Designing a High‑Performance Feeds System for Tencent Channel: Architecture, Strategies, and Optimizations

Tencent Channel is a community platform that aggregates posts, real‑time messages, audio/video, live streams, and bots, making the post subsystem a critical part of its overall architecture.

Compared with QQ groups, Weibo, and Xiaohongshu, the channel operates on a relationship graph of tens of millions of users, resulting in far greater complexity for its Feeds flow.

The system adopts a three‑layer architecture: a logical layer for data assembly and business logic, a proxy (storage‑agent) layer that abstracts storage and provides unified atomic or aggregated interfaces, and a storage layer that indexes and stores posts, comments, and related data.

Three guiding principles shape the design: (1) three‑layer separation to isolate concerns, (2) light‑heavy (core vs. non‑core) separation to avoid interference and reduce deployment cost, and (3) read‑write separation, with read‑heavy services deployed independently from write‑heavy services to improve scalability and consistency.

Feeds flow design focuses on three key list types—sub‑channel post list, post square, and personal feed—each modeled as a mapping problem and implemented using either write‑diffusion or read‑diffusion strategies depending on visibility and performance requirements.

For sub‑channel post lists, a write‑diffusion approach is used, but to support private or under‑review content a hybrid read‑write solution is introduced, preventing empty‑pull issues and ensuring that only publicly visible posts are indexed while private posts are merged at query time.

Post square and personal feed rely on read‑diffusion; to control the potentially huge diffusion volume, pruning techniques such as storing the latest publish timestamps per sub‑channel and caching them in cookies are applied, reducing the number of queried posts per page from thousands to a few hundred.

The article concludes that these design ideas—three‑layer architecture, separation principles, and optimized diffusion—provide a practical blueprint for building robust, high‑performance Feeds systems in large‑scale social applications.