Building a Kubernetes Cluster with Kubespray: A Learning Journey
Series Introduction
This blog series documents the journey of setting up a Kubernetes cluster using Kubespray for research and learning purposes. The series covers everything from hardware setup, networking, and OS installation, to Kubernetes deployment, troubleshooting, and lessons learned.
Each post focuses on a specific stage of the project, highlighting challenges, solutions, and best practices. The goal is to provide a step-by-step, real-world perspective on building and managing a small-scale Kubernetes cluster collaboratively, while also exploring resource usage, monitoring, and optimization.
Whether you’re a student, researcher, or aspiring DevOps engineer, this series offers practical insights and hands-on experience for learning Kubernetes and distributed system design.
Why We Built a Kubernetes Cluster for Research
Every project starts with a reason. For us, it wasn’t just about running Kubernetes , it was about learning, experimenting, and building something real for research. In this post, I’ll share how my journey from machine learning to DevOps led to the idea of building a cluster with my team, and why this setup matters for the work we’re doing.
Introduction
We decided to build a Kubernetes cluster using Kubespray to explore system design, networking, and distributed environments. The cluster would also support our individual research projects — mine being focused on resource usage and cost-optimized allocation in Kubernetes pods.
This post is the first in a series, documenting both the progress and the obstacles we face along the way. It’s not just a tutorial , it’s a record of our lessons, mistakes, and what we learned from them.
Our first goal: get the hardware ready, install Ubuntu, and prepare the base system for Kubernetes.
Background to the Research
I’m currently in my final year of a B.Sc. in Information Technology at the University of Vavuniya Sri Lanka, where we are required to complete both a research project and an industrial training.
From my first year, I was fascinated by machine learning and data science. Over three years, I worked on small projects in ML and analytics, constantly learning as a side passion. Like many students in IT, I struggled with finding my niche , starting with web development (frontend, backend, full stack), then legacy frameworks like Spring Boot, and sometimes even doubting whether I belonged in this field.
A keynote speaker at an IEEE event once said:
“To be an IT professional is to adapt. Flexibility is your survival skill.”
That stuck with me. By my third year, I grew interested in DevOps and system design. At the same time, I didn’t want to abandon ML entirely. That’s when I discovered MLOps , a fusion of DevOps and machine learning practices. It felt like the perfect direction to combine my skills.
This is how the idea of building a research cluster became central to my project.
The Team and the Plan
The cluster project started with four of my colleagues, guided by my co-supervisor.
Together, we planned to build a small, locally hosted cluster — not just to test and deploy applications, but also to analyze performance, resource allocation, and the practicality of running workloads in this environment.
In the next post , I’ll dive into the hardware setup, first installation attempt, and the BIOS issue that forced us to switch systems.