Scaling a Billion-User Online Banking Platform with DevOps

The Agricultural Bank of China (ABC) personal online banking system serves millions of individual customers with a comprehensive suite of financial services, including accounts, transfers, credit cards, loans, investments, and more, all built on a .NET stack and supported by continuous integration and deployment pipelines.

Configuration Management

All configuration files, build and deployment scripts, database change scripts, and environment settings are stored in a Git repository with strict naming and commit conventions. Using TFS, tags and branch operations are automated, and a permission model, metrics, monitoring, backup, and availability safeguards are in place.

The project follows an agile two‑week sprint cycle, treating each feature as the smallest deployable unit. Feature branches are created per functional layer and completed within three days.

An "aoneflow" branching model merges feature branches frequently into the main line, minimizing conflicts and keeping merge times within hours. Build artifacts, CDN packages, configuration packages, and SQL packages are managed in Artifactory, serving as a single source of truth for all environments.

All code and configuration changes originate from TFS work items, undergo automated and manual reviews—including pre‑build checks, code compliance scans, security scans, and unit‑test gates—and are linked automatically between TFS and the change‑management system for bidirectional traceability.

Build and Continuous Integration

TFS orchestrates pipeline tasks at an atomic level, providing visual build workflows that can be customized per scenario. Self‑service builds are enabled, and shared libraries are used across environments, with quality gates and parallel execution support.

Environment build scripts standardize configurations and allocate dedicated build resources. Pull‑request (PR) triggers automatically run builds, unit tests, and code scans; successful PR merges launch main builds that publish artifacts to Artifactory. Build failures are emailed to the PR author, and solutions are documented in a shared wiki for rapid resolution.

Test Management

A test pyramid tailored to the personal banking system increases the proportion of API automation, accumulating over 3,000 automated test cases. Test activities shift left, with testers collaborating on user stories during requirements gathering and conducting API tests during development.

Test data is decoupled from test cases to ensure independence. An organization‑wide automation platform links test cases to requirements, selects appropriate automated tests, and automatically associates defects.

Quality standards enforce security and compliance checks on every PR. Using Parasoft, developers receive detailed, line‑level code quality reports both locally and in the pipeline, enabling continuous remediation of technical debt.

Deployment and Release Management

Deployments target Windows servers. In test environments, TFS handles deployment verification; the pre‑production environment uses the "Star Cloud" platform to validate deployment strategies and versions. After validation, production (including gray‑release) deployments are performed with a one‑click automated process on Star Cloud.

Weekly releases are the norm. Application packages are separated from configuration packages, allowing independent deployment of each. Dedicated pipelines manage application and database changes, and all artifacts are stored in Artifactory.

The entire delivery flow is visualized in the pipeline, with release notes, test reports, and automated test links embedded in artifact metadata. Approvals are required before production deployment, and rollback is a single click.

Environment Management

Virtual machines provide the foundational infrastructure. ABC has built complete development, test, pre‑release, gray, and production environments, and a self‑developed automation tool installs and configures environments with a single command, ensuring consistent tool versions.

The automation tool also provisions test environments on demand, codifies environment and dependency configurations in Git, and stores service‑level dependency files alongside application code.

Data Management

Production data is exported, sanitized, and used to create baseline test data sets. Unit tests that modify databases automatically restore data after execution. Test data is isolated per test case, preventing cross‑test contamination.

Data change scripts (DDL/DML) are organized in separate Git directories with branch strategies and pipeline stages for multi‑phase validation. Scripts are linked to application artifacts, and change reviews enforce execution and rollback scripts.

The Star Cloud platform orchestrates data changes before application changes, ensuring that configuration data is applied only after the new application version is live, supporting zero‑downtime deployments and rollbacks.

Metrics and Feedback

Metrics are collected across seven dimensions—requirements, coding, building, testing, defects, environment, and deployment—totaling 43 indicators. A unified measurement platform generates automated reports, with drill‑down capabilities to raw data.

Teams can view pipeline‑level or activity‑level metrics in real time (or T+1). Issues identified by metrics are turned into work items and fed back into the development plan for continuous improvement.

Conclusion

After more than a year of DevOps pilot practice, ABC has amassed extensive automation assets, built a release pipeline with visualized deployment nodes, and transitioned from coarse‑grained to fine‑grained management. The personal online banking system now embraces continuous integration, testing, and delivery, positioning the bank for broader digital transformation across all online financial services.