Cinder

Short-lived log uploads for Omarchy. Boring, constrained, disposable.

Logs come in. Cinders are what's left after they burn out — and after 24 hours, even those are gone.

API

POST /
  multipart: file=@log.txt, expires=24 (optional, hours)
  → 201, body is "<absolute URL>\n"

GET /:slug
  → 200, text/plain; charset=utf-8, X-Content-Type-Options: nosniff
  → 404 for unknown, expired, deleted, or blocked

GET /up
  → 200 health check

Errors are plain text. No HTML, ever.

Limits

Setting	Default	Env var
Max upload size	10 MB	CINDER_MAX_UPLOAD_BYTES
Default TTL (hours)	24	CINDER_DEFAULT_TTL_HOURS
Min TTL (hours)	1	CINDER_MIN_TTL_HOURS
Max TTL (hours, hard cap)	24	CINDER_MAX_TTL_HOURS
Cleanup grace (minutes)	60	CINDER_CLEANUP_GRACE_MINUTES
Rate limit per minute per IP	5	(in controller)
Rate limit per hour per IP	25	(in controller)

Binary content, archives, and images are rejected with 415. Oversize uploads get 413. Invalid expires falls back to the default TTL.

Operator

bin/rails 'cinder:delete[slug]'   # soft-delete + purge file
bin/rails 'cinder:block[slug]'    # mark blocked (returns 404 publicly)
bin/rails cinder:cleanup          # run cleanup now
bin/rails cinder:stats            # counts by status

IP/User-Agent blocking lives at the Cloudflare edge, not in Rails.

Cleanup

CleanupExpiredUploadsJob runs every 15 minutes via Solid Queue (see config/recurring.yml). It soft-deletes uploads past expires_at + cleanup_grace. Read paths also enforce expiry, so even if cleanup is delayed, expired logs stop being served immediately.

In production, configure your object storage bucket with a 1–2 day lifecycle rule as a safety net.

Storage

Active Storage. Disk by default. Switch to S3/R2 by editing config/storage.yml and setting config.active_storage.service in config/environments/production.rb.

Deployment

This repo only publishes the application image. To deploy it, build your own thin overlay image (FROM ghcr.io/thinkoodle/cinder:<tag>) that adds whatever environment-scoped credentials your instance needs, and point Kamal at that overlay.

The image expects:

• RAILS_MASTER_KEY for environment-scoped credentials.
• SOLID_QUEUE_IN_PUMA=1 so the recurring cleanup job runs in-process — no separate worker.
• CINDER_HOSTS to lock the Host header (comma-separated).
• Object storage configured in config/storage.yml (R2 example included) and pointed at by config.active_storage.service in the right environment file.
• Cloudflare in front for WAF, IP/ASN blocks, and edge rate limits.
• Object-storage bucket lifecycle policy of 1–2 days as a safety net behind CleanupExpiredUploadsJob.

Releases are tagged vX.Y.Z; GitHub Actions builds and pushes ghcr.io/thinkoodle/cinder:vX.Y.Z (plus :main on every push to main).

Metadata retention

Log content lives only until expires_at + cleanup_grace. Upload metadata (slug, sha256, ip, ua, status) persists after the file is purged so operators can investigate abuse and prevent re-uploads. There is no automatic metadata purge — prune it via Rails console when the audit window expires.

Tests

bin/rails test