Optimization of Xianyu Mobile Message Push System

Xianyu is a second‑hand goods trading platform. When a buyer pays, the platform pushes a notification to the seller; when the seller ships, the buyer receives a push. Message push is crucial for timely transaction updates and user experience.

The push architecture consists of two layers: an underlying platform (ACCS/Agoo) that abstracts OS and vendor channels, and a business‑specific middle platform that implements Xianyu’s push logic.

Because Xianyu’s conversation model attaches multiple transaction sessions to a single user chat, a dedicated push system is required. Although the current system meets basic performance, growth has exposed three main issues: low acceptance rate of messages from the underlying platform, lower delivery rates compared to leading apps, and lack of full‑chain data collection for troubleshooting.

Core dependencies are Alibaba’s ACCS (full‑duplex, low‑latency channel) and Agoo (long‑connection push service). Devices bind to user IDs via the Agoo SDK, and push requests are routed through ACCS or vendor channels based on device status.

Optimization measures include maintaining a real‑time and full‑snapshot device registry to improve query success, upgrading the Agoo SDK for better device reporting, generating a unique messageId for each push to enable end‑to‑end tracing, and instrumenting key flow points for visibility.

After deployment, the acceptance rate rose by double digits, device UV increased as expected, and issue‑tracing time dropped by one‑third, meeting business goals.