diff --git a/homelab-overview.md b/homelab-overview.md index 0c18375..18cafab 100644 --- a/homelab-overview.md +++ b/homelab-overview.md @@ -119,6 +119,27 @@ Do NOT add per-subdomain A records — WebSupport rejects duplicate IP values zo --- +## Backup Policy (updated 2026-08-10) + +Backup storage: `spendlik-nas` (Proxmox storage ID, NFS-backed on NAS `192.168.1.12`). +Three separate backup jobs configured under Datacenter → Backup: + +| Job | Selection | Schedule | Retention | Notes | +|---|---|---|---|---| +| Nightly LXC | All except 110, 111 | Daily 23:59 | `keep-last=30` | Unchanged — original policy | +| Weekly VM | 110 (windows11) only | Sun 01:00 | `keep-last=2` (was 3) | VM 110 only used for Rhino work; project files live on Google Drive, so backup is a safety net, not primary storage. Reduced from 3 → 2 on 2026-08-10 to reclaim ~56GB. | +| Weekly Paperless | 111 (paperless) only | Sun 01:00 | `keep-last=2` | Carved out from nightly job on 2026-08-10 — was accumulating up to 30 nightly copies at ~28GB actual disk each, the single largest contributor to NAS backup growth. | + +**Pruning**: `keep-last` on a job only affects future runs. To reclaim space from existing backups immediately, run manually: +``` +pvesm prune-backups spendlik-nas --keep-last --vmid --type --dry-run +``` +Always dry-run first, then re-run without `--dry-run` to execute. + +⚠️ There's a gap in nightly LXC backup history from 2026-07-30 to 2026-08-03 — coincides with the pveproxy/NFS crash on 2026-07-28 (see `00_index.md` gotchas). Not a fault, just a known missed window. + +--- + ## Claude AI Integration The **mcp-server** container (ID 112) exposes a Model Context Protocol (MCP) server at `mcp.spendlik.sk`, providing Claude with tools to: