KubeRay

KubeRay is the industry-standard toolkit for running distributed Ray applications on Kubernetes via a custom operator and specialized controllers.

KubeRay manages the lifecycle of Ray clusters on Kubernetes by providing three primary Custom Resource Definitions: RayCluster, RayJob, and RayService. It automates complex infrastructure tasks like node autoscaling, fault tolerance, and multi-node networking for high-compute workloads. By integrating with the Kubernetes API, KubeRay allows teams to deploy large-scale machine learning frameworks (like PyTorch or XGBoost) and Python applications across hundreds of worker nodes with a single YAML manifest. It is the core architectural component for production-grade Ray deployments in cloud-native environments.

https://github.com/ray-project/kuberay