Repolode
Submit
kuberay-project

kuberay-project/kuberay

A Kubernetes operator that deploys and manages Ray applications through RayCluster, RayJob and RayService resources.

View on GitHubWebsite
Stars
2,739
Forks
894
Open issues
718
Last commit
2 days ago

What is it?

KubeRay is an open-source Kubernetes operator for running Ray applications on Kubernetes. It defines three custom resources: RayCluster, which manages the lifecycle of a Ray cluster including creation, deletion, autoscaling and fault tolerance; RayJob, which creates a RayCluster and submits a job once the cluster is ready, with an option to delete the cluster afterward; and RayService, which pairs a RayCluster with a Ray Serve deployment and supports zero-downtime upgrades.

The core component is maintained as part of the project. Optional ecosystem pieces include a kubectl plugin (beta), an API server (alpha) and a dashboard (experimental). The project integrates with Kubernetes tooling such as Prometheus, Grafana, Volcano, Kueue and Apache YuniKorn. It is a CNCF sandbox project licensed under Apache-2.0.

Key features

  • 01Manages RayCluster lifecycle, including autoscaling and fault tolerance.
  • 02Runs Ray jobs through RayJob, optionally deleting the cluster on completion.
  • 03Deploys Ray Serve applications with RayService and zero-downtime upgrades.
  • 04Offers an optional kubectl plugin for common Ray workflows.

Use cases

  • Running distributed training jobs on a shared Kubernetes cluster.
  • Serving machine learning models with Ray Serve on Kubernetes.
  • Running batch inference or data processing jobs that need temporary Ray clusters.

Quick start

  1. 1

    Open the RayCluster quickstart in the Ray documentation

  2. 2

    Follow the RayJob or RayService quickstart in the Ray documentation for those resources

Things to know

  • Requires a Kubernetes cluster, and setup is done through Ray documentation rather than a single install command.
  • The KubeRay APIServer is alpha and the dashboard is experimental and not production-ready.
  • The kubectl plugin is still in beta.

Topics

#ray#kubernetes#operator#distributed-computing#machine-learning#model-serving#artificial-intelligence#deep-learning#orchestration

FAQ

What is kuberay?
A Kubernetes operator that deploys and manages Ray applications through RayCluster, RayJob and RayService resources. KubeRay is an open-source Kubernetes operator for running Ray applications on Kubernetes. It defines three custom resources: RayCluster, which manages the lifecycle of a Ray cluster including creation, deletion, autoscaling and fault tolerance; RayJob, which creates a RayCluster and submits a job once the cluster is ready, with an option to delete the cluster afterward; and RayService, which pairs a RayCluster with a Ray Serve deployment and supports zero-downtime upgrades.
Is kuberay free to use? What license does it have?
kuberay is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is kuberay still maintained?
Actively maintained. The last commit to kuberay was on Oct 20, 2026.
What can I use kuberay for?
Running distributed training jobs on a shared Kubernetes cluster. Serving machine learning models with Ray Serve on Kubernetes. Running batch inference or data processing jobs that need temporary Ray clusters.
How do I install kuberay?
Open the RayCluster quickstart in the Ray documentation Follow the RayJob or RayService quickstart in the Ray documentation for those resources
Where can I get kuberay?
The source code and installation instructions are on GitHub: https://github.com/kuberay-project/kuberay

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects