A learning project exploring different ways to protect shared state when concurrent Node.js operations race to modify the same resource.

The project uses a simple locker checkout invariant: a locker can have at most one active checkout. Concurrent requests deliberately race against that invariant, then several strategies are implemented and tested to understand where each one provides its guarantee.

The strategies include:

• in-process keyed mutexes
• PostgreSQL advisory locks with READ COMMITTED isolation
• Redis-based distributed locks
• optimistic concurrency control
• PostgreSQL constraints

The goal is not just to make the race disappear, but to understand why each strategy works, what its coordination boundary is, and what changes when the application runs across multiple Node.js processes.

See also docs/PLAN.md for the full progression and docs/tests-overview.md.

npm ci
npm run docker:up   # Postgres (localhost:5433) + Redis (localhost:6379)
npm run migrate
npm run typecheck
npm test

• npm run docker:down stops and removes the containers. Postgres data is kept in the named postgres_data volume; Redis is intentionally ephemeral and starts empty the next time the containers are created.
• npm run docker:reset stops the containers and deletes the Postgres volume, dropping its schema and data. Re-run docker:up + migrate afterward to get back to a clean database.

Requires Node >=24.7 (see .nvmrc). Runtime commands such as migrate execute TypeScript directly through Node's built-in type stripping; Vitest transforms test files and their imports. The erasableSyntaxOnly compiler option keeps the application source compatible with Node's native execution. There is no build step.