Self-Healing CI/CD Pipelines with Feedback-Loop Automation: Building Fault-Tolerant CI/CD Systems Using Anomaly Detection and Automated Rollback Logic

Abstract

Modern continuous integration and delivery (CI/CD) pipelines are crucial for rapid software releases, yet they risk introducing failures into production. This paper presents a comprehensive study of self-healing CI/CD pipelines that incorporate feedback-loop automation to achieve fault tolerance. We detail an architecture that integrates real-time anomaly detection (including machine learning-based techniques) and automated rollback mechanisms into popular CI/CD platforms (e.g. TeamCity, GitHub Actions, Jenkins). The goal is to minimize production downtime and human intervention by enabling the pipeline to detect issues and revert to stable states autonomously. Grounded in large-scale industry deployments, our approach leverages continuous monitoring and intelligent decision-making to reduce mean time to recovery (MTTR) and improve reliability. We implement the system in a prototype and evaluate it with experiments simulating deployment anomalies. Results show significantly faster failure detection and recovery. MTTR improved by over 50%  as well as high anomaly detection accuracy and efficient rollbacks. We discuss design trade-offs, such as balancing false positives in detection versus safety, and highlight how feedback loops can continuously improve pipeline resilience. The findings demonstrate that self-healing CI/CD pipelines can substantially enhance scalability and reliability of software delivery while minimizing manual oversight, paving the way for more autonomous DevOps processes.