How KubeRay Is Evolving for Massive AI Workloads

Explore the latest advancements in KubeRay designed to dramatically improve the user experience for teams running Ray on Kubernetes in this 24-minute conference talk from Ray Summit 2025. Learn from Aaron Liang from Google and Jui-An Huang from Anyscale as they introduce a series of major RayJob enhancements that simplify and harden the end-to-end workload lifecycle. Discover deletion policies for predictable and controllable cleanup, cron scheduling to support recurring and automated Ray workloads, sidecar mode for easier integration with surrounding services and tooling, and background status checks that improve reliability and reduce operational overhead. Understand how these upgrades make Ray job orchestration on Kubernetes more intuitive, manageable, and robust for both developers and operators. Gain practical insights into how these new features streamline Ray workloads in production and see how KubeRay is evolving to make large-scale distributed computing easier than ever for massive AI workloads.