Feathr: An Enterprise-Grade, High Performance Feature Store
summary
Feathr is LinkedIn’s feature store running in production for 5+ years, open-sourced April 2022, supporting hundreds of models and thousands of features at petabyte scale. It uses Spark (Databricks/Synapse) for centralized compute, a registry backed by Apache Atlas or SQL Server, and materializes to Redis and Cosmos DB for online serving. Automated pipelines deliver backfilled point-in-time correctness, cut experimentation from weeks to days, and often outperform prior hand‑crafted join jobs. open-source April 2022
AI-written summary · about
chapters (15)
- 0:00Introduction to Feathr
- 1:26What is a feature store
- 3:15Music streaming analogy
- 6:07Problems with manual feature engineering
- 8:12Feathr's core principles
- 9:39Feathr at LinkedIn
- 11:31Feathr timeline and impact
- 13:16Feathr architecture overview
- 15:47Feathr on Azure demo intro
- 16:09Exploring features in Feathr UI
- 17:55Feature lineage and reuse
- 18:36Enterprise access control
- 19:38Consuming features in Python
- 21:43Retrieving offline features
- 22:05Materializing features online
related talks
33mFeathr, an Open Source, Battletested, Scalable and Enterprise-Grade Feature Store
32mBuilding a Cloud Native Feature Store with Feast on Kubeflow
32mA feature Store for Machine Learning
31mFeast as a Cloud Native Data System: Bridging Batch Processing and Online Serving...
27mFeast: feature store for Machine Learning
15mFeature Stores: Powering Machine Learning Infrastructure with Apache Flink
20mNavigating the Skies of Data: A Flight Through Feature Stores
44mFeature store