Feature store
summary
Search handles ~150 million products with ~10k RPS typical and 35k peaks, pulling up to 2000 candidates per query into ranking. Redis hotspotting with up to 18 to 3 replicas motivates switching to sharded standalone Redis, adding Kafka for category moves and a cold Postgres shard set for slower consistency. A feature-revision AB system stores experimental feature sets in separate shards, cutting memory and CPU use and enabling live experiments while keeping latency and availability targets.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 6:25Old architecture problems
- 12:21New storage requirements
- 16:37Redis Standalone solution
- 21:13Experiments and A/B testing
- 24:49Results and summary
- 31:56Q&A
related talks
19mML feature store a cornerstone in modern data architecture for AI/ML
25mBuilding A Feature Store For Hyper Growth
33mFeathr, an Open Source, Battletested, Scalable and Enterprise-Grade Feature Store
15mFeature Stores: Powering Machine Learning Infrastructure with Apache Flink
08mFeature Store Usage Patterns - From a Single Data Scientist to an Enterprise
33mEvolution and Scaling of Feature Store at Uber
29mIntegrating High Performance Feature Stores with KServe Model Serving
23mFeathr: An Enterprise-Grade, High Performance Feature Store