Implementing Continuous Integration at Baidu: CI Capability Model and Practices

Continuous Integration (CI) is a software development practice where team members frequently integrate their work, often multiple times per day, with each integration validated by automated builds, tests, and deployments to quickly detect integration errors.

Baidu began rolling out CI across major product lines such as Fengchao and Wangmeng in 2010, achieving good results. As the company expanded, more product lines sought CI to ensure quality and speed, leading to the creation of a dedicated CI team that extracts best practices, builds a CI toolchain, and serves all product lines.

The CI team’s approach includes four steps: (1) establishing a CI capability model to provide direction and implementation guidance; (2) creating CI guidance documents for product lines to follow; (3) having team members participate in new product line CI roll‑outs to refine the model; and (4) building supporting toolchains tailored to product types and languages.

A CI talent development ecosystem was also built, focusing on four aspects to create a CI ecosystem where talent uses CI technology, refines tools, and drives mutual improvement of product lines and Baidu’s overall CI capability.

The CI capability model is based on three core factors: comprehensive test coverage, an efficient build system, and standardized project processes. Test coverage includes local, trunk, daily, and pre‑online testing stages, each with specific test types and a measurable coverage metric. The build system is evaluated by compile time, overall build time, trunk success rate, abnormal build rate, and daily success rate. Team habits are measured by trunk‑first development, CI frequency, and mean time to recovery (MTTR).

All metrics are quantifiable, and a CI capability score is calculated by summing dimension scores. The model is graded into levels to guide product lines, and Baidu conducts a seven‑stage quarterly assessment (admission, preliminary evaluation, interview, discussion, communication, rating, summary) for each product line.

Since the CI capability model’s release in 2015, all product lines have improved their CI scores by over 20 points, significantly boosting R&D efficiency. The scores and grades serve as a benchmark rather than an end goal; the ultimate aim is continuous improvement of product line iteration efficiency.

The article concludes by noting that future posts will cover detailed CI implementation solutions.