One manual EC2 box to containers on Fargate

The company

4 Elements Music is a music company whose platform ran on a single EC2 instance, deployed by hand. Their developers were mid-rewrite at the time — moving the application off Ruby on Rails onto Django — so the infrastructure had to serve both the thing that existed and the thing being built.

The problem

One manually-deployed instance is a performance ceiling and an availability risk in the same object, and it makes every change a decision about whether the site can afford to be down. Running a framework migration across that is worse again: you cannot stage a rewrite you cannot deploy twice.

They wanted a modern, scalable platform, and the networking done properly this time.

What I built

Terraform modules written from scratch, published to Terraform Cloud, so the infrastructure was reproducible rather than remembered.

Networking with an actual design. A custom VPC with three subnet tiers — public, private, database — and VPC endpoints to S3, ECR, Systems Manager, CloudWatch and OpenSearch, keeping service traffic off the public internet.

Compute and data. Containers on ECS Fargate, with S3, RDS (PostgreSQL) and OpenSearch behind them. I ran the data migration onto the new stores.

Development parity. A multistage Dockerfile and an improved Docker Compose setup, so what ran on a laptop resembled what ran in production.

Delivery. GitHub Actions with AWS CodeDeploy, using blue/green deployment — the point being that a release stops being an event you schedule around.

Edge. Cloudflare for DDoS protection and bot mitigation, and I migrated the 4elementsmusic.com DNS zone from Route 53 to Cloudflare.

Result

A standardised delivery workflow accelerated the release cycle, and least privilege was enforced across the infrastructure rather than asserted. The platform launched on documented, diagrammed foundations that someone other than me can maintain.