STUDY OF VARIABILITY OF ARCHITECTURAL METRICS DURING THE EVOLUTION OF A SOFTWARE SYSTEM IN CI/CD CONDITIONS

Abstract

Ensuring the architectural quality of software systems in continuous development is an urgent task of modern software engineering. The work is devoted to the development of a method for quantitatively assessing the metrics of the evolutionary architecture of software systems in continuous integration and delivery (CI/CD). An approach is proposed to construct a generalized fitness function that aggregates six structural quality metrics - SpotBugs violations, critical violations, line coverage by tests, code smells, line duplication density and Checkstyle violations - into a single quantitative assessment that is calculated automatically at each execution of the CI/CD pipeline. The pipeline normalizes each component by comparing the current value of the metric with the base value of the main branch stored in the cloud database. The weighting coefficients of the metrics are determined in accordance with the classification of quality characteristics priorities according to the ISO/IEC 25023 standard, where reliability, testability and maintainability receive relative priorities, respectively. An algorithm for integrating the architectural metrics assessment process into a CI/CD pipeline based on GitHub Actions has been developed, which ensures timely detection of signs of architectural degradation for each pull request and on schedule for the main branch. The approach has been experimentally tested on a real monolithic Java/Spring Boot system within 12 controlled scenarios covering improvement, degradation, and neutral changes in architectural quality. Statistical analysis confirmed the ability of the fitness function to qualitatively distinguish between different types of architectural changes. Comparison with the SonarQube Quality Gate mechanism showed the advantage of the proposed method: unlike the binary “pass/fail” signal, the fitness function quantifies the direction and magnitude of the variability of architectural metrics and avoids false positives in purely structural refactoring.