stylua public service

side project, trying to get unrust at rust (pun intended). basically a public stylua service for luau, web based, publicly accessible, with a studio plugin on top

since most plugins either compile rust --> wasm --> lua through wasynth and pack each plugin distribution with stylua (super slow), or they require an external/local web server making them super annoying to setup. if i just host one plugin-facing public api the whole community can use it, and bonus points i get to unrust

how it works

three crates:

• crates/api - axum server, formats inline when theres capacity, otherwise queues the job and hands you a session id
• crates/worker - drains the queue, visibility timeout + reaper so dead workers dont eat jobs
• crates/shared - types, redis store, config parsing

config is just stylua config as json. unknown keys get ignored, actually invalid values (like a garbage enum variant) get you a 400 instead of silently formatting with defaults

endpoints

• GET /health
• POST /v1/format - { "code": "...", "config": { ... } }, 256kb max. formats inline when it can, 202 + session id when it cant
• POST /v1/sessions - same body, always queues
• GET /v1/sessions/{id} - session status
• GET /v1/sessions/{id}/result - formatted code, 409 if not ready yet, 422 if formatting failed
• DELETE /v1/sessions/{id}

running it

needs redis, then:

REDIS_URL=redis://localhost:6379 cargo run -p api
REDIS_URL=redis://localhost:6379 cargo run -p worker

env vars, everything except REDIS_URL is optional:

• PORT - api port, default 3000
• MAX_QUEUE - queue cap before the api starts returning 503, default 1000
• VISIBILITY_TIMEOUT_SECS - how long a worker can hold a job before the reaper requeues it, default 30
• POLL_INTERVAL_MS - worker poll rate, default 200
• FORMAT_TIMEOUT_SECS - per job format timeout, default 5
• REAP_INTERVAL_SECS - default 10

Dockerfile builds the api, Dockerfile.worker builds the worker