Proportion Plugin User Guide

Introduction

The Proportion scheduling plugin ensures fair resource sharing across different queues based on their configured weights. It dynamically controls the maximum resources a queue can consume, preventing a single team or workload from starving others in a multi-tenant cluster. When the cluster is under heavy load, the plugin allocates resources proportionally to each queue’s weight.

Environment setup

Update scheduler configmap

1. Ensure the proportion plugin is enabled in the volcano-scheduler-configmap.

2. Define multiple Queue resources with different weights.

   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: proportion
     - name: gang
   

How to use Proportion Plugin

First, create the Queues with specific weights. The weight represents the relative share of the cluster’s resources that the queue is entitled to.

Step 1: Create Queues

Create a file named queues.yaml:

apiVersion: scheduling.volcano.sh/v1beta1
kind: Queue
metadata:
  name: team-a-queue
spec:
  weight: 2
  capability:
    cpu: "10"
    memory: "20Gi"
---
apiVersion: scheduling.volcano.sh/v1beta1
kind: Queue
metadata:
  name: team-b-queue
spec:
  weight: 1
  capability:
    cpu: "10"
    memory: "20Gi"

In this scenario, team-a-queue is entitled to twice as many resources as team-b-queue when there is resource contention.

Step 2: Submit Jobs to the Queues

Create a file named proportion-jobs.yaml:

apiVersion: batch.volcano.sh/v1alpha1
kind: Job
metadata:
  name: job-team-a
spec:
  queue: team-a-queue
  schedulerName: volcano
  tasks:
    - replicas: 10
      name: worker
      template:
        spec:
          containers:
            - image: nginx
              name: nginx
              resources:
                requests:
                  cpu: "1"
          restartPolicy: OnFailure
---
apiVersion: batch.volcano.sh/v1alpha1
kind: Job
metadata:
  name: job-team-b
spec:
  queue: team-b-queue
  schedulerName: volcano
  tasks:
    - replicas: 10
      name: worker
      template:
        spec:
          containers:
            - image: nginx
              name: nginx
              resources:
                requests:
                  cpu: "1"
          restartPolicy: OnFailure

Verification

1. Apply the Queues and the Jobs:

   kubectl apply -f queues.yaml
   kubectl apply -f proportion-jobs.yaml
   

2. If your cluster has exactly 9 available CPUs, Volcano will schedule pods proportionally according to the queue weights (2:1).

   • team-a-queue (Weight 2) will get roughly 6 pods running.
   • team-b-queue (Weight 1) will get roughly 3 pods running.
     

3. You can verify this by checking the number of running pods for each job:

   kubectl get pods -l volcano.sh/job-name=job-team-a
   kubectl get pods -l volcano.sh/job-name=job-team-b