FlightOps is a flight booking web application, built in Python (Flask) with PostgreSQL, that I’m taking all the way to production-style delivery on Google Kubernetes Engine. Users can register, search flights between US states, book tickets and manage their bookings. This project is in progress, and this page separates what’s built from what’s planned.
At a glance #
| Role | Solo |
| When | September 2026 onwards |
| Status | In progress. App, database and container done; GKE, GitOps and monitoring under way |
| Stack so far | Flask, Gunicorn, PostgreSQL, Docker, Docker Compose, Kubernetes manifests |
| Planned | Terraform (VPC, GKE, IAM, DNS, secrets), GitHub Actions, Artifact Registry, Argo CD with Argo Rollouts, Gateway API, Secret Manager CSI, OPA Gatekeeper, Prometheus and Grafana |
Why I’m building it #
My earlier projects each covered one slice of delivery. FlightOps is meant to go end to end on one application with real state: a database with migrations, user accounts and bookings, delivered with GitOps and progressive rollouts, and guarded by policy. The kind of app where a careless deploy actually loses data.
What’s built so far #
- The application: Flask, with sign-up and sign-in, flight search, booking, payment and ticket pages.
- Database migrations: four ordered, idempotent SQL migrations (states, flights, users, tickets), so they can run again safely.
- A production container: a two-stage build on
python:3.12-slim, running Gunicorn as a non-root user with a read-only app directory,tinias init, and a health check. - Local stack: Docker Compose with PostgreSQL on a named volume, so data survives restarts.
- PostgreSQL on Kubernetes: a StatefulSet, Service, ConfigMap, PodDisruptionBudget, and a Job that runs the migrations.
- Decision records: written from the start, beginning with how the database is administered and persisted.
Planned architecture #
- Terraform provisions the VPC, a GKE Standard cluster, IAM, DNS and secrets.
- GitHub Actions builds and pushes the image to Artifact Registry, then updates the image tag in Git.
- Argo CD deploys from Git, with Argo Rollouts for progressive delivery of the app.
- Gateway API on a Google load balancer, with Certificate Manager for TLS.
- Secret Manager injects secrets through the CSI driver, OPA Gatekeeper enforces policy, and Prometheus and Grafana watch it.
Key decisions so far #
psqlover pgAdmin. The CLI is already in the Postgres image, adds no credentials, port or memory, and is what’s used to debug production. Recorded as ADR-0001.- Named volumes over bind mounts for local data, because they behave like Kubernetes persistent volumes.
- Migrations as a Kubernetes Job, separate from the app’s startup, so schema changes run once and visibly.
Related #
- Project Aether, where several of these patterns already run on my homelab
- GitOps best practices in my knowledge base