How Mobile Taobao Scaled to 100M DAU: Architecture Evolution and Lessons

Development Stages

Starting in 2009, Mobile Taobao grew from 1 million to over 100 million DAU, prompting four architectural phases:

Phase 1: Early WAP site with HTML templates and a single application for rapid publishing.

Phase 2: Rapid DAU growth across WAP, Android, and iOS required a unified API gateway for multi‑platform business replication and control.

Phase 3: Further DAU increase led to a full HTML5 solution, mixing HTML5 and native components, with an optimized and extensible API gateway.

Phase 4: At 100 M DAU, the API gateway was deployed across many IDC rooms, demanding systematic architecture governance and effective R&D, integration, and monitoring.

API Gateway

Initially there was no API gateway for WAP. As applications multiplied, a unified gateway became necessary to simplify API management and avoid complex, tightly‑coupled RPC dependencies. The gateway also adds security, auditing, logging, and other functions. With distributed IDC deployments for events like Double 11, multiple gateways are used, and a central gateway directs clients to the appropriate regional gateway.

Mobile Side

1. Bundle

In the latter half of the previous year, the team reorganized the architecture. Business bundles—deployable units containing UI, services, and middleware—are managed by a central thread and run inside containers. Each business line (e.g., Juhuasuan) builds its own bundle, which is packaged and deployed together.

2. WebApp

The HTML5 framework consists of a custom runtime container that hosts various WebApps. A unified publishing system performs performance checks, CDN validation, and HTML validation before releasing to CDN. The container receives runtime commands, updates configurations, loads new WebApps, and can intercept URLs for navigation or interaction.

3. PackageApp

PackageApp is the latest construction, built on the previous system. It hides the download process from users by pre‑fetching HTML5/WebApp resources to the client. When a user clicks an icon, the system checks for a newer version; if available, it asynchronously updates from CDN, otherwise it runs the existing version. This approach supports standard URLs, container‑defined specifications, and adaptive updates for different network conditions (full vs. differential, Wi‑Fi vs. cellular).

A load‑time comparison before and after adopting PackageApp shows significant performance improvement.

Support System

1. R&D Support

Comprehensive tools include unified UI libraries, pre‑release environments, and “coloring” clusters for targeted debugging. Code passes through pre‑release, then is deployed to production, with optional coloring clusters to isolate problematic user sessions for deeper analysis.

2. Testing Support

Beyond unit tests, stability, performance, and automation are critical. Automated API regression tests run against the gateway, while client‑side scripts verify that new code does not break existing functionality. Static code analysis also prevents low‑level defects.

3. Operations Support

Mobile‑specific operations monitor performance, stability, business metrics, and public sentiment. Since apps are distributed across many markets, sentiment monitoring aggregates user feedback from app stores and social media, categorizes issues (e.g., payment, detail page, refund), and helps prioritize fixes.

4. Release Support

Release management supports internal and external gray releases, allowing targeted rollouts to specific user groups, regions, or device segments. Critical issues can be fixed via bundle replacement or hot‑patches (class‑loader replacement on Android, experimental solutions on iOS).

Client Monitoring

Minute‑level metrics capture success/failure counts and rates for user actions, enabling real‑time business availability monitoring.

Sentiment Platform

The sentiment platform collects feedback from the app, app stores, and social media, performs keyword clustering, and classifies topics to identify dominant issues per version. This data drives rapid problem resolution and validates the impact of technical improvements.