Tellus: A User‑and‑Item Task System for Xianyu’s Light Publishing

Business background: Xianyu, a C2C app, allows sellers to publish items quickly with minimal information, leading to insufficient product details and unreasonable pricing, which reduces conversion.

Solution idea: Introduce a task‑driven approach that guides sellers to complete missing item information or adjust prices while preserving the lightweight publishing flow.

Technical implementation: Developed Tellus, a task system built on both user and item dimensions. Existing internal task systems were user‑only, so a custom solution was required.

System architecture: The lifecycle includes task creation, display, and completion. Task creation decouples data sources via MetaQ (Alibaba’s RocketMQ‑based message queue) to handle diverse event streams. Task display supports single and composite tasks, with AB testing for various task types. Completion logic combines active (success/failure) and passive triggers (e.g., editing an item adds a color attribute, automatically completing the related task).

Technology choices: For massive task data (hundreds of billions of records) Table Store (Alibaba Cloud NoSQL) was selected for its PB‑scale storage, high TPS, low latency, strong consistency, and managed service benefits. Metadata, being low‑volume, is cached with Guava Cache.

Business impact: Tellus’s configurability enabled rapid rollout of features such as “Super Exposure” and “Inventory Diagnosis,” doubling task completion rates. Template‑based and configurable task designs reduced development effort and allowed non‑engineers to launch new tasks via configuration.

Future plans: Integration with Omega (real‑time event collection) and Hermes (push‑based user engagement) to generate and deliver tasks automatically, plus incentive mechanisms using Xianyu coins.