Audit-driven; goal: one user can't exhaust the box or starve the shared Traefik stack.
Resource containment:
- compose: mem_limit 512m, memswap_limit, pids_limit 200, cpus 1.0 (blast radius = container)
- Enforce per-user cloud quota (default 30MB, admin override) on /api/save + /api/characters → 413
- Cap published-character size (2MB); cap rooms (200), images/room (40), image bytes/room (40MB)
- Cap WS image dataUrl length in the wire schema; per-IP WS connection cap (20)
- MAX_USERS registration ceiling → 503 when full
Availability + correctness:
- Async crypto.scrypt (was scryptSync) so password hashing can't freeze the event loop;
constant-time on unknown users to avoid username probing
- trustProxy so req.ip is the real client behind Traefik — rate limits are per-IP, not global
- Tighter auth-endpoint bucket (10/min) on /register + /login; prune stale rate buckets
- O(1) token->user index; log persist() failures instead of swallowing them
- Trim /healthz info; surface quota in /api/usage + the admin dashboard storage bar
- Client surfaces 413 (quota) / 429 (rate) clearly
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
NilsBriggen
0892b8f19c