Improving Site Build Efficiency for Didi International Travel: Architecture, Application Modeling, and Automated Delivery Framework

Since 2018, Didi's international travel business has expanded to multiple overseas markets (Brazil, Mexico, etc.). Rapid growth required frequent large‑scale site provisioning across various data‑center and public‑cloud environments (IDC, AWS, GCP). The existing manual process involved thousands of modules, tens of thousands of tables, and required hundreds of engineers from RD, SRE, and QA, taking more than three months per site and causing high risk and missed market windows.

In Q4 2021 the team launched a project to dramatically improve site‑building efficiency. The desired future state includes:

Highly automated delivery that can be performed by a small number of SREs and completed within weeks.

Highly reliable delivery with automated acceptance testing to reduce unexpected production impact.

Platform‑agnostic delivery that works across hybrid‑cloud environments (self‑built IDC and public clouds).

Strong anti‑corrosion capability so that infrequent deliveries (typically yearly) do not degrade the delivery pipeline.

The solution is built around three core ideas:

Application‑centric modeling : Define a standard application description using the Open Application Model (OAM) to capture components, traits, and the full application configuration.

Automated delivery framework : Implement a delivery engine based on KubeVela (an OAM‑native cloud‑native application platform) that translates the OAM description into declarative Kubernetes resources, integrates Terraform for multi‑cloud component provisioning, and supports extensible CUE‑based component definitions.

Environment‑information management : Separate environment‑specific configuration (e.g., MySQL connection strings, region settings) from code, store it in a dedicated configuration repository, and inject it during CI using placeholder substitution, thus eliminating manual configuration changes.

Key technical components include:

Application modeling : Use OAM to describe services, databases, object storage, and configuration bundles as Components, Traits, and an ApplicationConfiguration file.

Delivery platform UI & management center : Provide a front‑end for users to submit delivery requests, manage permissions, and monitor delivery status.

Adaptors : Terraform providers for cloud resources and custom adaptors for internal services.

Stability and anti‑corrosion measures : Automated test suites for configuration changes, metric reporting at each injection point, and static analysis to prevent configuration drift.

The implementation roadmap consists of three phases:

MVP construction & pilot (≈6 months): Build a minimal KubeVela‑based solution for a few components and validate on a small IDC.

Capability expansion & productization (≈1 year): Integrate additional components, embed the solution into the existing DevOps pipeline, and release a platform product.

Full‑scale rollout (≈6 months): Deploy the platform across all international travel services, achieving an 80 % reduction in manual effort for site provisioning.

After rollout, the platform consists of two main products: an Application Center for managing and delivering applications, and an Environment Management service for handling environment‑specific configuration. Real‑world deployments in 2023 demonstrated a dramatic decrease in labor cost and improved reliability.

Future work includes extending support to more infrastructure types, broader resource component coverage, and promoting the platform to other Didi business lines, as well as leveraging the application description for architecture governance and GitOps‑based continuous delivery.