Inside Alibaba’s City‑Level Active‑Active Architecture: A Complete Guide

City Active‑Active Overview

City active‑active deploys two independent data centers within the same city, each running a full set of services that are simultaneously active and receive traffic. When either data center fails, traffic is automatically switched to the other, providing uninterrupted service.

Architecture Layers

Traffic Layer

The traffic entry layer uses a global load balancer to distribute user requests to data center A or B based on IP geolocation or round‑robin algorithms. An application gateway at each data‑center entrance parses a route key (e.g., user_id) and determines the target cell. If a request lands in the wrong data center—such as an odd user_id that should be handled by A but arrives at B due to DNS drift—the gateway performs an internal cross‑data‑center redirect to the correct location.

Application Layer

Services are designed to be stateless. Typical practices include:

Stateless Web/API services

Centralized session storage

Unified task‑scheduling center

Data Layer

The data layer is the key to a successful active‑active deployment. Common patterns are:

Master‑slave replication with automatic failover (a basic one‑master‑one‑slave setup, not true active‑active)

Dual‑master architecture where both centers can write, requiring conflict resolution and strong consistency mechanisms

Sharding by business domain or user segment, placing different business lines or user groups in separate centers to reduce cross‑center write conflicts

Typical request flow:

User request
│
┌────────┴────────┐
│                │
│   IDC‑A        IDC‑B
│   (Active)      (Active)
│                │
│   Application Cluster
│   Application Cluster
│                │
│   Redis Cluster   Redis Cluster
│   MySQL Cluster   MySQL Cluster
│                │
│   │            │
│   │            │
│   └────────────┘

By separating responsibilities across the traffic, application, and data layers and applying the appropriate data‑layer pattern, continuous availability and seamless failover are achieved for services running in the same city.