Meituan Instant Logistics: Distributed System Architecture Evolution and Challenges

Background : Meituan's instant logistics has grown over five years, accumulating experience in building high‑concurrency distributed systems, focusing on extremely low fault tolerance and latency.

Key takeaways : (1) The system must be distributed, scalable, and fault‑tolerant; (2) AI is integrated across pricing, ETA, dispatch, capacity planning, subsidies, accounting, voice interaction, LBS mining, operations, and monitoring to reduce cost, improve experience, and support scale.

Technical challenges include massive order‑rider matching, traffic spikes during holidays or adverse weather, strict availability requirements (no downtime or lost orders), and high sensitivity to data latency and anomalies.

Architecture overview : The platform revolves around three pillars—SLA calculation (ETA, pricing), multi‑objective rider matching, and rider‑side decision support (intelligent voice, route recommendation, store reminders). It relies on Meituan's common components: HLB for traffic load balancing, OCTO for service registration, discovery, load balancing, fault tolerance, and gray releases, message queues (Kafka, RabbitMQ), Zebra for distributed database access, CAT for distributed logging and monitoring, Squirrel+Cellar for caching, and Crane for task scheduling.

Practices for scalability and consistency : Stateful nodes were transformed into stateless ones, enabling parallel computation and rapid horizontal scaling. Consistency across DB and cache is ensured via Databus, a low‑latency, high‑availability change‑data‑capture system that streams binlog changes to downstream stores such as Elasticsearch and KV systems.

Reliability measures : Full‑link stress testing, periodic health checks, chaos engineering (random fault injection), real‑time alerts (performance, business, availability), rapid fault localization (single‑machine, cluster, IDC, component, service), systematic rollback, throttling, circuit breaking, degradation, and a “nuclear‑option” fallback.

Deployment strategies : Single‑IDC rapid deployment and disaster recovery with automatic traffic switching; multi‑center approach using virtual centers composed of multiple IDC partitions for capacity expansion; unit‑based architecture for finer‑grained scaling and fault isolation, with routing based on region or city and synchronized data across sites.

Intelligent logistics platform : A one‑stop machine‑learning platform for model training and deployment, and JARVIS, an AIOps platform that consolidates alarms, filters noise, and automates fault analysis to improve incident response efficiency.

Future challenges : As microservices grow, services become “fat”, leading to network amplification from latency; complex service topologies demand faster fault detection and handling; and the shift from cluster‑level to unit‑level operations poses new deployment and stability challenges.

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 research.