본문으로 건너뛰기

Launch multinode jobs with Volcano

This tutorial will guide you through the process of launching multinode training jobs with W&B and Volcano on Kubernetes.

Overview​

In this tutorial, you will learn how to use W&B Launch to run multinode jobs on Kubernetes. The steps we will follow are:

  • Ensure that you have a Weights & Biases account and a Kubernetes cluster.
  • Create a launch queue for our volcano jobs.
  • Deploy a Launch agent into our kubernetes cluster.
  • Create a distributed training job.
  • Launch our distributed training.

Prerequisites​

Before you get started, you will need:

  • A Weights & Biases account
  • A Kubernetes cluster

Create a launch queue​

The first step is to create a launch queue. Head to wandb.ai/launch and in the top right corner of your screen, hit the blue Create a queue button. A queue creation drawer will slide out from the right side of your screen. Select an entity, enter a name, and select Kubernetes as the type for your queue.

In the configuration section, we will enter a volcano job template. Any runs launched from this queue will be created using this job specification, so you can modify this configuration as needed to customize your jobs.

This configuration block can accept a Kubernetes job specification, volcano job specification, or any other custom resource definition (CRD) that you are interested in launching. You can make use of macros in the configuration block to dynamically set the contents of this spec.

In this tutorial, we will use a configuration for multinode pytorch training that makes use of volcano's pytorch plugin. You can copy and paste the following config as YAML or JSON:

kind: Job
spec:
  tasks:
    - name: master
      policies:
        - event: TaskCompleted
          action: CompleteJob
      replicas: 1
      template:
        spec:
          containers:
            - name: master
              image: ${image_uri}
              resources:
                limits:
                  nvidia.com/gpu: 1
              imagePullPolicy: IfNotPresent
          restartPolicy: OnFailure
    - name: worker
      replicas: 1
      template:
        spec:
          containers:
            - name: worker
              image: ${image_uri}
              resources:
                limits:
                  nvidia.com/gpu: 1
              workingDir: /home
              imagePullPolicy: IfNotPresent
          nodeSelector:
            cloud.google.com/gke-nodepool: k80-pool
          restartPolicy: OnFailure
  plugins:
    pytorch:
      - --master=master
      - --worker=worker
      - --port=23456
  minAvailable: 1
  schedulerName: volcano
metadata:
  name: wandb-job-${run_id}
  labels:
    wandb_entity: ${entity_name}
    wandb_project: ${project_name}
apiVersion: batch.volcano.sh/v1alpha1

Click the Create queue button at the bottom of the drawer to finish creating your queue.

Deploy your launch agent​

Now that you have created a queue, you will need to deploy a launch agent to pull and execute jobs from the queue. The easiest way to do this is with the launch-agent chart from W&B's official helm-charts repository.

Get an API key for the agent​

If you created a queue in your personal entity, head to wandb.ai/authorize to get a personal API key.

If you created a queue for a W&B, you will need to create a service account in that team or use an API key from a prior service account. For more information on generating service accounts, see these docs.

Keep the API key you plan to deploy your agent with handy for later steps.

Create a config for the agent​

Create a yaml configuration file for the agent you will deploy. Refer here for complete documentation of the agent configuration file. The config should contain (at least):

entity: <your-entity>
queues: [<your-queue>]

Deploy with helm​

Make sure you have helm and kubectl installed and pointing at the cluster where you want to install your agent. To add the wandb/helm-charts repository, run:

helm repo add wandb https://wandb.github.io/helm-charts

Next, run the following command (substituting your own api key and agent config file):

helm install volcano-agent wandb/launch-agent --set agent.apiKey=<your-api-key> --set-file launchConfig=<path-to-your-launch-config>

If the commands above succeed, you should see the status of the queue you created change from Not running to Active.

Create a training job​

Volcano's pytorch plugin will automatically configure the necessary environment variables for pytorch ddp to work, e.g. MASTER_ADDR, RANK, WORLD_SIZE, etc. As long as your write your pytorch code uses DDP correctly, everything else should just work. Refer to pytorch's documentation for more details on how to use DDP in your custom python code.

팁

Volcano's pytorch plugin is also compatible with multinode training via the PyTorch Lightning Trainer.

Launch 🚀​

Now that our queue and cluster are set up, it's time to launch some distributed training! To start off with we will use a job trains a simple multi-layer perceptron on random data using volcano's pytorch plugin. You can find the source code for the job here.

To launch this job, head to the job's page and click the Launch button in the top right corner of the screen. You will be prompted to select a queue to launch the job from.

  1. Set the jobs parameters however you like,
  2. Select the queue you created earlier.
  3. Modify the volcano job in the Resource config section to modify the parameters of your job. For example, you can change the number of workers by changing the replicas field in the worker task.
  4. Click Launch 🚀

You can monitor the progress and if necessary stop your job from the W&B UI.

On this page
Was this page helpful?👍👎