SHIELD: A Self-Healing AI Engine for Intelligent Log Diagnosis and Autonomous Remediation in CI/CD Pipelines

Abstract

Continuous Integration / Continuous Delivery (CI/CD) pipelines are at the heart of modern soft-ware delivery. However, CI/CD pipelines regu-larly fail due to various reasons, such as resource or library dependencies clashing, configuration errors, or flaky tests. Unfortunately, these pipe-lines often need to be manually fixed and re-started. While there are tools that help detect pipeline failures, there are limited solutions for automatically recovering failed pipelines, which has resulted in DevOps teams constantly blocking on failed pipelines. In this paper, we introduce SHIELD: Self-Healing AI Engine for Intelligent Log Diagnosis and Autonomous Remediation. SHIELD automatically detects, diagnoses, and remedies failures in CI/CD pipelines without any human intervention. SHIELD consists of a Natural Language Processing-powered intelligent log parser, a supervised machine learning-based failure classifier, and an autonomous remediation module that instantaneously implements the best solution to remedy the failure. Furthermore, we introduce a reinforcement learning loop as feed-back to improve our failure classifier continuously. We evaluate SHIELD and show that it drastically reduces Mean Time to Recovery (MTTR) with limited human debugging intervention.