Kuaishou Big Data Analytics Practices Driven by NoETL

Overview – The talk introduces Kuaishou's big‑data analytics framework and its functional practice under the NoETL paradigm.

1. Kuaishou Data Analytics System – The data platform aims to improve decision‑making efficiency by providing BI tools (KwaiBI) and specialized products, offering both product‑level and service‑level capabilities. Current usage reaches millions of QPS and tens of thousands of weekly active users.

2. Pain Points of Traditional ETL – Long development cycles, low efficiency, high cost, and inflexible delivery hinder rapid analytics, especially when adding new metrics or dimensions.

3. NoETL Insights – NoETL does not eliminate ETL but promotes Smart/Auto ETL, increasing self‑service and automation across the analysis chain.

4. Functional Practices

Standardized metric‑center platform: one‑definition‑multiple‑reuse for metrics and dimensions.

Data standardization: logical tables abstract physical tables, enabling model generation (star/snowflake) and dataset creation.

Analysis‑chain standardization: OAX language translates user intent into dataset‑based logical plans and physical SQL.

Custom fields: flexible creation of new metrics without ETL, supporting simple arithmetic, concatenation, case‑when, and LOD calculations.

Automated modeling: metadata‑driven discovery of fact‑dimension relationships, automatic construction of virtual wide tables (datasets), reducing manual modeling effort.

Automated acceleration: identifies frequently co‑used metric‑dimension pairs, builds accelerated tables, and selects optimal storage engines (ClickHouse, Druid, etc.) to improve query performance.

5. Metric‑Center Deployment – The platform has improved quality, efficiency, and reuse, laying the foundation for NoETL across the analysis pipeline.

6. Typical Scenarios – Flexible metric/dimension addition, virtual wide‑table construction, and accelerated query paths address the three common pain points.

7. Future Plans – Enhance intelligent acceleration, expand advanced analytics functions, and integrate standardized metrics with production systems.

Q&A – Answers clarify that end users do not need to understand star or snowflake models and explain cost‑benefit considerations for automated acceleration.

Overall, the presentation demonstrates how NoETL concepts and a metric‑center platform can streamline big‑data analytics, reduce ETL workload, and boost efficiency at Kuaishou.