Samah Kansab ¹, Matthieu Hanania ¹, Francis Bordeleau ¹ and Ali Tizghadam ², ¹Ecole de Technologie Superieure (ETS), Canada, ²TELUS, Canada

GitLab's Merge Request (MR) mechanism is a cornerstone of DevOps, traditionally used for code review and analysis. This study broadens its scope by leveraging MR data to explore multiple facets of the DevOps workflow. Using a dataset of 26.7k MRs from 116 projects across four teams in a networking software company, we first examine the impact of environmental and process changes. We analyze how external factors, such as the COVID-19 pandemic, and internal adaptations, like the migration to OpenShift, influence effort, productivity, and collaboration. Our findings indicate that while review effort increased during the pandemic, productivity remained stable, with up to 70% of weekly activities occurring outside standard working hours. Similarly, the OpenShift migration initially disrupted workflows then showed a successful adaptation, with stabilized performance metrics over time. Next, we analyze branch management practices, revealing that stable branches, particularly those linked to new releases, are prioritized, leading to faster review completion. Finally, we apply machine learning (ML) to explain the time to complete code reviews, highlighting the roles of bots and human reviewers in industrial context. While bots accelerate review initiation, human reviewers play a crucial role in reducing time to complete code review. Additional factors, such as the number of commits and reviewer experience, also significantly influence review efficiency. Our analysis contributes to extending the use of MR data beyond code review by demonstrating how it provides deeper insights into software development workflows, team collaboration, and process adaptations, offering a framework for leveraging MR dynamics to optimize DevOps practices.

Software process, DevOps, Merge request, GitLab, Code review