Evolution and Refactoring of Autohome Mobile Backend Architecture

In the era of mobile internet, Autohome faced explosive growth in traffic and functional complexity, prompting a shift from a simple PC‑centric monolithic architecture to a robust mobile backend capable of handling high reliability, scalability, and low cost.

2010‑2012: The initial mobile team built a basic app that directly mirrored PC forum content, using an ALL‑IN‑ONE server without clear architectural layers, which was acceptable due to low traffic.

2013‑2014: After the company's IPO, mobile traffic surged (2.5× increase by end of 2014), exposing the limitations of the monolithic design, frequent crashes, and high maintenance costs. The first refactor introduced C# and SQL Server, separating the code into API, Service, and DAO layers, adding CDN acceleration, read‑write separation, and clustering.

2015‑Present: New challenges included massive request volume, numerous dependencies (Redis, DB, RPC, HTTP, MongoDB, MQ), vertical business coupling, frequent promotional activities, bi‑weekly releases, and reliance on costly Microsoft technologies. To address these, the team migrated to Linux, MySQL, Tomcat, and Java, and performed a comprehensive redesign.

Refactoring Strategy: The redesign emphasized team and service decomposition, horizontal scaling (multi‑cluster, multi‑data‑center), vertical splitting of business modules, business sharding, stateless services, reusable abstract business services, layered resource design, loose coupling with self‑protection, circuit‑breaker, rate‑limiting, and isolated deployment. Core services were kept lightweight for stability, while auxiliary services used asynchronous processing.

Implementation Highlights:

Adopted a single‑service structure illustrated in Figure 3.

Introduced a Java‑based RPC framework Trubo (multi‑language support, ZooKeeper registration, client load‑balancing, fault tolerance, Spark‑based monitoring).

Developed an asynchronous request component AIS4J to decouple external resources and integrate circuit‑breaker logic.

Built a configuration service on ZooKeeper for dynamic feature toggles, along with monitoring, tracing, and logging services.

Implemented distributed tracing (Trace) with unique Trace ID and Transaction ID, visualized via Spark dashboards (Figures 6‑8) and service call chain diagrams (Figure 9).

Enabled a debug mode exposing request‑level timing and resource chain information (Figure 10).

Established real‑time alarm mechanisms that notify via SMS and email, linking alerts to Trace IDs for rapid issue localization.

The comprehensive architectural overhaul enabled Autohome to sustain rapid traffic growth, improve system resilience, and support continuous delivery for its mobile platform.