zchengcantstop/scheduler-extender-master

k8s自定义调度扩展器

sample-scheduler-extender

A sample to showcase how to create a k8s scheduler extender.

if you cluster is installed by kubeadm，maybe your kube-scheduler's yaml like this:

apiVersion: v1
kind: Pod
metadata:
  creationTimestamp: null
  labels:
    component: kube-scheduler
    tier: control-plane
  name: kube-scheduler
  namespace: kube-system
spec:
  containers:
  - command:
    - kube-scheduler
    - --authentication-kubeconfig=/etc/kubernetes/scheduler.conf
    - --authorization-kubeconfig=/etc/kubernetes/scheduler.conf
    - --bind-address=127.0.0.1
    - --kubeconfig=/etc/kubernetes/scheduler.conf
    - --leader-elect=true
    - --config=/etc/kubernetes/scheduler-extender.yaml
    - --v=9
    image: gcr.azk8s.cn/google_containers/kube-scheduler:v1.16.2
    imagePullPolicy: IfNotPresent
    livenessProbe:
      failureThreshold: 8
      httpGet:
        host: 127.0.0.1
        path: /healthz
        port: 10251
        scheme: HTTP
      initialDelaySeconds: 15
      timeoutSeconds: 15
    name: kube-scheduler
    resources:
      requests:
        cpu: 100m
    volumeMounts:
    - mountPath: /etc/kubernetes/scheduler.conf
      name: kubeconfig
      readOnly: true
    - mountPath: /etc/kubernetes/scheduler-extender.yaml
      name: extender
      readOnly: true
    - mountPath: /etc/kubernetes/scheduler-extender-policy.yaml
      name: extender-policy
      readOnly: true
  hostNetwork: true
  priorityClassName: system-cluster-critical
  volumes:
  - hostPath:
      path: /etc/kubernetes/scheduler.conf
      type: FileOrCreate
    name: kubeconfig
  - hostPath:
      path: /etc/kubernetes/scheduler-extender.yaml
      type: FileOrCreate
    name: extender
  - hostPath:
      path: /etc/kubernetes/scheduler-extender-policy.yaml
      type: FileOrCreate
    name: extender-policy
status: {}

You must mount scheduler-extender.yaml 和 scheduler-extender-policy.yaml to Pod.

Notes

• Because k8s.io/kubernetes pkg is not intended to be used with go get,so we need use go modules replace to depend these modules, so you must replace go.mod to your local k8s path .

• Prioritize webhook won't be triggered if it's running on an one-node cluster. As it makes no sense to run priorities logic when there is only one candidate:

// from k8s.io/kubernetes/pkg/scheduler/core/generic_scheduler.go
func (g *genericScheduler) Schedule(pod *v1.Pod, nodeLister algorithm.NodeLister) (string, error) {
    ...
	// When only one node after predicate, just use it.
	if len(filteredNodes) == 1 {
		metrics.SchedulingAlgorithmPriorityEvaluationDuration.Observe(metrics.SinceInMicroseconds(startPriorityEvalTime))
		return filteredNodes[0].Name, nil
    }
    ...
}