KubeRay

KubeRay is an open-source Kubernetes operator for deploying and managing Ray applications on Kubernetes. It is hosted under the ray-project organization on GitHub and licensed under Apache-2.0. The core component provides three custom resource definitions: RayCluster, RayJob, and RayService. The latest listed release is v1.7.1.

What It Is

KubeRay lets teams run Ray workloads on Kubernetes using declarative custom resources instead of managing Ray clusters by hand. Its core is described as the official, fully maintained component. Optional ecosystem components add a kubectl plugin, an API server, and a dashboard.

Core Custom Resources

• RayCluster: KubeRay manages the cluster lifecycle, including creation and deletion, autoscaling, and fault tolerance.
• RayJob: KubeRay creates a RayCluster and submits a job once the cluster is ready. It can be configured to delete the cluster after the job finishes.
• RayService: Combines a RayCluster with a Ray Serve deployment graph, and offers zero-downtime upgrades and high availability.

Ecosystem Components and Integrations

The optional components vary in maturity. The kubectl ray plugin is Beta, the KubeRay APIServer is Alpha, and the KubeRay Dashboard is Experimental and not yet production-ready, according to the README. KubeRay integrates with Kubernetes ecosystem tools such as Prometheus, Grafana, py-spy, Volcano, Apache YuniKorn, Kueue, and Nginx.

Documentation and Examples

User-facing documentation and examples are hosted in the Ray documentation. The examples cover training, LLM online inference, and batch inference. The project is a Cloud Native Computing Foundation sandbox project.