markandrewkamau/infraforge

Infraforge is a small, self-built Platform-as-a-Service that fits on a single laptop. Its purpose is educational: the author wanted to study how large internal developer platforms work, with Atlassian's Micros named as the reference example, and built a miniature version from scratch in Go to exercise the same control loop. A user describes a desired resource over HTTP, the system files the request, an asynchronous worker picks it up, and a real, network reachable Postgres container is created and reported back with its connection details. The README walks through that loop in six steps. A client posts a JSON body like service_name checkout, resource postgres to the broker on port 8080. The broker validates the request, writes a job record into Redis with a 24 hour TTL, pushes a reference to that job onto a Redis Stream, and returns immediately with a job id. A separate worker process is reading from that stream as part of a consumer group, which gives at least once delivery. When it pulls a job, it shells out to the docker CLI to run a postgres:16-alpine container named infraforge-pg-<jobID> on a random port bound to 127.0.0.1, then writes the outcome, ready or failed, back into Redis. The client polls GET /v1/provision/<id> until it sees ready, at which point the response carries a connection object. The code is partitioned so each concern is replaceable behind an interface. cmd/broker and cmd/worker are the two entry points. Inside internal there is an api package with HTTP handlers, a model package with shared Job, Status, and ConnectionInfo types, a store package that ships both an in-memory and a Redis-backed implementation of the same interface, a queue package for Redis Streams, a provisioner package whose current implementation drives docker but whose interface allows swapping in Kubernetes later, and a worker package that contains the control loop and is tested with fakes. A docker-compose.yml file brings up Redis. A GitHub Actions workflow runs go vet, go build, and go test with the race detector on every push. Quick start is three terminals: make deps-up and make broker, make worker in the second, and curl in the third to post a provision request and poll for status. Once a database is ready, docker exec into the infraforge-pg-<jobID> container with psql talks to it. make clean-pg removes every Postgres container the system created, and make deps-down stops Redis. All runtime settings are environment variables.