Setu

Overview

Setu is an ambitious end-to-end Platform-as-a-Service (PaaS) modeled after Vercel's core deployment infrastructure. The problem it solves is providing a self-hosted, scalable environment where developers can push their code and have it automatically built, containerized, and deployed to a live URL without manual server configuration. It bridges the gap between raw compute resources and developer-friendly continuous deployment pipelines. The architecture of Setu is heavily distributed and container-native. The ingestion pipeline accepts project source code and places a build job onto a Redis-backed queue. I chose Redis for this queueing layer to ensure that concurrent deployments are handled safely, completely preventing race conditions across simultaneous builds for the same project. Worker nodes pick up these jobs, orchestrate isolated Docker containers to execute the build process securely, and then push the compiled artifacts to a MinIO cluster (an S3-compatible object storage solution). For traffic routing, I implemented an Nginx reverse proxy that dynamically maps incoming subdomains to the correct internal service, enabling automatic multi-tenant routing. A critical challenge was providing real-time visibility into the build process, similar to Vercel's deployment logs. To solve this, I integrated WebSocket streaming. As the Docker build container executes, its stdout and stderr streams are captured and piped over WebSockets directly to the user's dashboard, offering a live, interactive experience. Another significant hurdle was managing the lifecycle and cleanup of ephemeral build containers to prevent resource exhaustion on the host machines, which required strict timeout policies and automated garbage collection scripts. The outcome is a robust, highly educational platform demonstrating applied knowledge of distributed systems, multi-tenancy, and cloud-native infrastructure.