Storage & Backup ZFS pools on hpve, what each holds, and what is (and isn't) backed up

All bulk storage lives on hpve as ZFS pools. This page is the authoritative map of what’s on each pool and, more importantly, what has a second copy and what doesn’t. Last verified 2026-07-27.

Pools at a glance
Pool Raw Redundancy Used Role
Biggest 18.2 T mirror (survives 1 disk) 81% Backup/archive target — holds nvmepool-backup, Kiwix, Loki, Maple, archive
mediapool2 18.2 T RAIDZ1 (survives 1 disk) 54% Primary movies + TV (Plex library). 5×4 TB, Thunderbolt enclosure
nvmepool 10.9 T stripe — NO redundancy 75% Primary working data: music, photos, ingest, Sync replica, CT100 disk, GIS
backups 464 G single disk 29% Proxmox vzdump VM/CT backups + ISOs
mediapool 10.9 T RAIDZ1 ~0% (empty, ~7.1 TB usable) Spare/free — leftover from the 2026-06-01 cutover
What's on each pool

mediapool2 (primary video)

  • movies — 6.68 TB (Radarr)
  • tv — 1.30 TB (Sonarr)
  • Auto-snapshots disabled 2026-07-27 (com.sun:auto-snapshot=false) after 204 stale snapshots were pinning ~4.8 TB. Pool dropped 81% → 54%.

nvmepool (fast working storage)

  • music 1.82 TB · photos 1.56 TB (Immich) · ingest 2.81 TB (seedbox landing / staging) · sync 1.33 TB (server-side Syncthing replica of ~/Sync) · container-data 267 GB · vms 293 GB (CT100 disk = 287 GB) · gis 12 GB · plus books, audiobookshelf, bookshelf, video, courses, podcasts

Biggest (backup/archive)

  • nvmepool-backup — nightly rsync of nvmepool data · Kiwix (ZIM offline wikis) · loki (log store) · Maple · archive · backups

backups — Proxmox vzdump dumps + install ISOs. mediapool — empty; ~7.1 TB usable available for reuse.

Backup coverage — what has a second copy
Data Primary Backup Status
Photos, Music, GIS, Books, Sync, container-data nvmepool Biggest/nvmepool-backup (nightly rsync) ✅ Covered
Proxmox VM/CT config & disks (each pool) backups pool (vzdump) ✅ Covered
Movies + TV mediapool2 none ⚠️ Single copy — accepted risk
Seedbox ingest (staging) nvmepool none — Transient by design
Anything (on-box) offsite ⚠️ No offsite backup

The one deliberate gap: the movie/TV library on mediapool2 has no second copy. RAIDZ1 survives exactly one disk failure; a second disk, a pool-level fault, or accidental deletion loses it. Ed reviewed this on 2026-07-27 and accepted the risk rather than replicate (the media is large and largely re-acquirable, though some back-catalogue is no longer available on current indexers). The idle mediapool (~7.1 TB usable) could hold a movies replica if that ever changes.

Second gap: nvmepool has no ZFS-level redundancy (it’s a stripe). Its protection is entirely the nightly rsync to Biggest — so a nvmepool disk failure means restoring from Biggest with up to ~24 h of changes lost, not zero-loss. Fine for its content (working/reproducible data), but worth knowing.

Maintenance & thresholds
  • Keep pools under ~80% full. ZFS (copy-on-write) shifts to a slower space-conserving allocator past ~80% and fragmentation accelerates. Biggest (81%) and nvmepool (75%) are the ones to watch.
  • Do NOT defragment — SSDs gain nothing from it and ZFS has no defrag tool. frag% is free-space fragmentation; the lever is free capacity, not a defrag pass. TRIM (autotrim/zpool trim) is the relevant SSD maintenance.
  • Scrubs run monthly (see Proxmox page); all pools last scrubbed clean (0 errors) in July 2026.
  • Snapshots: zfs-auto-snapshot runs on most datasets. It was disabled on mediapool2 (media doesn’t need point-in-time rollback and the snapshots were silently hoarding space). Check zfs list -o name,usedbysnapshots if a pool’s “used” looks larger than its file contents.