- db-Volume auf /var/lib/postgresql (v18 legt PGDATA in .../18/docker; Mount auf
.../data laesst v18 nicht starten -> db unhealthy -> App-Rollback)
- Backup-Sidecar: Shell-Scheduler statt Alpine-crond (/etc/crontabs fehlt im
Debian-postgres:18-Image -> Crash-Loop)
Verifiziert auf TrueNAS: Custom-App gerbilmanager RUNNING, 2370 Tiere, Port 8090.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
deploy/truenas/compose.yaml:
- 4 services: db (postgres:17), api (.NET), frontend (nginx), backup sidecar
- Named volumes as NAS bind-mounts (paths configured in .env)
- Healthchecks: db pg_isready, api /health, frontend waits on api
- ANTHROPIC_API_KEY placeholder for FEAT-12a AI stub
deploy/truenas/.env.example: POSTGRES_PASSWORD, PORT, REGISTRY, dataset paths
deploy/truenas/scripts/:
- entrypoint.sh: installs daily 03:00 cron, starts crond in foreground
- backup.sh: pg_dump -> .sql + tar czf photos -> .tar.gz + rotation
- restore.sh: psql < dump + tar xzf photos; confirmation prompt (-f to skip)
Predecessor logic (commit 6a37f9e) translated from PowerShell to POSIX sh
for Alpine containers. Same pg_dump/restore semantics, same KEEP_DAYS rotation.
.gitea/workflows/ci.yml (DRAFT -- inactive until Julian confirms Actions):
dotnet test + npm test/build on push/PR to main;
docker build+push to Gitea registry on main push only
docs/ops.md:
German ops guide -- architecture diagram, first-install checklist,
start/stop/update commands, backup runbook (manual + auto), ZFS-snapshot
layering strategy, CI activation steps, open questions, troubleshooting table
Co-Authored-By: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>
