An Overview of CloudKitchens's Ray-Powered ML Platform
summary
City Storage Systems moves its ML stack onto Kubernetes and Ray, replacing legacy Trino/Flink pieces and proving Ray can host ETL, training, and serving with fewer moving parts. Engineers build a Ray control plane plus a proxy for multi‑team, multi‑region cluster isolation and autoscaling, add a Metaflow Ray step and a Poetry plugin to submit user Python as Ray jobs, and adopt Daft over Ray Data for joins and performance. The system runs on CubeRay, works at scale today, and some components remain internal.
AI-written summary · about
chapters (6)
- 0:00Introduction and background
- 10:23Platform vision and principles
- 15:01Ray infrastructure overview
- 18:39Data preparation with Ray
- 21:14Integrating tools with Metaflow
- 24:46Q&A session
related talks
26mIntroduction to Distributed ML Workloads with Ray on Kubernetes
44mPanel: ML Platform on Ray
22mRay, a Unified Distributed Framework for the Modern AI Stack
28mRay on Public-Cloud k8s - experiments, lessons & best practices
28mScaling AI & Machine Learning Workloads With Ray on AWS, Kubernetes, & BERT
31mNext-Generation Data Science Workflows Using Ray
35mOperating a Production-Scale Physical AI Data Platform with Ray
19mScaling Frontier AI with Ray on Kubernetes