How Meituan Waimai Supports Ten Million Daily Orders: Evolution of Its Backend Architecture

Meituan Waimai, a daily ten‑million‑order business, has undergone several backend architecture iterations to sustain massive traffic. The article, originally from InfoQ, interviews Fang Jianping, Technical Director of Meituan‑Dianping, summarizing the history and future direction of the system.

Stage 1 – Exploration (until early 2015): A small team (~10 engineers) built a simple single‑system web application to quickly launch and validate the delivery model.

Stage 2 – Explosion (starting 2015): Rapid growth required a shift to a distributed service architecture. Key actions included:

Splitting services and data layers (order, user, merchant, logistics, etc.) and creating dedicated micro‑services such as browsing, ordering, and recommendation.

Introducing distributed caches and sharding tables (order, store, user) to handle read/write pressure.

Establishing systematic operations: SLA definition, full‑link stress testing, fault‑injection drills.

Stage 3 – Expansion (later years): New categories (flowers, groceries, supermarkets, errands) demanded a platform‑oriented approach. Common e‑commerce logic was abstracted into a public service platform, with plugins or workflow templates for each category. Projects include:

Set‑based deployment to partition traffic by region and control capacity.

Intelligent operations (AIOps) for automatic anomaly detection, root‑cause analysis, and automated remediation.

High‑Concurrency Strategies: The system employs redundant capacity (1.5× peak), dynamic resource scaling via Meituan’s private cloud containers, self‑protection mechanisms (rate limiting, degradation, circuit breaking), and adaptive traffic scheduling based on real‑time data.

Current Bottlenecks: Horizontal scalability limits at the service‑cluster and storage‑cluster levels (e.g., replica count constraints). Set‑based partitioning is expected to double capacity and support 3‑5 years of growth.

Priority Strategy: Core order‑transaction chain (store browsing, product display, order placement, payment, delivery) is protected first; non‑core services are degraded during incidents. Over 100 degradation switches are managed via an incident response system.

Architecture Optimization: Balancing scaling and performance tuning, leveraging distributed middleware (caches, message queues, service governance), stateless service design, and asynchronous processing.

Splitting Principles: Traffic‑based (C2C vs B2B), synchronous vs asynchronous, functional closure (Conway’s law), and platform services (shared middleware, algorithm platform).

Core Link Identification: The order‑transaction flow (store browse → product select → payment → delivery) defines the critical path; other services are classified as non‑core and can be degraded.

AI Exploration: Meituan Waimai applies AI in OCR for merchant documents, voice‑assistant ordering, intelligent assistants in WeChat, and delivery robots.

Guest Introduction: Fang Jianping has over a decade of experience in large‑scale distributed backend systems, previously at Baidu, now leading Meituan Waimai’s backend architecture.