Motan RPC Framework: Architecture, Hybrid‑Cloud Enhancements, and Traffic Management

Motan is a high‑performance, lightweight RPC framework originally created by Sina Weibo engineers to meet the platform's service‑oriented and hybrid‑cloud deployment needs, now serving hundreds of services and billions of daily calls.

The article first reviews RPC concepts and contrasts them with HTTP, RMI, and Web Service, highlighting two main RPC families: service‑governance‑focused (e.g., Dubbo) and cross‑language‑focused (e.g., Thrift, gRPC). Motan belongs to the former, offering rich governance capabilities while remaining simple.

Core Motan features include service discovery, multiple high‑availability strategies (failover, failfast, server isolation), various load‑balancing algorithms (least‑concurrent, consistent‑hash, random, round‑robin), extensibility via SPI, and auxiliary functions such as call statistics and logging.

Motan’s architecture consists of five main modules—Register, Transport, Serialize, Protocol, and Cluster—each extensible via SPI. The Register module interacts with a registry (ZK, Consul, or the internal Vintage KV store) for service publishing, subscription, and heartbeat. Transport uses Netty TCP long connections, Serialize defaults to Hessian2, Protocol manages RPC description and filters, and Cluster aggregates available servers and applies HA/load‑balancing policies.

In a hybrid‑cloud environment, Motan must handle three scaling scenarios: expanding clients, expanding servers, or proportionally scaling both. Cross‑datacenter calls consume dedicated line bandwidth, so Motan introduces mechanisms to control cross‑room traffic ratios.

To reduce bandwidth, Motan compresses RPC payloads by replacing full method signatures with 16‑byte hashes and caching static attachment data. This reduces a simple long‑parameter request from ~280 bytes to ~94 bytes, achieving ~60 % compression in high‑QPS cases. For large responses, optional GZIP (or QuickLZ) compression cuts size by 30‑70 % at the cost of extra CPU.

Dynamic traffic adjustment is achieved through group‑based routing and a JSON‑encoded command system stored in the registry. An example command is wrapped in { "index": 1, "version": "1.0", "dc": "yf", "pattern": "*", "mergeGroups": ["openapi-tc-test-RPC:12", "openapi-yf-test-RPC:1"], "routeRules": [], "remark": "Switch 50% traffic to another datacenter" } . Clients parse these commands to subscribe to multiple groups and distribute calls according to specified weights, with granularity down to 1 %.

Additional optimizations include trimming exception stack traces to only the cause, enhancing registry resilience (supporting ZK, Consul, and a Redis‑based Vintage store), and preparing Motan for open‑source release with reduced external dependencies.

Overall, Motan demonstrates a balance of simplicity, extensibility, and performance, making it a practical RPC solution for large‑scale, hybrid‑cloud microservice architectures.