How Xianyu Scaled from Zero to Millions: Architecture Secrets of Alibaba’s Community Marketplace

Recently at ArchSummit Beijing, Alibaba senior wireless technology expert Sun Bing (nickname “Jiu Gai”) delivered a talk titled “Grid Community – Xianyu Technical Architecture”. Sun joined Alibaba in 2011, worked on B2B, Taobao, Mobile Taobao, and now leads the technology for the innovative business Xianyu, sharing the evolution of its community architecture.

Sun explains that an architect’s work must be guided by clear purpose and vision; without a compelling reason, architectural choices become mere technical selections. He describes Xianyu’s mission as building a P2P trading community, emphasizing that only those who truly believe in the mission stayed through early hardships.

He recounts the project’s early days in 2014, when Alibaba focused on large‑scale apps like Mobile Taobao and Alipay. Xianyu started with a ten‑person team, no testing, and reused components from Taobao. The initial architecture was simple, built quickly to prove the concept.

During the first stage, speed was prioritized over quality. Rapid feature delivery was essential, but Apple’s App Store review process caused delays. To address dynamic content needs, Xianyu adopted a HyBird architecture using H5 as a fallback via a URLRoute mechanism, allowing the client to route requests to H5 pages when native implementations were unavailable.

A switch center was introduced to enable flexible configuration, version control, and A/B testing without requiring client updates. APIs were versioned, and a gateway ensured backward compatibility across multiple API versions.

For dynamic UI components, Xianyu built a Poplayer system to inject promotional assets (e.g., seasonal banners) into pages without hard‑coding resource slots, facilitating close front‑end/back‑end collaboration.

As the product grew, the focus shifted from pure speed to efficiency and capability. The team emphasized communication‑free collaboration, automated contracts, and a MBaaS layer that allowed the client to perform one‑off operations (such as likes or lottery participation) without adding new server APIs.

The architecture evolved toward a Service Bus that abstracts service calls behind an SDK, providing a unified parser for rendering diverse data types on the client. Data flows from client logs to an algorithm platform, feeding real‑time and batch models that drive search and recommendation.

Offline‑online integration remains challenging; Xianyu implemented identity verification and device checks to link offline user actions with online profiles.

Data‑driven development is central: product KPIs align with user‑scale goals, and multi‑objective optimization guides resource allocation. Experiments showed that repeated daily visits are more valuable for retention than a single transaction.

In the Q&A, Sun clarified SDK version compatibility, explaining that a single version identifier governs API compatibility, and described the MBaaS approach for dynamic schema creation and personalized data handling.