Alibaba's Continuous Delivery Model and Practices for High‑Quality Software Delivery

Alibaba, with over 30,000 employees, millions of online applications and billions of users, updates hundreds of applications daily; the continuous delivery system safeguards this high‑speed, high‑risk environment.

In modern development enterprises, especially internet companies, effective continuous delivery is crucial. Alibaba, based on years of R&D management experience, has proposed a new model and practice for continuous delivery.

Product expert Dai Meng from Alibaba Technology Assurance Department shared Alibaba Cloud's continuous delivery experience at the 2016 Cloud Expo Shanghai Summit, analyzing Alibaba's continuous delivery theory and how tools enhance R&D management.

The talk outlines why continuous delivery is needed, especially during massive traffic events like Double‑11, where updates must be seamless and immediate. Two key factors affecting release quality are identified: "no failure" and "failure" scenarios.

When no failure occurs, activities include code review, testing, gray‑release, and deployment. When failures happen, steps involve incident localization, rollback, fault analysis, and fixing—each integrated into the continuous delivery process.

Continuous delivery comprises integration, continuous processes, deployment, and delivery. Integration means merging code and logic continuously; deployment ensures integration independence; multiple integrations lead to a delivery.

Habit formation is essential for continuous delivery. A 1967 experiment showed collective habits can be established quickly, and with proper tools and planning, teams can cultivate the necessary delivery habits.

Two delivery pipelines are illustrated. The "bad" pipeline shows isolated code commit, compilation, unit testing, and separate deployment and integration testing, leading to feedback delays. The "good" pipeline merges code checkout with compilation and unit testing, combines deployment with integration testing, and adds smoke testing stages before production.

Images depict the simple bad pipeline (Fig 1), the improved pipeline framework (Fig 2), and the data theory of the delivery line (Fig 3), highlighting atomic activities, expanding coverage, increasing cost, and decreasing frequency.

Continuous delivery tools encompass project management, code hosting, build management, and release automation, forming a complete delivery line.

Agile principles (individuals & interactions over processes & tools, working software over documentation, customer collaboration over contract negotiation, responding to change over following a plan) guide tool selection.

The platform provides dynamic documentation, unified story cards for agile development, and encourages role‑agnostic collaboration within a single document.

Effective communication—face‑to‑face, phone, instant messaging, and email—is integrated into the continuous delivery platform, enhancing development efficiency.

Alibaba offers a comprehensive code management service with features such as integrated compile/test/release plugins, full Git command support, private repositories, and online code browsing, all highly available with unlimited storage.

Additional services include high‑speed Maven repository mirroring, component management, and backup mechanisms (1:9 for code, 1:3 for components).

To build a resilient (anti‑fragile) system, the continuous delivery platform (CRP – Continuous Release Platform) provides end‑to‑end lifecycle services: project, requirement, defect, code, component, environment, delivery line, dependency, test, one‑click deployment, monitoring, and team collaboration.

Images of the platform architecture and component management diagrams are included throughout the article.

Overall, Alibaba's continuous delivery ecosystem enables rapid, safe, and efficient software releases in large‑scale cloud environments.