Environmental Impact of CI/CD Pipelines

TL;DR

This study quantifies the significant environmental impact of GitHub Actions CI/CD pipelines, revealing substantial carbon and water footprints, and proposes mitigation strategies to reduce resource waste and environmental harm.

Contribution

It provides the first large-scale estimation of the carbon and water footprints of GitHub Actions, using over 2.2 million workflow runs, and offers practical mitigation recommendations.

Findings

Estimated carbon footprint in 2024 ranges from 150.5 to 994.9 MTCO2e.

Water footprint ranges from 1,989.6 to 37,664.5 kiloliters.

Most likely scenario estimates 456.9 MTCO2e carbon footprint.

Abstract

CI/CD pipelines are widely used in software development, yet their environmental impact, particularly carbon and water footprints (CWF), remains largely unknown to developers, as CI service providers typically do not disclose such information. With the growing environmental impact of cloud computing, understanding the CWF of CI/CD services has become increasingly important. This work investigates the CWF of using GitHub Actions, focusing on open-source repositories where usage is free and unlimited for standard runners. We build upon a methodology from the Cloud Carbon Footprint framework and we use the largest dataset of workflow runs reported in the literature to date, comprising over 2.2 million workflow runs from more than 18,000 repositories. Our analysis reveals that the GitHub Actions ecosystem results in a substantial CWF. Our estimates for the carbon footprint in 2024 range from 150.5 MTCO2e in the most optimistic scenario to 994.9 MTCO2e in the most pessimistic scenario, while the water footprint ranges from 1,989.6 to 37,664.5 kiloliters. The most likely scenario estimates are 456.9 MTCO2e for carbon footprint and 5,738.2 kiloliters for water footprint. To provide perspective, the carbon footprint in the most likely scenario is equivalent to the carbon captured by 7,615 urban trees in a year, and the water footprint is comparable to the water consumed by an average American family over 5,053 years. We explore strategies to mitigate this impact, primarily by reducing wasted computational resources. Key recommendations include deploying runners in regions whose energy production has a low environmental impact such as France and the United Kingdom, implementing stricter deactivation policies for scheduled runs and aligning their execution with periods when the regional energy mix is more environmentally favorable, and reducing the size of repositories.