Insights from Data Platform Experts: Distributed Transactions, Aurora, and HBase

Data Platform & Data Middle Platform Overview

Speaker Zhang Maosen, chief engineer at Didi, distinguished the concepts of a data platform versus a data middle platform, emphasizing that a data middle platform should be "thousand bodies, thousand faces"—a fully digitalized layer that extracts value from data and adapts to organizational change. He highlighted that the decision to build a data middle platform or a business middle platform depends on a company's development stage, citing Alibaba's early focus on a business middle platform.

Distributed Transaction Solutions for Finance

Yan Along, senior R&D engineer at Aikesheng, reviewed common distributed transaction models such as XA, TCC, and Sagas, providing criteria for selecting the appropriate pattern. He introduced the open‑source financial‑grade transaction framework txle, built on ServiceComb Pack, which ensures eventual consistency. Yan compared workflow, exception handling, global transaction error processing, and service degradation scenarios, noting that txle was open‑sourced on October 24.

Deep Dive into Amazon Aurora (MySQL‑compatible)

AWS architect Lv Lin explained Aurora’s design from three angles: performance, availability & durability, and manageability. Aurora follows the philosophy "log is database" and separates storage from compute, a key differentiator in public cloud. The system maintains six replicas, requiring only four successful writes to consider a transaction committed, enabling flexible scaling for both read and write workloads while remaining MySQL‑compatible.

SQL Audit Rules Design and Implementation

Fu Yi, senior database expert at Xinju Network, discussed the challenges of modern database management and the need for automated, intelligent SQL audit. He described the principles behind audit rule design, integration into the work‑order process, and continuous monitoring that enables self‑discovery, self‑diagnosis, and self‑optimization, thereby enforcing SQL development standards before and after deployment.

JD’s HBase‑Based NoSQL Platform

Wu Yiran, head of JD’s HBase platform, shared three main topics: trends in NoSQL, JD’s evolution and use cases of HBase, and practical experiences. He highlighted the massive data volume and high‑concurrency requirements that demand millisecond‑level response times. The platform’s construction addressed multi‑tenant isolation, permission management, active‑active (dual‑data‑center) design, and comprehensive monitoring and alerting.

Event Recap

The salon attracted a full house, with attendees crowding even the seats beyond the main hall. Participants engaged actively during breaks, surrounding the speakers to ask detailed technical questions. The organizers thanked the speakers and participants, and announced future events for continued knowledge sharing.