Multi-Cloud Smart Deploy: An AI-Based CI/CD Optimization with Kubernetes Rollback Strategy

Shaikh, Sahil (2025) Multi-Cloud Smart Deploy: An AI-Based CI/CD Optimization with Kubernetes Rollback Strategy. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (1MB) | Preview
Preview	PDF (Configuration Manual) Download (3MB) | Preview

Abstract

With the rise of cloud-native applications, we are once again faced with new challenges in terms of deployment orchestration, performance optimization, rollback strategies across multi-cloud environment, etc. The research focuses on a motivating problem statement around efficient deployment of containerized applications in multiple environments including GCP, Azure and AWS with effective monitoring and AI driven performance analysis. Continuous integration and continuous deployment. CI/CD pipelines are a must-have in the industry right now, but they are still largely confined to single-cloud scopes. In this article, we focus on innovative multi-cloud CI/CD deployment with Kubernetes cluster roll back integrated with auto AI based evaluation. The solution developed in this project is utilized as a GitHub Actions driven CI/CD pipeline responsible for the dynamic build, test, and deployment of a Django web application to Kubernetes clusters hosted over GCP, Azure, and AWS. Logs collected automatically and merged include post-deployment metrics such as startup time, rollout duration, and resource usage. Their inputs are logged from four machine learning models running inside Azure Machine Learning (Azure ML), which analyze cloud performance according to training accuracy and inference metrics. Under this systems pipeline architecture, intelligent decisions can be made on which cloud provider is best in the various scenarios. The implementation also provides Kubernetes rollback functionality to reverse bad deployments leading to enhancing reliability and service availability. The project enables key DevOps practices and machine learning workflows in a fully automated manner. The results show consistent and correct training through the ML pipeline, and evidence-based cloud selection. Overall, the proposed system provides greater resiliency, better performance visibility, and automatic rollback, as compared to traditional static deploy. Therefore, we strongly encourage the use of the proposed system in industrial multi-cloud deployments. In sum, the system shines on all axes we tested, and we leave more fine-grained resource profiling and predictive scaling for future work. This new effort is a major step forward in automating and intelligently delivering cloud-agnostic applications.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Gupta, Shaguna UNSPECIFIED
Uncontrolled Keywords:	CI/CD Pipeline; DevOps; Multi-Cloud Deployment; AI-Based Cloud Evaluation; Azure Machine Learning (Azure ML); Kubernetes Rollback
Subjects:	Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science Q Science > QH Natural history > QH301 Biology > Methods of research. Technique. Experimental biology > Data processing. Bioinformatics > Artificial intelligence Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Artificial intelligence Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning
Divisions:	School of Computing > Master of Science in Cloud Computing
Depositing User:	Ciara O'Brien
Date Deposited:	25 Nov 2025 17:51
Last Modified:	25 Nov 2025 17:51
URI:	https://norma.ncirl.ie/id/eprint/8959

Actions (login required)

View Item