Baidu's Continuous Integration Capability Model and Implementation Approach

Continuous integration (CI) is a software development practice where team members frequently integrate their work, typically at least once a day, with each integration validated by automated builds, tests, and releases to quickly detect integration errors and accelerate cohesive software development.

Since 2010, Baidu has gradually applied CI to major product lines such as Net Alliance and Fengchao, achieving good results. With rapid product iteration and market competition, many product lines now seek CI to ensure quality and efficiency. To shorten CI adoption time and avoid redundant effort, Baidu formed a CI team composed mainly of members from product lines with successful CI implementations, aiming to extract best practices, build CI toolchains, and serve all product lines.

The following diagram shows a generic CI layered architecture, dividing CI into pipelines and standards, execution engines, and various services. The article focuses on the conceptual and standard aspects, with future articles covering other dimensions.

To address direction and implementation, Baidu established a CI capability model that provides guidance for product lines and reflects CI characteristics across multiple dimensions.

After defining the direction, Baidu creates CI guidance plans for product lines to follow, enabling reuse of proven solutions and avoiding pitfalls.

Based on the guidance plan, Baidu builds corresponding toolchain platforms tailored to product type and programming language to accelerate CI rollout.

The CI capability model and guidance plan are continuously refined through participation in new product lines, ensuring ongoing improvement.

Finally, Baidu has built a CI talent development system covering four aspects to create a CI ecosystem where talent uses CI technology to support product lines, continuously polishing the CI toolchain, and achieving mutual growth.

Existing CI‑enabled business lines have subjective criteria for evaluating CI success, making it hard to view the company's overall CI capability. New lines lack solid references, leading to potential pitfalls. Therefore, a quantifiable, executable, and improvable CI standard is essential to objectively assess each line, guide new lines, and help existing lines improve.

CI Model Characteristics

Based on accumulated CI experience, CI has the following characteristics:

CI is a development practice.

The goal is to ensure a system that can work at any time.

Small, incremental changes gradually build the system.

Integration occurs at least daily.

Work is performed on the main trunk.

Automated testing is used.

The overall CI goal is to guarantee quality while maximizing efficiency, ensuring the product can be released at any moment. The three core factors are comprehensive test coverage, an efficient build system, and standardized project processes.

The following diagram illustrates the CI capability model for server‑type products.

Key metrics are described below:

Test Coverage : Measured horizontally (local, trunk, daily, RB, pre‑online stages), vertically (different test types per stage), and by effect (coverage rate and test closure).

Build System : Evaluated by build time (≤1 hour), trunk success rate, and abnormal build rate.

Team Habits : Assessed via trunk development (reducing branch work), CI frequency (frequent commits), and MTTR (mean time to recovery for failed tasks).

These three dimensions form the CI evaluation model, visualized in the following diagram.

The algorithm for scoring aggregates module‑level CI data up to product‑line and company levels using an N‑ary tree structure, minimizes manual data entry, and highlights weak spots such as low build success rates or long build times.

CI capability scores are summed across dimensions and mapped to five maturity levels:

Level

Basic Description

Core Points

Score ≥

L1

Entry

Establish basic automation and testing

20

L2

Intermediate

Open automation capabilities and form standards

40

L3

Advanced

Increase automation breadth and depth

60

L4

Proficient

All team members contribute to CI; build system is complete

80

L5

Elite

Product can be released at any time with controlled risk

90

Each level’s requirements are summarized as follows:

L1: Build basic automation for packaging, deployment, and testing.

L2: Extend automation to earlier (code entry) and later (gray release) stages, ensuring stability and efficiency.

L3: Refine automation to meet specialized stage needs, achieving comprehensive and effective automation.

L4: Achieve full‑team participation, with PM, RD, QA, and Ops aligning to continuously improve CI efficiency and stability.

L5: Enable seamless handoff from development to testing to release, achieving risk‑controlled, anytime‑ready deployments.

Conclusion

Since Baidu announced its CI capability model in 2015, all product lines have adopted it, raising the company’s CI score by over 20 points and significantly improving development efficiency. Quarterly assessments (admission, preliminary review, interview, discussion, communication, grading, summary) track each line’s CI capability.

The model’s scores and levels are not end goals but benchmarks to guide continuous improvement; the ultimate aim is to enhance product‑line iteration efficiency.

Author Introduction

Zhu Hualiang

Baidu senior test engineer with years of experience in large‑scale continuous integration solutions, platform development, and implementation, possessing rich theory and practice in the CI field.