Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Discover how Ray's new Label Selector API revolutionizes scheduling on heterogeneous GPU and accelerator clusters in this 13-minute conference talk from Ray Summit 2025. Learn from Glenn Lockwood of VAST Data as he demonstrates how this robust solution eliminates the need for fragile workarounds like custom resource tags or accelerator_type annotations that often break across environments or cause scheduling failures. Explore the API's key capabilities including support for both static and auto-scaling RayClusters, per-bundle label selectors for precise placement in multi-node workloads, fallback strategies for resource scarcity scenarios, and full integration with the Anyscale platform, Ray Dashboard, and KubeRay. Master practical applications through real-world use cases such as selecting specific GPU types, restricting workloads to nodes with local NVMe storage, matching topology-specific requirements, and isolating latency-sensitive inference tasks. Watch a live demonstration showing how the Label Selector API enhances flexibility, reliability, and developer experience when scheduling Ray workloads on heterogeneous infrastructure, providing identical behavior across all deployment environments from local clusters to production-grade autoscaling fleets.
