Kube Scheduler

2024-12-17

创建调度器配置文件

NodeResourcesFit 是一个调度插件, 检查节点是否拥有 Pod 请求的所有资源, 得分可以使用以下三种策略之一: LeastAllocated (默认)、MostAllocated 和 RequestedToCapacityRatio

实现了多个扩展点: preFilter、filter、preScore、score

我这里自定义使用 MostAllocated 策略, 优选分配比率较高的节点

# /etc/rancher/rke2/kube-scheduler-config.yaml
apiVersion: kubescheduler.config.k8s.io/v1
kind: KubeSchedulerConfiguration
clientConnection:
  kubeconfig: /var/lib/rancher/rke2/server/cred/scheduler.kubeconfig
profiles:
  - schedulerName: default-scheduler
    pluginConfig:
      - name: NodeResourcesFit
        args:
          scoringStrategy:
            type: MostAllocated
            resources:
              - name: cpu
                weight: 1
              - name: memory
                weight: 1

修改 rke2 配置文件

修改 /etc/rancher/rke2/config.yaml

kube-scheduler-arg:
+ - config=/etc/rancher/rke2/kube-scheduler-config.yaml

重启 rke2-server

会重新生成 kube-scheduler 的 static pod manifest 文件 /var/lib/rancher/rke2/agent/pod-manifests/kube-scheduler.yaml

会挂载 /etc/rancher/rke2/kube-scheduler-config.yaml 文件到 pod 中