The VPS is shared with other workloads and has no swap. On 2026-10-28 a
Docker daemon restart stopped every container; those with restart:
unless-stopped (all of Celtic) stayed down ~65 minutes until started by hand.
- restart: always on all services so the stack returns after a daemon restart
- mem_limit per service (db 512m, api 768m, backup 256m, web/caddy 128m)
- oom_score_adj -500 on db and api so the kernel kills other workloads first
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>