Skip to content

Latest commit

 

History

History
90 lines (57 loc) · 3.23 KB

README.md

File metadata and controls

90 lines (57 loc) · 3.23 KB
Raw

Deploying Ray on Open Data Hub via the KubeRay Operator

The code found here is a subset of https://github.com/ray-project/kuberay. Specifically, the manifests needed to enable KubeRay with an Open Data Hub deployment.

Components of the KubeRay deployment

  1. Custom Resource Definitions
  2. Operator
  3. Roles and Service Accounts

Installation

There are two ways to install KubeRay. You can either install it directly into your cluster alongside an existing Open Data Hub deployment using Kustomize OR you can include it as an additional component in your Open Data Hub KfDef. Below we will walk through both approaches.

Install via Kustomize:

This method requires that you have kustomize and the oc client installed in your environment. Then, from within the ray/ directory of this repo you can run the following:

cd operator/base
oc kustomize > deploy_kuberay.yaml
oc create -f deploy_kuberay.yaml

Install via KfDef:

Alternatively, you can include the KubeRay components in your ODH KfDef. This will ensure that all of the components above get installed as part of your ODH deployment.

 # KubeRay
  - kustomizeConfig:
      repoRef:
        name: manifests-contrib
        path: ray/operator
    name: ray-operator

You can find a minimal KfDef here, that can be used to just install the core ODH components, along with everything you'll need to use KubeRay.

Confirm the operator is running:

Once installed, you can confirm the KubeRay operator has been deployed correctly with oc get pods.

$ oc get pods
NAME                               READY   STATUS    RESTARTS        AGE
kuberay-operator-867bc855b7-2tzxs      1/1     Running   0               4d19h

Create a test cluster

Now that the operator is running, let's create a small Ray cluster and make sure the operator can handle the request correctly. From whatever namespace you want to use you can run the following command:

$ oc apply -f tests/resources/ray/ray-test-cluster-test.yaml

Confirm the cluster is running

$ oc get RayCluster 
NAME                   DESIRED WORKERS   AVAILABLE WORKERS   STATUS   AGE
kuberay-cluster-test   1                 2                   ready    13s

Once the cluster is running you should be able to connect to it to use ray in a python script or jupyter notebook by using ray.init('ray://kuberay-cluster-test-head-svc:10001').

Make sure that the version of ray in your environment matches the version of ray that is running in your cluster. If you used the ray-test-cluster-test.yaml above then you should be using ray==2.1.0.

pip install ray==2.1.0

Then, from your python environment you can connect to the cluster with the following code:

import ray
ray.init('ray://kuberay-cluster-test-head-svc:10001')

Delete your test cluster

Now that you've confirmed everything is working feel free to delete your Ray test cluster. oc delete RayCluster kuberay-cluster-test

That's it! You should now be able to use Ray as part of your Open Data Hub Deployment for distributed and parallel computing.