Architecture Evolution and Challenges of Meituan's Code Hosting Platform

Meituan's self‑developed Code platform manages millions of Git repositories and handles tens of millions of daily Git requests.

1. Evolution stages

Single‑machine deployment: simple web‑app serving all requests, limited by disk I/O.

Multi‑machine deployment: read‑write separation with a master node for writes and slave nodes for reads, round‑robin routing, and agent‑based sync, doubling read throughput.

Distributed architecture: application‑level sharding, three‑replica active‑active mode, gRPC communication, and logical‑physical separation via a routing service.

2. Key challenges

Scalability – storage saturation (80% SSD usage) and CPU/IO load >95% during peaks.

Availability – need for zero‑downtime upgrades, reliable backup, and cross‑region resilience.

3. Solution components

Proxy modules (SSH Proxy, HTTP Proxy) built with Go‑Crypto and Gin for request filtering, traffic control and seamless migration.

Routing (Shard) service maintaining repository‑to‑shard mapping, fast (<15 ms) gRPC lookups, read‑repair and strict arbitration (N=3, R=W=2).

Application services providing Git APIs, event hooks, and stateless scaling.

Storage layer based on GitLab components, with Replication Manager, Code Core, Git Core and Git Command Factory.

Active‑active three‑node Git clusters with multi‑region deployment, hot‑standby replication, and log compression (e.g., main -> a … compressed to main -> f ).

4. Operational practices

Data inspection, automatic hot‑backup, version‑based routing, and gradual migration tools ensure zero‑impact user experience.

5. Results and outlook

The platform now supports horizontal storage expansion, automatic load balancing, and high availability; future work will focus on automated operations, best‑practice tooling, and tighter code security integration.