Data Governance and Active Metadata Practices at JD Retail

JD Retail faces multiple data management challenges: growing data volume creates inefficient and redundant models, weak asset awareness makes tables hard to find and use, shared accounts cause change‑management risks, and expanding table counts increase compute and storage consumption.

To address these issues, JD Retail proposes a holistic data governance solution comprising four pillars: establishing data standards, upgrading data architecture for agility, isolating development and production environments for safety, and implementing storage and compute governance to reduce resource waste.

The governance framework includes standard governance (defining a unified data language, asset certification, and systematic metadata collection), architecture governance (logical virtual tables, intelligent materialization using HBO/CBO/RBO models, and lake‑warehouse integration), development governance (account, table, and queue isolation), and resource governance (lifecycle management, invalid table/task identification, and optimization of compute tasks).

Active metadata is central to JD Retail's practice. By continuously collecting and analyzing runtime metadata, the system can automatically generate recommendations for data lifecycle values, identify storage optimization opportunities, and automate back‑fill processes using production lineage information.

Intelligent lifecycle evaluation models quantify storage and compute costs, incorporate non‑quantitative factors (e.g., data tier, certification), and provide visual dashboards for self‑service analysis, achieving over 70% acceptance and saving billions of yuan annually.

Future directions focus on further automation, AI‑driven task optimization, semantic asset recognition, and extending active‑metadata‑based governance across the organization.