Building scalable backend services with Rust and PostgreSQL

Sylvain Kerkour distills the patterns he uses to build medium-sized backend services (10K+ lines, ~100 endpoints) in Rust, on top of PostgreSQL. He concedes Rust isn't as productive as Go for backend work, but argues its rich type system, compiler-enforced correctness, and zero-cost abstractions pay off in fewer business-logic bugs and higher performance — and that enums alone make it hard to go back.

The article walks the full HTTP stack top to bottom: tokio as the async runtime, rustls (over openssl/boring) for TLS because it's Rust-native and statically linked, hyper as the protocol brain that turns bytes into Request/Response, and axum as the ergonomic layer on top (routing, extractors, middleware). reqwest is the client-side counterpart, tower the shared middleware abstraction, and tracing the converged observability story.

Code is organized into three layers that may only talk to their immediate neighbors: an HTTP/scheduler/worker layer, a service layer holding all business logic, and a repository layer that wraps database queries. Caching lives exclusively in the service layer — the repository "must stay dumb." Background jobs use a Postgres-backed queue, cron jobs use Postgres advisory locks for leader election across replicas, and single-page apps are served from the same binary via rust-embed to dodge the "CORS tax."

The closing advice is pragmatic: don't overthink the stack decision, just try Rust rather than planning meetings and months of research for a 20-endpoint service.