Lean Development Practices and DevOps Implementation at Didi: Coding, Testing, Monitoring, and Ecosystem

Lean production, originally from the Toyota Production System, aims at low‑cost, high‑quality, multi‑variety manufacturing. Its core metric is lead time, which directly impacts product quality, customer satisfaction, and employee happiness. The article explores how these lean principles are applied to software development, especially business‑level R&D.

The author emphasizes that DevOps embodies lean thinking by continuously optimizing the development pipeline. However, many foundational services are under‑utilized, leading to duplicated work and resource waste. To address this, the team analyzes each DevOps stage and builds standards, guidance, and reusable services to support business needs, acting as a “gear” that aligns business and infrastructure teams.

Coding : The first step is establishing coding standards, particularly package management. By adopting SemVer (MAJOR.MINOR.PATCH) and automating dependency upgrades, the team isolates libraries, enables fast bug‑fix releases, and supports gradual rollout of non‑compatible changes. This foundation allows the creation of common components and a company‑wide custom framework called NUWA , which aggregates SDKs and functional components. The framework’s core is an RPC layer that originally supported HTTP, Thrift, and gRPC across Go and PHP services. After evaluating compatibility, maintainability, and future extensibility, the team consolidated to HTTP and gRPC, dropping Thrift.

Testing : Business services sit at the top of the backend stack, making testing costly due to numerous dependencies. The team introduced traffic recording and replay to eliminate the need for mocks and separate test environments, achieving more accurate tests. Additionally, a “sim sidecar” was added to the simulation environment: it extracts the Trace‑ID from full‑link tracing, uses it for routing, and enables per‑developer isolated environments while preserving a shared stable pre‑environment.

Monitoring : Standard monitoring was insufficient for fine‑grained analysis such as GC time, latency breakdowns, and flame graphs. The team extended the company’s monitoring platform with custom metrics and collection tools, allowing detailed health checks, proactive risk mitigation, and faster issue localization.

Ecosystem : By integrating coding standards, testing infrastructure, and monitoring into a unified platform, Didi built the NUWA ecosystem, which propagates best practices across the organization and drives continuous improvement.

The article concludes with personal reflections on the importance of learning‑oriented organizations and invites interested engineers to apply for positions at Didi.