Martins Eglitis

I built a Kubernetes platform that was automated, scalable, replicable, safe, and convenient for developers in the company. The requirements for the development, testing, and production environments were similar, but they had some differences in configuration, such as cluster policies, domains, and security credentials. Building these environments from scratch every time was a tedious and error-prone activity.

The platform was developed with the best practices in mind, incorporating automation using GitLab CI/CD, GitOps using ArgoCD, IaC using Terraform, secrets management using SOPS and Sealed Secrets, user and access management using AWS IRSA, Authelia SSO, and Pinniped login, cluster policies using Kyverno, and monitoring using Grafana, Prometheus, and Loki.

This approach ensured reproducibility, safety, and convenience, streamlining the process of setting up environments and eliminating the need to start from scratch with each deployment. The developers could use the platform to quickly and reliably create the required environments for their application workloads, enhancing the overall efficiency and security of the development workflow.

This sandbox project was of great importance to me as it allowed me to explore and experiment with building Kubernetes clusters on bare virtual machines, optimizing costs, and finding the most suitable cloud provider.

Through thorough research, I identified Hetzner as the optimal choice, prioritizing cost-effectiveness and availability over certain AWS features and scale. Crafting the clusters using kubeadm, Cilium as CNI, and Hetzner's container storage interface was a rewarding challenge that deepened my technical expertise.

In this project, I embraced Terraform to configure the infrastructure, using modules, custom scripts, and automation for better management. Running Terraform from a GitLab CI/CD pipeline further streamlined the deployment process, ensuring a smooth and efficient workflow.

The company's deployment process faced significant challenges due to a sprawling 1,000+ line bash script that had become unwieldy and unreliable, leading to frequent failures during code deployment to production servers.

To address this issue, I took the initiative to rewrite the entire delivery tool in Go. Embracing Cobra and Viper frameworks, I ensured seamless support for noninteractive pipelines and interactive use cases alongside conditional flags and arguments. Moreover, I enhanced the tool by implementing an AWS DynamoDB-based locking mechanism, preventing parallel deliveries and promoting stability.

Additionally, I laid the foundation for SlackOps integration, enabling the delivery tool to be effortlessly triggered directly from Slack, streamlining the collaboration and communication aspects of the deployment process. This transformation significantly improved the reliability and maintainability of the delivery process, fostering a more robust and efficient code deployment workflow.

As part of an educational project, I developed a brute-force recursive Sudoku solver for 9x9 puzzles. The project comprises three distinct sub-projects.

The core of the solver is a Go-based library, meticulously tested using conventional and fuzzy test cases to ensure accuracy and reliability. To enhance user experience, I built a user-friendly website enabling effortless grid input, solving, result downloads, and API access. The website, written in Go, includes a Dockerfile for easy image building and deployment.

In the realm of operations, the third sub-project encompasses GitLab CI/CD pipelines, Terraform configurations for AWS, and Kubernetes configurations. These components harmoniously orchestrate the deployment and management of the Sudoku solver on cloud infrastructure.