Evolution and Implementation of Continuous Delivery at Gaode

For engineering teams, building a sustainable, multi‑dimensional quality‑assured delivery system can act as a highway for rapid business value delivery while safeguarding quality throughout the process. This article introduces the evolution and practical rollout of a continuous delivery system at Gaode.

Continuous Delivery Concept

According to Martin Fowler (2013), Continuous Delivery is a software development discipline where software can be released to production at any time. The key criteria include:

Your software is deployable throughout its lifecycle.

Your team prioritizes keeping the software deployable over adding new features.

Anyone can receive fast, automated feedback on production readiness whenever a change is made.

You can perform push‑button deployments of any version to any environment on demand.

Based on these principles, the following focus areas are essential when establishing a continuous delivery system:

Standardized workflow.

Rapid, accurate, automated feedback for changes.

Prioritizing deployment issues over feature development.

One‑click releases.

Gaode’s Continuous Delivery Initiative

The initiative aligns with the group’s vision of a 2‑1‑1 delivery model, emphasizing a 1‑hour packaging‑deployment‑verification window, a 1‑week development cycle, and a 2‑week demand delivery cycle. Milestones include:

2018/08 – Concept and engineering capability building.

2018/12 – Initial rollout and pilot.

2019/04 – Platform integration and expansion to six business lines.

2019/09 – Review and future outlook.

Future – Micro‑innovations and deep‑dive into bottlenecks such as coverage, cloud‑song case platform, and code scanning.

Pre‑Adoption Delivery Process

Before adopting continuous delivery, Gaode’s backend followed a multi‑stage iteration cycle (code change → CI smoke test → feature branch merge → daily environment testing → QA testing → pre‑release testing → gray release → production). Issues identified in this process included:

Accumulated demand causing batch testing and release delays.

Opaque quality standards and delayed feedback.

Manual involvement in deployment and stage transitions.

Manual supervision in multi‑datacenter production verification.

Implementation Measures

Establish a standardized flow aligned with the group’s release process.

Build a backend quality‑standard system to eliminate manual checks.

Enable fast feedback loops and control the release pipeline, making change results visible at any time.

Reduce human participation to achieve fully unattended releases.

Key outcomes include a 50%‑90% efficiency gain in formal releases, coverage of most critical backend services, and the establishment of 12 quality metrics across code, testing, monitoring, and issue‑tracking dimensions.

Future Directions

The roadmap focuses on:

Early “vehicle inspection” – shifting quality checks left to the development stage (code gate, static/security scans, unit tests, code review).

Vehicle performance enhancement – precise regression, scenario coverage, rapid problem localization, synchronized test‑code development, and data‑noise reduction.

Big‑data‑driven insights – leveraging production logs to model real‑world scenarios, improve test case generation, and reduce test data preparation effort.

By continuously refining the delivery pipeline, Gaode aims to achieve faster, higher‑quality business value delivery with lower manual costs, strengthening its competitive edge.