docs/infra/storage.md

# Storage layout

> Verified state as of 2026-07-15 (live audit during the CT 104 disk-full incident). Source of truth for which service lives on which storage class.
> Earlier revisions were stale on Garage, n8n, Arcane, Karakeep/Memos, Pocket-ID, arr-stack, Nextcloud's storage protocol, and missing Coder/Gitea. `volumes.md` was reconciled in the same pass.

## Storage classes

| Class | Substrate | Where | Typical use |
|---|---|---|---|
| Local zfs (rootfs) | NVMe in nuc | `/opt/stacks/<stack>/` on each LXC | Tier-1 hot state: databases, SQLite, secret stores, anything latency-sensitive |
| Local zfs (subvol) | NVMe in nuc | `rpool/data/subvol-<NNN>-disk-0` per LXC | Container rootfs, image cache |
| Docker named volume | NVMe in nuc | `/var/lib/docker/volumes/...` on each Docker CT | Per-container persistent state managed by Docker |
| UNAS over **NFSv3** | `192.168.1.31` | `/mnt/pve/unas` on CTs 101, 103, 104, 105, 111 | Bulk media, workspace home dirs, document storage |

## What lives where (verified 2026-05-21)

### UNAS NFS — `/mnt/pve/unas/services/...`

Bulk + media + non-latency-sensitive app data.

| Service | CT | Path on UNAS → in container | Backed up? | Notes |
|---|---|---|---|---|
| Immich (uploads, thumbs, derived) | 104 | `services/immich/{encoded-video,profile,thumbs}` + `media/images` + `backup/immich` | Immich own pg_dump → `media/images/db-dumps`; no off-host copy | |
| Paperless-ngx (documents) | 104 | `media/documents/public/paperless-ngx/{consume,export,library}` | none | |
| Paperless-AI | 104 | `services/paperless-ai` → `/app/data` | none | |
| Traccar (logs, config) | 104 | `services/traccar/{logs,traccar.xml}` | none | data dir reverted to local `/opt/stacks/traccar/data` |
| ~~Memos~~ | ~~104~~ | ~~`services/memos`~~ | — | **SUNSET 2026-07-10** — service + data + UNAS dir wiped. See [[CHANGELOG]] |
| Arr-stack (Prowlarr, RDTClient, Shelfarr, Audiobookshelf) | 104 | configs on `services/arr-stack/{prowlarr,rdtclient/config,shelfarr/storage,audiobookshelf}`; library + downloads on `media/{audiobooks,ebooks,podcasts,Torrents}` | none | **fully on UNAS incl. SQLite configs** (verified running 2026-07-15). Radarr/Sonarr not deployed. Stale local `/opt/stacks/arr-stack/media` (27 GB pre-migration copy) deleted 2026-07-15. |
| Gluetun (VPN) | 104 | `services/gluetun/data` → `/gluetun` | none | |
| Gitea | 111 | `services/gitea` → `/data` | none | new 2026-05-20 |
| Coder workspace home dirs | 111 | `services/coder/<user>/<workspace>` → `/home/<user>` | none | new 2026-05-20 |
| Backrest's jottacloud-mirrored data | 103 | `media/data-dir` only | **rclone → jottacloud (Backrest)** | only path with off-host backup |

### Local zfs — `/opt/stacks/...`

Tier-1 state and anything that should NOT be NFS-backed.

| Service | CT | Local path → in container | Backed up? | Notes |
|---|---|---|---|---|
| Pocket-ID | **110** | `/opt/stacks/pocketid/data` → `/app/data` | none | CT 110 has no NFS mount at all. Earlier doc said UNAS — incorrect. |
| Garage (S3 meta + data) | 104 | `/opt/stacks/shared-db/garage/{data,meta}` → `/var/lib/garage/*` | none | **moved off NFS** 2026-05-19 after WAL-G outage. Metadata is **SQLite** (`db_engine = sqlite`) — do NOT put on NFS. Data 82 GB on disk (2026-07-15), almost all WAL-G PG backups — see bucket table below. |
| Shared Postgres (multi-stack) | 104 | docker named volume `shared-db_shared-pgdata` | **WAL-G → local Garage** (configured; outage 2026-05-19 prompted move) | Garage itself is single-disk local — no off-host copy. |
| n8n | 104 | `/opt/stacks/n8n/data` → `/home/node/.n8n` | none | **back to local** after PG migration attempt failed. Stale 408 MB `database.sqlite` left on UNAS (May 15) — clean up. n8n now uses `shared-postgres` as its actual DB. |
| ~~Arcane~~ | ~~104~~ | ~~`/opt/stacks/arcane/data`~~ | — | **DECOMMISSIONED 2026-05-26** (replaced by Portainer). Compose renamed `.DECOMMISSIONED-2026-05-26`; data was local, not UNAS (volumes.md previously said "migrated" — wrong). |
| ~~Karakeep + Meilisearch~~ | ~~104~~ | ~~`/opt/stacks/karakeep/localdata/{data,meili}`~~ | — | **SUNSET 2026-07-10** — service + local + UNAS data wiped. See [[CHANGELOG]] |
| Immich Postgres (pgvector) | 104 | `/opt/stacks/immich/postgres` → `/var/lib/postgresql/data` | Immich pg_dump → UNAS | tier-1; PG demands local disk |
| Homepage | 104 | `/opt/stacks/homepage/{config,icons}` | none | decommissioned — replaced by Homarr on CT 109 |
| Dozzle | 104 | `/opt/stacks/dozzle/dozzle_data` | none | |
| ~~LiteLLM config~~ | ~~104~~ | ~~`/opt/stacks/ai/litellm-config`~~ | — | **SUNSET 2026-05-26** — replaced by Bifrost |
| SearXNG config | 104 | `/opt/stacks/ai/searxng` | none | |
| Flaresolverr | 104 | `/var/lib/flaresolver` | none | |
| AdGuard / Zoraxy / DNS / Shepard / Backrest binaries | 102/108/103/101 | local zfs only | none | Backrest itself has no self-backup |
| Vaultwarden (attachments + key material) | 104 | `/opt/stacks/vaultwarden/data` → `/data` | none | **moved off NFS 2026-05-22**; DB is on CT 113 postgres; stale `db.sqlite3` deleted |
| Nextcloud config + sidecars | 105 | local zfs (CT rootfs) | none | app data on NFS — see below |

