Meituan Instant Logistics: Evolution of Distributed System Architecture and Practices

Background

Meituan's instant logistics has grown for over five years, accumulating distributed high‑concurrency system experience. Two main takeaways are the need for ultra‑low fault and latency tolerance and the integration of AI across pricing, ETA, dispatch, capacity planning, subsidies, accounting, voice interaction, LBS mining, operations, and monitoring to boost scale, preserve experience, and reduce cost.

Instant logistics demands near‑zero tolerance for failures and high latency, requiring distributed, scalable, and fault‑tolerant architecture.

AI is deeply embedded in core functions such as pricing, ETA, dispatch, capacity planning, subsidies, accounting, voice interaction, LBS extraction, operations, and monitoring.

Meituan Instant Logistics Architecture

The platform focuses on three aspects: providing SLA (ETA, pricing) to users, matching the most suitable rider under multi‑objective (cost, efficiency, experience) optimization, and offering rider‑side decision‑support (voice, route recommendation, store arrival reminders).

Distributed architecture underpins the platform, ensuring high availability and concurrency. Services are deployed across multiple nodes, communicating via internal RPC (OCTO), message queues (Kafka, RabbitMQ), distributed databases (Zebra), monitoring (CAT), caching (Squirrel+Cellar), and task scheduling (Crane).

Single IDC Rapid Deployment & Disaster Recovery

After a single IDC failure, entrance services detect faults and automatically switch traffic; rapid scaling synchronizes data and pre‑deploys services before opening traffic. All data‑sync and traffic‑distribution services support automatic fault detection and removal, enabling IDC‑level scaling.

Multi‑Center Attempts

Multiple IDC units form a virtual center; services are deployed uniformly across centers. When a center reaches capacity, new IDC resources are added to expand.

Unit‑Based Attempts

Unit‑level routing (by region or city) improves disaster recovery and scaling. Data synchronization across regions may incur latency, and SET disaster recovery ensures seamless failover.

Core AI Logistics Platform

The machine‑learning platform provides end‑to‑end model training and algorithm deployment, addressing challenges of diverse algorithm scenarios, data quality inconsistencies, and low iteration efficiency.

JARVIS, an AIOps platform, stabilizes operations by consolidating noisy alerts, automating fault detection, and accelerating root‑cause analysis for distributed clusters.

Future Challenges

Future hurdles include service bloat as micro‑services proliferate, network amplification from minor latency, complex service topologies hindering rapid fault localization, and the shift from cluster‑level to unit‑level operations after unitization.

Author Bio

Song Bin, senior technical expert at Meituan, leads the instant logistics backend team, focusing on distributed system architecture, high‑concurrency stability, and AIOps.