Restauranty

A food-ordering platform built as three independent Node.js services and a React frontend, deployed and operated as a real system rather than a one-time exercise.

It runs in two places on purpose. AWS EKS is the cloud-native version — cluster and networking provisioned with Terraform, TLS issued automatically by cert-manager, traffic restricted with Kubernetes NetworkPolicies (tested live, not just written), and monitored with Prometheus and Grafana deployed through Terraform’s own Helm provider. Hetzner is the permanent home — a self-managed VPS running the same containers behind Traefik, with its own CI/CD pipeline, its own monitoring stack, daily automated database backups with a tested restore, and a public status page.

Both environments deploy from the same Docker images through a GitHub Actions pipeline that builds, tests, and pushes on every change, then waits for a manual approval before touching either live server.

A few of the real problems hit along the way: a shared JWT secret that was wired into one service’s Kubernetes config but missed on the other two, causing a crash loop that took a kubectl logs dive to trace; a security group gap that only surfaced after a routine pod reschedule sent traffic across nodes for the first time; and a stubborn health-check failure that survived a retry loop with backoff — the retries not helping was itself the clue that the real bug was a protocol mismatch, not a timing issue.

Live demo: https://restauranty.kelenva.com Status page: https://status.kelenva.com/status/kelenva Monitoring: https://grafana.kelenva.com