NodeOrder Plugin User Guide

Introduction

The NodeOrder plugin is responsible for scoring nodes during the scheduling process to find the absolute best node for a pod. Rather than just finding any node that fits the pod’s requirements, NodeOrder ranks the nodes based on configurable scoring dimensions, such as affinity rules or resource availability.

Environment setup

Update scheduler configmap

The nodeorder plugin must be enabled in the scheduler configuration.

kind: ConfigMap
apiVersion: v1
metadata:
  name: volcano-scheduler-configmap
  namespace: volcano-system
data:
  volcano-scheduler.conf: |
    actions: "enqueue, allocate, backfill"
    tiers:
    - plugins:
      - name: priority
      - name: nodeorder
      - name: gang

How to use NodeOrder Plugin

The NodeOrder plugin scores nodes automatically based on built-in Kubernetes concepts like NodeAffinity, PodAffinity, and PodAntiAffinity.

Example YAML

Create a file named nodeorder-job.yaml:

apiVersion: batch.volcano.sh/v1alpha1
kind: Job
metadata:
  name: nodeorder-job
spec:
  minAvailable: 1
  schedulerName: volcano
  tasks:
    - replicas: 3
      name: worker
      template:
        spec:
          affinity:
            nodeAffinity:
              preferredDuringSchedulingIgnoredDuringExecution:
              - weight: 100
                preference:
                  matchExpressions:
                  - key: disktype
                    operator: In
                    values:
                    - ssd
          containers:
            - image: nginx
              name: nginx
              resources:
                requests:
                  cpu: "1"
          restartPolicy: OnFailure

In this example, the nodeAffinity specifies a preferredDuringSchedulingIgnoredDuringExecution rule. The NodeOrder plugin takes this preference into account. It will give a higher score (boosted by the weight of 100) to any node that has the label disktype=ssd.

Verification

1. Label one of your nodes with disktype=ssd:

   kubectl label nodes <your-node-name> disktype=ssd
   

2. Apply the job:

   kubectl apply -f nodeorder-job.yaml
   

3. Check where the pods were scheduled:

   kubectl get pods -o wide | grep nodeorder-job
   

4. You will observe that the Volcano scheduler strongly prefers to place the pods on the node labeled with disktype=ssd due to the high score calculated by the NodeOrder plugin.