Martins Eglitis

Developers in the company needed multiple environments (development, testing, and production) to run their application workloads. The requirements for the environments were similar but had some differences in configuration, such as cluster policies, domains, and security credentials. However, building these environments from scratch every time was a tedious and error-prone activity.

To address the problem, I built a Kubernetes platform, which is automated, scalable, replicable, safe, and convenient. The platform is built with the best practices in mind: automation (GitLab CI/CD), GitOps (ArgoCD), IaC (Terraform), secrets management (SOPS and Sealed Secrets), user and access management (AWS IRSA, SSO with Authelia, and login with Pinniped), cluster policies (Kyverno), and monitoring (Grafana, Prometheus, and Loki).

I built Kubernetes clusters on bare virtual machines to optimize costs. The requirements were to find the most optimal cloud provider in terms of cost, availability, and region. Based on research, I concluded that Hetzner is the optimal cloud provider, even though it lacks some AWS features and scale. I built the clusters with kubeadm, Cilium as CNI, and CSI provided by Hetzner. I wrote the configuration in Terraform, using modules and some custom scripts and configuration for better automation. The Terraform was then run from a GitLab CI/CD pipeline.

The company used a 1000+ line bash script to deploy code to production servers. This bash script had become unmaintainable and sometimes failed for no apparent reason, making the entire delivery process very flaky.

I rewrote the delivery tool in Go. It was based on Cobra and Viper because it had to support noninteractive use cases (from pipelines), interactive use cases, and conditional flags and arguments. I later improved the delivery tool and implemented a locking mechanism using AWS DynamoDB so that no deliveries could happen in parallel. I laid the groundwork for SlackOps so that the delivery tool could be triggered directly from Slack.

Sudoku solver is a brute-force recursive solver for 9x9 Sudoku puzzles, which I built for an educational project.

It consists of three sub-projects:

1. The brute-force sudoku solver library was written in Go and tested with conventional and fuzzy tests.

2. The website allows the end user to easily enter the numbers in the grid, solve it, download the results, or use the endpoints exposed as API. I wrote the website in Go, and it contains a Dockerfile for building the image.

3. The third sub-project is related to operations and consists of GitLab CI/CD pipelines, Terraform configuration for AWS, and Kubernetes configuration.