Data collected by live audit on 2026-07-11. Regenerate rather than hand-edit resource numbers.
⚠️ Partly outdated (checked 2026-10-09). The Ceph figures, diagram and tables below date from 2026-07-11 and have not been regenerated. Since then: the cluster has 3 nodes (mini-pc is gone); 13 OSDs / 45 TB raw, 14% used, HEALTH_OK, all 993 PGs active+clean — NAS 7 HDDs (osd.4–9, 11), M910q 1 × 16 TB HDD (osd.1), and SuperPC 5 NVMe SSDs (osd.0, 2, 3, 12, 13; ~2.8 TB). Replication is now size 3 for
ceph-storage,boot-disksandcephfs_metadata, and size 2 for the other pools (cephfs_data,backup-storage,ceph-frigate,jellyfin-media,immich-media,hanna-properties,stalwart-mail). All pools still share one CRUSH rule with no device-class split, so by capacity the data lands mostly on HDDs and the SSD OSDs are not a separate fast tier. See SuperPC.
ceph health detail)| Pool | Stored | Used by |
|---|---|---|
| ceph-storage | 415 GiB | LXC rootfs (RBD), Immich data disk |
| cephfs_data / _metadata | 2.2 TiB | /mnt/pve/cephfs |
| ceph-frigate | 43 GiB | unclear — Frigate media actually lives on cephfs; investigate & reclaim |
| .mgr | 14 MiB | Ceph internal |
The immich-data storage entry in storage.cfg is just an alias pointing at the same ceph-storage pool (not its own pool).
| Location | Size | Purpose |
|---|---|---|
| M910q /mnt/disk1 | 17 T (423G used) | Storj node data (deliberately outside Ceph) |
| SuperPC local-lvm | 40G | Frigate LXC rootfs (not migratable) |
| mini-pc local-lvm | 100G | VM disks (Jellyfin, amp-worker) |
| each node local | ~65G | Proxmox root |
| Dir | Consumer |
|---|---|
| media/ | Arr-Stack mp0, Jellyfin VM (virtiofs "media") |
| downloads/ | Arr-Stack mp1 |
| frigate/ | Frigate mp0 (recordings/clips/snapshots/exports) |
| monitoring/ | LXC 119 /opt/monitoring |
| wikijs/ | LXC 120 /opt/wikijs |
| dump/ | vzdump backups (weekly job) |
| jarvis/ | Jarvis LXC 130 mp0 (/srv/models): LLM model files — re-downloadable, not backed up |
| jarvis-notes/ | Jarvis LXC 130 mp1 (/srv/notes): Jarvis's git-versioned markdown notebook |
| template/cache/ | LXC templates (debian-13) |
| downloads/, import/, uploads/, snippets/ | misc |
Persistent service data lives on cephfs and is bind-mounted (mp0, shared=1) so guests can migrate and rebuilds are cheap. Exceptions today: OpenProject (named docker volumes on rootfs) and Frigate's config dir (on non-ceph rootfs) — both flagged in Audit Findings.
38 TiB free. Biggest growers: cephfs media (Arr), Frigate recordings (21-day motion retention), Immich. SuperPC has no OSDs — adding disks there would balance the NAS-heavy CRUSH map (75% of capacity on one host means NAS downtime = most replicas offline).