Predicates Plugin User Guide

Introduction

The Predicates plugin is the primary filtering engine for Volcano. It evaluates each node against the requirements of a pod to filter out nodes that are fundamentally incapable of running the pod. This includes checking if a node has enough CPU/Memory available, evaluating node selectors, matching node affinities, and checking if the node has the requested GPUs.

Environment setup

Update scheduler configmap

The predicates plugin is a fundamental plugin and should almost always be enabled in your volcano-scheduler-configmap.

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: predicates
      - name: gang

How to use Predicates Plugin

You do not need to do anything special to “trigger” the Predicates plugin. It automatically evaluates standard Kubernetes pod constraints.

Example YAML (Filtering by NodeSelector)

Create a file named predicates-job.yaml:

apiVersion: batch.volcano.sh/v1alpha1
kind: Job
metadata:
  name: predicates-job
spec:
  minAvailable: 1
  schedulerName: volcano
  tasks:
    - replicas: 1
      name: worker
      template:
        spec:
          nodeSelector:
            accelerator: nvidia-tesla-v100
          containers:
            - image: nginx
              name: nginx
              resources:
                requests:
                  cpu: "1"
          restartPolicy: OnFailure

In this example, the nodeSelector requires a node with the label accelerator=nvidia-tesla-v100. The Predicates plugin will filter out all nodes in the cluster that do not have this exact label.

Verification

1. Apply the job:

   kubectl apply -f predicates-job.yaml
   

2. If you do not have any nodes labeled with accelerator=nvidia-tesla-v100, the pod will remain in the Pending state.

3. Describe the pod to see the Predicates plugin at work:

   kubectl describe pod -l volcano.sh/job-name=predicates-job
   

   You will see an event from the Volcano scheduler stating that 0 nodes were available because none matched the node selector.

4. Label a node to satisfy the predicate:

   kubectl label nodes <your-node-name> accelerator=nvidia-tesla-v100
   

   The Predicates plugin will now allow this node to pass the filter, and the pod will be scheduled.