Automating Deployment and Operation of a Scalable Bare-Metal Kubernetes Cluster

The growing reliance on cloud-native technologies has made Kubernetes a cornerstone of modern software infrastructures. However, deploying and managing Kubernetes clusters directly in bare-metal environments still poses significant challenges. The lack of native provisioning mechanisms, the diversity of hardware configurations, and the need to coordinate manually upgrades and scaling operations make these environments difficult to maintain. For organizations and research institutions that depend on on-premise resources—whether for performance, security, or data sovereignty reasons—achieving the same level of automation and resilience found in public clouds remains an open problem. This thesis proposes an automated and extensible framework that brings cloud-grade manageability to bare-metal Kubernetes clusters. The framework follows a declarative and reproducible design that integrates the traditionally separate infrastructure and orchestration layers into a single, coherent workflow. Metal3 handles bare-metal provisioning, while KubeSpray leverages Ansible to automate cluster deployment and lifecycle operations such as scaling and upgrading. The networking layer is enhanced through the adoption of Cilium, providing advanced observability and ensuring efficient communication across workloads. At the operational level, Argo CD enables GitOps-based continuous delivery, ensuring version-controlled deployments and consistent synchronization of environments. The resulting system demonstrates a scalable and maintainable approach to operating on-premise Kubernetes infrastructures while greatly reducing the need for manual intervention. It shows that automation principles typically associated with cloud providers can be effectively reproduced in self-managed settings, substantially decreasing administrative complexity while preserving flexibility and performance. The framework also proves its extensibility by enabling the seamless integration of additional functionalities, exemplified by the incorporation of GPU workload management, which highlights its adaptability to evolving computational requirements.