Kuaishou Data Service System: Modeling, Architecture, and Future Directions

The presentation introduces Kuaishou's data service system, beginning with a domain modeling approach that abstracts raw data sources (e.g., Hive, MySQL) into structured tables and semantic data models to address development and operational challenges.

It then outlines the evolution of the service platform through three stages—customized services, platform‑based services, and unified services—highlighting improvements in development efficiency, reduction of duplicated effort, and the introduction of three core capabilities: data query, data preparation, and data integration.

The technical architecture is described as "1 engine + 1 platform > 2 services." The Octo query engine provides a unified federation query language (FQL), supports cross‑engine federation, and leverages Apache Arrow for efficient data exchange. The data preparation platform offers multi‑level caching, intelligent cache pre‑heating, and both visual and automated data modeling to streamline table and metric preparation.

Two primary services are built on this foundation: a Data API service that delivers high‑performance, low‑latency table queries with configurable, form‑driven development, and a Data Analysis service that enables headless BI and the Open Analysis Expression (OAX) language for semantic, multi‑dimensional analytics.

Future directions focus on increasing intelligence and performance, including serverless on‑demand scaling and continued enhancements to the Octo engine to achieve sub‑2 ms latency and support massive query volumes.