Design Practices for Large-Scale Microservice Frameworks

Du Huan, a senior expert engineer at Didi, shares his experience of rebuilding the company’s backend services with Go and designing a large‑scale microservice framework, presented at the Gopher China conference.

He first outlines the typical pain points of complex service development: tight schedules, large teams, rapid business growth, uneven developer skill levels, and the difficulty of integrating numerous internal tools without excessive cost.

He then reviews the historical evolution of service frameworks—from early PHP and ASP.NET, through the MVC boom of Django, Ruby on Rails, and the explosion of Go‑based RPC frameworks—highlighting how frameworks have gradually become more like operating systems that abstract away low‑level details.

Guided by the "Rule of Least Power" principle, the framework focuses on exposing only the essential, stable abstractions, avoiding over‑engineering while keeping the design extensible where truly needed.

The design goals include providing a unified interface layer, transparent I/O handling, automatic service governance, and a composable architecture that isolates business code from underlying drivers, allowing seamless upgrades of underlying infrastructure.

Implementation details cover a toolchain that generates code from IDL, runtime reflection to build routing, protocol hijacking (e.g., Thrift) to inject context and logging, a low‑precision timer pool to avoid timer/channel leaks, and fail‑fast mechanisms that propagate upstream timeouts to downstream services.

Business benefits reported are dramatically higher development speed, reduced operational incidents, transparent framework upgrades, and the ability to scale to thousands of services and millions of lines of code without sacrificing reliability.

Version management is treated like an operating system: the framework itself never changes its public interfaces, so all services always run the latest version without compatibility concerns.

The article concludes with future work (enhanced toolchains, deeper infrastructure integration) and a brief Q&A where the speaker discusses Go process model, service dependency management via IDL, timeout handling, and real‑world failure scenarios.