Design and Implementation of JD.com's Multi‑Active Distributed Architecture

JD.com’s journey from a single Beijing data center to a multi‑region, multi‑active architecture is described, highlighting the business drivers that require high availability, disaster‑recovery capability, and scalable network design as enterprises mature through startup, growth, and mature phases.

The need for geographically distributed active‑active sites stems from limited resources in a single IDC (cabinet, bandwidth, power) and the desire to improve user experience across China’s three major carriers. The solution places services close to users, avoids cross‑network traffic, and leverages BGP, GSLB, and intelligent DNS for optimal routing.

The network backbone adopts a leaf‑spine architecture with three‑layer BGP inter‑connection, eliminating Layer‑2 limitations. Each POD consists of ~200 racks, interconnected via high‑speed edge links, supporting horizontal scaling and Docker‑native automation through JDOS.

Data consistency is achieved by limiting replicas to three copies across at least two centers, using JD’s proprietary high‑speed synchronization product Tube (based on RBR binary‑log capture) and the JIMDB cache for asynchronous, near‑real‑time replication. Critical services rely on these layers to hide disaster‑recovery and latency concerns from developers.

Application deployment follows a closed‑loop model: services are locally deployed with dependencies satisfied, traffic is balanced across the three regional centers, and scheduling uses HTTPDNS and GSLB to direct users to the best‑performing site. The architecture also supports CDN‑like “just‑in‑time” user placement for seamless failover.

Monitoring spans nationwide network quality APIs feeding GSLB decisions, as well as fine‑grained intra‑center health checks on devices, links, and servers. In the event of catastrophic failures (power, cooling, network), an emergency multi‑active plan enables rapid traffic switchover while preserving data integrity through cache‑based consistency.

In summary, JD.com’s multi‑active strategy improves high availability, disaster resilience, and user experience by reducing cross‑center calls, enforcing closed‑loop read/write principles, guaranteeing eventual data consistency, and applying tailored solutions per business need.