WeChat Backend Architecture: High Availability, Strong Consistency, and Scalable Microservices

WeChat's backend supports a wide range of services such as instant messaging, social networking, and financial payments, serving over one billion active users and handling billions of requests per day.

The architecture has evolved from a first‑generation storage system in 2011 to a custom, highly available storage and computation framework called PaxosStore, which separates consensus, compute, and storage layers and supports both multi‑master and traditional failover designs.

Four major challenges drive the design: massive storage with fault tolerance, strong consistency for ten‑billion‑user data, handling traffic spikes (e.g., holidays and viral events), and supporting over one hundred billion data accesses per minute.

To meet these goals, WeChat targets five‑nine availability for financial services and six‑nine overall availability, employing multi‑master replication, cross‑data‑center load balancing, and rapid failover avoidance.

Key technical components include:

Multi‑master PaxosStore providing high‑throughput, ACID‑compatible storage with support for various data models (key‑value, FIFO, 2‑D tables).

A microservice framework offering service definition, discovery, error retry, monitoring, gray‑release, and configuration management.

Libco, a coroutine library built on epoll and a time‑wheel, enabling synchronous‑style programming with asynchronous execution, reducing complexity compared to traditional event‑driven models.

Practical outcomes show the core data storage achieving six‑nine availability, with the design and algorithms published in a VLDB paper and the PaxosStore code open‑sourced on GitHub.

Overall, the system demonstrates how a large‑scale, high‑availability backend can be built using custom consensus protocols, multi‑master replication, and coroutine‑based concurrency to support continuous, low‑latency service for billions of users.