The Azure Kubernetes Service (AKS) team at Microsoft has shared guidance for running Anyscale's managed Ray service at scale. They focus on three key issues: GPU capacity limits, scattered ML storage, ...