓扑上的关联与反关联）

翻译

podAffinity

requiredDuringSchedulingIgnoredDuringExecution

apiVersion: v1
kind: Pod
metadata:
  name: test1
spec:
  affinity:
    podAffinity:
      requiredDuringSchedulingIgnoredDuringExecution:
      - labelSelector:
          matchExpressions:
          - key: "mysvc"
            operator: In
            values:
            - S1
        topologyKey: kubernetes.io/hostname
  containers:
  - name: dao-2048
    image: daocloud.io/daocloud/dao-2048
    imagePullPolicy: IfNotPresent

1. 通过该文件创建得到的pod，一直处于pending状态，错误原因如下图
   ，可知当使用requiredDuringSchedulingIgnoredDuringExecution参数时，如果无法满足affinity条件，pod无法启动
2. 执行kubectl label po test1 mysvc=S1``kubectl label po test1 mysvc=S1修改pod标签使pod满足条件，通过kubectl get pod，可观察到pod状态变为running。再执行kubectl label po test1 mysvc-删除pod的标签，使pod不满足affinity条件，发现pod转态仍为running。由此确定schedule状态包含pending而且在IgnoredDuringExecutio
n条件下，在执行期间发生条件改变，对pod的状态是不会产生影响的。

preferredDuringSchedulingIgnoredDuringExecution

apiVersion: v1
kind: Pod
metadata:
  name: test1
spec:
  affinity:
    podAffinity:
      preferredDuringSchedulingIgnoredDuringExecution:
      - weight: 100
        podAffinityTerm:
          labelSelector:
            matchExpressions:
            - key: security
              operator: In
              values:
              - S1
          topologyKey: kubernetes.io/hostname
  containers:
  - name: dao-2048
    image: daocloud.io/daocloud/dao-2048
    imagePullPolicy: IfNotPresent

通过配置文件产生pod，观察可得pod的状态为running，由此可知preferredDuringScheduling下的affinity为非必须条件。

podAntiAffinity

requiredDuringSchedulingIgnoredDuringExecution

apiVersion: v1
kind: Pod
metadata:
  name: test4
  labels:
    security: S1
spec:
  affinity:
    podAntiAffinity:
      requiredDuringSchedulingIgnoredDuringExecution:
      - labelSelector:
          matchExpressions:
          - key: security
            operator: In
            values:
            - S1
        topologyKey: kubernetes.io/hostname
  containers:
  - name: dao-2048
    image: daocloud.io/daocloud/dao-2048
    imagePullPolicy: IfNotPresent

使用该文件创建3个pod，可以发现3个pod运行在不同的node上。。如果再创建一个pod，可以观察到该pod处于pending状态。因为在该环境只有三台worknode，符合条件的所有worknode均被占用。

preferredDuringSchedulingIgnoredDuringExecution

apiVersion: v1
kind: Pod
metadata:
  name: test3
  labels:
    security: S1
spec:
  affinity:
    podAntiAffinity:
      preferredDuringSchedulingIgnoredDuringExecution:
      - weight: 100
        podAffinityTerm:
          labelSelector:
            matchExpressions:
            - key: security
              operator: In
              values:
              - S1
          topologyKey: kubernetes.io/hostname
  containers:
  - name: dao-2048
    image: daocloud.io/daocloud/dao-2048
    imagePullPolicy: IfNotPresent

使用该配置模板创建三个pod，可以发现pod依旧分配到了不同的节点上。当创建第四个pod时，第四个pod能够被顺利创建，说明preferredDuringScheduling在podAntiAnffinity下也是不严格匹配规则。

podAntiAffinity的对称性

apiVersion: v1
kind: Pod
metadata:
  name: test3
  labels:
    security: S1
spec:
  containers:
  - name: dao-2048
    image: daocloud.io/daocloud/dao-2048
    imagePullPolicy: IfNotPresent

以及

apiVersion: v1
kind: Pod
metadata:
  name: test2
  labels:
    security: S1
spec:
  affinity:
    podAntiAffinity:
      requiredDuringSchedulingIgnoredDuringExecution:
      - labelSelector:
          matchExpressions:
          - key: security
            operator: In
            values:
            - S1
        topologyKey: kubernetes.io/hostname
  containers:
  - name: dao-2048
    image: daocloud.io/daocloud/dao-2048
    imagePullPolicy: IfNotPresent

先使用第二个配置文件创建两个pod，再用第一个文件创建一个pod，可以观察到，三个pod在三个不同的node上。，说明当设置test1对test2排他，不用设置test2对test1排他。