Coscheduling

This guide describes how to enable gang scheduling with the Coscheduling plugin in Kubeflow Trainer.

The Coscheduling plugin ensures that a group of Pods in the same training job start together only when all required resources are available,

Prerequisites

You have to install and enable the Coscheduling plugin in your Kubernetes cluster before enabling the Coscheduling gang scheduling policy.

Enable Gang Scheduling with Coscheduling

The Coscheduling policy configures gang scheduling with Coscheduling plugin

podGroupPolicy:
  coscheduling:
    scheduleTimeoutSeconds: 30