跳转至

FilterPods

官方文档:

在驱逐 Pods 的时候,有时并不需要所有 Pods 都被驱逐,descheduler 提供了两种主要的方式进行过滤:命名空间过滤和优先级过滤

命名空间过滤

该策略可以配置是包含还是排除某些名称空间。可以使用该策略的有:

  • PodLifeTime
  • RemovePodsHavingTooManyRestarts
  • RemovePodsViolatingNodeTaints
  • RemovePodsViolatingNodeAffinity
  • RemovePodsViolatingInterPodAntiAffinity
  • RemoveDuplicates
  • RemovePodsViolatingTopologySpreadConstraint
  • RemoveFailedPods
  • LowNodeUtilization and HighNodeUtilization (Only filtered right before eviction)

比如只驱逐某些命令空间下的 Pods,则可以使用 include 参数进行配置,如下所示:

apiVersion: "descheduler/v1alpha1"
kind: "DeschedulerPolicy"
strategies:
  "PodLifeTime":
     enabled: true
     params:
        podLifeTime:
          maxPodLifeTimeSeconds: 86400
        namespaces:
          include:
          - "namespace1"
          - "namespace2"

又或者要排除掉某些命令空间下的 Pods,则可以使用 exclude 参数配置,如下所示:

apiVersion: "descheduler/v1alpha1"
kind: "DeschedulerPolicy"
strategies:
  "PodLifeTime":
     enabled: true
     params:
        podLifeTime:
          maxPodLifeTimeSeconds: 86400
        namespaces:
          exclude:
          - "namespace1"
          - "namespace2"

优先级过滤

所有策略都可以配置优先级阈值,只有在该阈值以下的 Pod 才会被驱逐,我们可以通过设置 thresholdPriorityClassName(将阈值设置为指定优先级类别的值)或 thresholdPriority(直接设置阈值)参数来指定该阈值

默认情况下,该阈值设置为 system-cluster-critical 这个 PriorityClass 类的值

比如使用 thresholdPriority

apiVersion: "descheduler/v1alpha1"
kind: "DeschedulerPolicy"
strategies:
  "PodLifeTime":
     enabled: true
     params:
        podLifeTime:
          maxPodLifeTimeSeconds: 86400
        thresholdPriority: 10000

或者使用 thresholdPriorityClassName 进行过滤:

apiVersion: "descheduler/v1alpha1"
kind: "DeschedulerPolicy"
strategies:
  "PodLifeTime":
     enabled: true
     params:
        podLifeTime:
          maxPodLifeTimeSeconds: 86400
        thresholdPriorityClassName: "priorityclass1"

不过需要注意不能同时配置 thresholdPrioritythresholdPriorityClassName

如果指定的优先级类不存在,则 descheduler 不会创建它,并且会引发错误