Full-Stack Observability in Ray: Infrastructure Signals in Ray Dashboard

on YouTube ↗15m

" summary

Describes Ray’s platform events integration that ingests Kubernetes events (for RayCluster, RayJob, RayService custom resources plus pod evictions, spot preemptions, and OOM kills) and surfaces them in the Ray Dashboard by correlating ephemeral pod lifecycles with stable Ray worker identities to enable immediate root-cause analysis without running kubectl.

AI-written summary · about

" related talks

• 22mDebugging and Observability for Distributed Ray ApplicationsSangBin Cho · Ray Summit 2020
• 26mIntroduction to Distributed ML Workloads with Ray on KubernetesAbdel Sghiouar · AI_dev: Open Source GenAI & ML Summit Europe 2024
• 39mSailing Ray Workloads with KubeRay and Kueue in KubernetesKante Yin, Jason Hu · KubeCon + CloudNativeCon + Open Source Summit China 2023
• 31mNext-Generation Data Science Workflows Using RayErik Erlandson · Kubernetes AI Day North America 2022
• 32mAn Overview of CloudKitchens's Ray-Powered ML PlatformSantosh Jha, Ammar Alrashed · Ray Summit 2024
• 28mRay on Public-Cloud k8s - experiments, lessons & best practicesMadhuri Yechuri, Anne Marie Holler, Chi Su · Cloud Native Rejekts NA 2021
• 24mSkyRay: Seamlessly Extending KubeRay to Multi-Cluster Multi-Cloud OperationAnne Marie Holler · Cloud Native + Kubernetes AI Day NA 2024
• 22mHidden Signals in K8s ClustersGary Edgar · SREday Redmond 2025 Q2