SELF-OPTIMIZING CI/CD PIPELINES USING REINFORCEMENT LEARNING IN CLOUD-NATIVE ENVIRONMENTS: A HYBRID SIMULATION AND CASE-BASED PERFORMANCE ANALYSIS

Abstract

Background: Continuous Integration and Continuous Deployment (CI/CD) pipelines are central to modern DevOps practices, enabling rapid software delivery in cloud-native environments. However, traditional pipelines rely on static configurations that fail to adapt dynamically to changing workloads, leading to inefficiencies in deployment time, failure rates, and resource utilization.

Objective: This study aims to develop and evaluate a reinforcement learning (RL)-based framework for self-optimizing CI/CD pipelines using a hybrid approach combining simulated environments and real-world cloud log-based case analysis.

Methods: A hybrid experimental design was employed, integrating simulated CI/CD workflows with anonymized cloud deployment logs. A Deep Q-Network (DQN)-based RL agent was trained to optimize pipeline configurations, including resource allocation, test prioritization, and deployment scheduling. Performance metrics such as deployment time, failure rate, rollback frequency, and resource utilization were compared between traditional pipelines and RL-optimized pipelines using statistical analysis (independent t-test and regression modeling).

Results: The RL-optimized pipeline demonstrated a significant reduction in average deployment time (28.4%), failure rate (21.7%), and rollback frequency (18.9%), along with improved resource utilization efficiency (p < 0.01). The hybrid model showed strong generalizability across simulated and real-world datasets.

Conclusion: Reinforcement learning offers a robust and adaptive solution for optimizing CI/CD pipelines in cloud-native environments. The proposed hybrid framework enhances operational efficiency and reliability, providing a scalable approach for next-generation DevOps systems