Scaling WeChat Moments: Architecture, Capacity Planning, and Flexible Strategies for High Traffic

1. Introduction

WeChat Moments consists of two business architectures: image and video. Image traffic is massive and CPU‑intensive, while video mainly consumes bandwidth. Data is stored permanently, and rapid business growth leads to increasing storage, bandwidth, and device consumption, especially during holidays, putting pressure on operations.

2. Related Articles

Links to PPT and video presentations about the massive technology behind WeChat Moments.

3. Software Architecture Guarantees

Overall architecture diagram (image). The system is divided into OC (outside‑network data centers) and IDC (internal data centers). IDC hosts storage; OC provides external access and caching. Users download from the nearest OC; if a cache miss occurs, the request is routed to IDC.

4. Disaster Recovery and Retry Mechanism

Automatic removal of faulty machines is achieved by a master server managing IP lists and heartbeat detection. Example of front‑end single‑machine removal is shown (image). When an OC or IDC fails, manual switch or retry mechanisms are used.

Download retry: after two failures, the client retries with a different, geographically distant IP list, ensuring cross‑region retry. During peak holidays, retries may be disabled or manually turned off if IDC failure rate exceeds 20%.

Front‑end retry control interface (image).

5. Hardware Guarantees

5.1 Capacity Assessment and Expansion

Before major holidays, capacity is evaluated and devices are expanded based on bandwidth, CPU, memory, and disk I/O metrics.

5.2 Spring Festival Upload Load

Upload traffic is expected to increase 9×, download 1×. Excess requests are rejected; some modules, especially the compress module, cannot scale without additional VMs, so flexible strategies are applied.

6. Flexible Strategies Overview

Two layers: coarse‑grained (rate limiting) and business‑specific (reducing image/video quality, delaying updates).

7. Flexible Practice: Compress Module

Switching from HEVC to JPEG reduces CPU load by 80% (to 20% of original), supporting 5× growth, but increases average file size. A compromise reduces quality from 70 to 50 while using JPEG, keeping user perception unchanged.

8. Flexible Practice: Short Video Bitrate

Bitrate reduced from 1800 kbps to 1200 kbps, cutting average size from 2.1 MB to 1.3 MB, with negligible user impact; changes take about four hours to propagate.

9. Flexible Practice: TSSD Buffer Pools

Two buffer pools are added to absorb burst upload requests; one buffers overflow for zone module, the other protects the pre‑upload module and TFS storage.

10. Flexible Practice: Timeline Proportion

Timeline updates are cached and not pushed to users, reducing download requests. Risks include user complaints and potential traffic spikes if caching duration is too long.

11. Spring Festival Manual Flexibility Steps

Operational steps illustrated (image).

End of article with invitation to share and join the architecture community.