K8S-POD优先级与抢占

1、概念

在定义pod时，可以在spec.priorityClassName中指定PriorityClass，根据其中定义的优先级在PrioritySort中排序，优先调度优先级高的pod，优先级相同的pod根据进入队列的时间戳先后调度，当未找到

合适的运行节点时，调度器会将POD转为pending状态，并为其启动“抢占”过程，在集群中删除一个或者多个低优先的POD，让节点满足该优先级高的POD调度。

pod优先级使用32位正整数，可用值为[0,1000000000],值越大优先级越高，大于1000000000的优先级预留给系统级的关键pod，以防止这些pod被驱逐。

例如：API-server，Controller-manager，Scheduler和etcd 的pod直接使用system-cluster-critical PriorityClass，优先级为：2000000000

           metric-server,CoreDNS、Dashboard等使用system-node-critical 的pod使用system-node-critical PriorityClass,优先级为：2000001000

在定义pod时不定义PriorityClass时，默认优先级的值为：0

2、PriorityClass定义

 

如果集群上存在多个设定了全局默认的优先级的PriorityClass对象，优先级小的会生效，如： 

定义默认优先级的demoappv11

apiVersion: apps/v1
kind: Deployment
metadata:
  name: demoappv11
spec:
  replicas: 4
  selector:
    matchLabels:
      app: demoappv11
  template:
    metadata:
      labels:
        app: demoappv11
    spec:
      affinity:
        podAntiAffinity:
          requiredDuringSchedulingIgnoredDuringExecution:
          - labelSelector:
              matchExpressions:
              - {key: app, operator: In, values: ["demoappv11"]}
            topologyKey: kubernetes.io/hostname
      containers:
      - name: demoappv11
        image: harbor.myland.com/baseimages/centos/centos-tsinghua:7.9.2009 
        command: ["sleep","1000"]
        resources:
          requests:
            memory: 2Gi
            cpu: 150m

 

优先级为222222的demoappv12

kind: PriorityClass
apiVersion: scheduling.k8s.io/v1
metadata:
  name: demoappv12
value: 222222
description: "demoappv12"
globalDefault: false
preemptionPolicy: PreemptLowerPriority
---
apiVersion: apps/v1
kind: Deployment
metadata:
  name: demoappv12
spec:
  replicas: 4
  selector:
    matchLabels:
      app: demoappv12
  template:
    metadata:
      labels:
        app: demoappv12
    spec:
      priorityClassName: demoappv12
      affinity:
        podAntiAffinity:
          requiredDuringSchedulingIgnoredDuringExecution:
          - labelSelector:
              matchExpressions:
              - {key: app, operator: In, values: ["demoappv12"]}
            topologyKey: kubernetes.io/hostname
      containers:
      - name: demoappv12
        image: ikubernetes/demoapp:v1.2
        resources:
          requests:
            memory: 2Gi
            cpu: 150m

由于内存不足，调度器开始清理优先级低的demoappv11

 

直到demoappv11处于pending状态