### Garage S3 buckets (on local zfs, CT 104) — verified 2026-07-15

82 GB on disk, dominated by WAL-G Postgres backups. Trim via WAL-G retention on the source hosts, not by moving Garage to NFS.

| Bucket | Logical size | Objects | Use |
|---|---|---|---|
| `ct113-pg-backup` | 66.7 GiB | 1556 | WAL-G for CT 113 shared Postgres fleet |
| `immich-pg-backup` | 36.3 GiB | 4388 | WAL-G / WAL archive for Immich Postgres (+60 MiB unfinished multipart to clean) |
| `ct109-portainer-backup` | 1.9 GiB | 49 | Portainer backups |
| `shared-pg-backup` | 638 MiB | 4986 | shared-postgres WAL-G |
| `chat-artifacts` | — | — | AI chat output artefacts — `https://chat-artifacts.s3.nuclide.systems/<key>` |
| `owui-files` | 0 B | 0 | Open WebUI file storage |
| ~~`lobe-pg-backup`~~ | 151 MiB | 511 | **stale** — LobeChat decommissioned 2026-05-26; safe to delete |
| ~~`lobe-files`~~ | 0 B | 0 | **stale** — LobeChat |
| ~~`memos`~~ | 3.5 MiB | 4 | **orphan** — Memos sunset 2026-07-10; bucket delete blocked by quorum decode error, key revoked, needs Garage repair pass |

### Docker named volumes

Local on `/var/lib/docker/volumes/`. Mostly databases and caches.

| Volume | Service / CT | Backed up? |
|---|---|---|
| `shared-db_shared-pgdata` | shared-postgres / 104 | WAL-G → local Garage |
| `paperless-ngx_pgdata`, `_redisdata`, `_data` | paperless-ngx / 104 | none |
| `lobe-postgres` + `lobe-redis` | lobehub / 104 | none |
| `nuc-ai-core_rustfs-data` | lobehub stack / 104 | none |
| `pgadmin-data` | pgadmin / 104 | none (regenerable) |
| `immich_model-cache` | immich-ml / 104 | regenerable |
| `coder-db` + `gitea-db` (docker volume `coder-db`) | / 111 | none |
| (orphaned) `daytona-minimal_db_data` + runner anon vol | / 104 | none — clean up post-decommission |

### Nextcloud (CT 105) — data on NFS

Nextcloud AIO mounts `/mnt/pve/unas/services/nextcloud` via NFSv3 (same share as all other CTs). Migrated from CIFS on 2026-05-22 after the CIFS mount caused a crash-loop. The Postgres + Redis sidecars stay on local docker volumes.

## Backup reality

**There is effectively no off-host backup of service data.** The only configured Backrest plan covers `/mnt/pve/unas/media/data-dir → jottacloud` — i.e., Backrest backs up *one specific UNAS path*, not the services that write to UNAS.

Coverage:

- Immich pg_dump → UNAS (same filer; survives container loss, not filer loss)
- shared-postgres WAL-G → Garage (same host; survives container loss, not disk loss)
- Everything else: zero

This is a known gap. Plans:
1. Extend Backrest plans to snapshot tier-1 paths (Pocket-ID sqlite, Vaultwarden data, Gitea repos, Coder workspace homes) to jottacloud nightly.
2. Once the second NVMe lands (see `infra/proxmox-state.md` §6), mirror `rpool` so a single disk death doesn't take everything.

## ZFS snapshots (CT 104)

`rpool/data/subvol-104-disk-0` uses `refquota=200G` (not `quota`) — so **snapshots do not count against the container's 200 G limit, but they DO consume pool space**. A stale snapshot can silently exhaust the pool while the container's own `df` looks fine.

- Watch: `zfs list -t snapshot -o name,used rpool/data/subvol-104-disk-0`
- As of the 2026-07-15 incident, `@pre-tier1-backups-20260521-003708` held **97.9 GB** (pinning everything deleted since 2026-05-21, incl. the Karakeep/Memos sunset and docker prunes). Pool free was down to 57 GB. Destroying stale snapshots is the fastest pool-level reclaim.

## Drift cleanup TODO

- [x] ~~Remove stale `services/n8n/database.sqlite` from UNAS~~ — gone (verified 2026-05-22)
- [x] ~~Remove stale `services/shared-db/garage/` copy from UNAS~~ — gone (verified 2026-05-22)
- [x] ~~Remove orphaned `daytona-minimal_db_data` Docker volume on CT 104~~ — gone (verified 2026-05-22)
- [x] ~~Vaultwarden `data/` on UNAS~~ — migrated to local zfs 2026-05-22. Stale NFS copy at `services/vaultwarden/` can be cleaned up.
- [x] ~~Add a Backrest plan for Pocket-ID DB~~ — done 2026-05-22. Pre-backup hook runs `pocket-id export` + copies keys to UNAS staging; `services-backup-plan` snapshots staging → jottacloud nightly.