JD Mini Program Data Center: Architecture, Milestones, and Real‑time Analytics Solutions

JD Mini Program provides an open, secure bridge for brand developers to connect with JD's core products, enabling a single codebase to run across multiple apps with cloud‑delivered pages.

The Data Center has evolved through four stages: building foundational data capabilities, expanding metrics, drilling down to user‑level analysis, and achieving intelligent, data‑driven operations.

Its business panorama covers functional analytics (operation and monitoring data), presentation layers (developer console, management backend, mobile assistant, open APIs), and domain‑specific analyses such as user behavior, transaction paths, user profiling, churn monitoring, and traffic monitoring.

Technical architecture addresses three core problems: data ingestion, storage, and analysis. Ingestion uses client‑side (Zhongwu line) and server‑side SDKs; storage selects appropriate sources (real‑time vs. offline) based on latency requirements, employing systems like JED, JimDB, ES, HBase; analysis leverages reusable data models and scalable pipelines.

Real‑time monitoring of mini‑program crashes and performance uses Flink with configurable alarm rules stored in Zookeeper, custom sliding windows via WindowAssigner, and broadcast variables to efficiently distribute configurations across tasks.

For custom analytics, a ClickHouse‑based OLAP engine stores event data using ReplicatedMergeTree tables with daily partitions, enabling fast ad‑hoc queries and supporting both offline and real‑time data flows via DataBus and Flink streaming.

User‑behavior analysis is powered by Elasticsearch, creating daily indices via templates, using index/create operations to manage unique user counts, and leveraging full‑text search and aggregation for metrics like PV, UV, new and cumulative users, and follower counts.

Future directions include consolidating industry‑specific data solutions, reusing the technology stack for other platforms (e.g., RN), and incorporating AI/ML techniques such as time‑series forecasting, collaborative filtering, and large‑language models to enhance predictive alerts and intelligent data operations.

PUT miniapps/_doc/1

PUT miniapps/_create/1