Proxmox¶

Nodes¶

PVE1 (Primary)¶

• IP: 192.168.1.199
• URL: pve1.mdhmedia.uk
• Role: Primary hypervisor

Hardware¶

Storage¶

Network¶

Since 2026-07-28 the sole uplink is the USB3 RTL8153 gigabit adapter (enx3c18a0d4eddd), running at 1 Gb/s. The onboard eno1 is an unplugged 10/100 RTL8106E (working, just slow and not connected) and was split out of vmbr0 the same day. A boot remediation chain — restart-usb.service → usbup.sh → /usr/local/sbin/nic-link-up.sh — guards against the USB NIC renegotiating at 100 Mb; see Network.

flowchart LR
    subgraph Bridge[vmbr0 - 192.168.1.199/22]
        usb[enx3c18a0d4eddd<br/>USB3 RTL8153 - 1 Gbit]
    end
    eno1[eno1<br/>RTL8106E 10/100 - unplugged, out of vmbr0]
    Bridge --> GW[Gateway<br/>192.168.1.254]

PVE2 (Secondary)¶

• IP: 192.168.1.151
• URL: pve2.mdhmedia.uk
• Role: Secondary hypervisor, heavy workloads
• BMC: Dell R710-era iDRAC6 at 192.168.1.150 — remote power/console independent of the host OS (confirmed 2026-08).
• Fans: Managed by the live proxmox/fan-control/ systemd service using in-band IPMI (/dev/ipmi0). Its thermostatic curve sets 12/18/25/35/50% at 0/45/55/63/70 C; any error, service stop, or 75 C ceiling restores iDRAC automatic control.
• Power: the proxmox/power-monitoring/ reader publishes node_power_watts{source="ipmi"} from the in-band /dev/ipmi0 "System Level" sensor (whole-system, ~160-200 W). pve1 instead reports source="rapl" (a CPU-package proxy from RAPL, read as root because the Platypus-CVE mitigation makes energy_uj root-only and Alloy runs non-root). Both feed the Power & Energy Grafana dashboard (mdh-power), costed by the editable Home Assistant input_number.electricity_unit_rate (24.0952 p/kWh) - HA's prometheus integration on .240 exposes it, the homeassistant scrape job pulls it, and the electricity_unit_rate_gbp_per_kwh recording rule reads it (constant fallback). Repoints to the new HA VM at the HAOS cutover.

Hardware¶

Storage¶

Network¶

flowchart LR
    subgraph vmbr0[vmbr0 - 192.168.1.151/22]
        eno1[eno1]
    end
    subgraph vmbr1[vmbr1 - 192.168.1.152/22]
        enp4s0f0[enp4s0f0]
    end
    subgraph vmbr2[vmbr2 - 192.168.1.153/22]
        enp4s0f1[enp4s0f1]
    end
    vmbr0 --> GW[Gateway<br/>192.168.1.254]

PVE3 (Wake-on-LAN Backup Node)¶

• IP: 192.168.1.28
• URL: 192.168.1.28:8006
• Role: Wake-on-LAN backup node (ZFS storage + Proxmox Backup Server) — normally powered off since 2026-07-30

Joined to the homelab cluster on 2026-07-24. On 2026-07-26 its two former NTFS disks were built into ZFS pools — tank (1.81 TB, VM/CT + ISO storage) and bak (464 GB, the PBS datastore) — and the node now hosts the Proxmox Backup Server VM (see Virtual Machines and Backups). Since 2026-08-04 the 500G bak disk is passed whole into the PBS VM, which imports the bak pool itself — the host no longer mounts it, and the pve3-bak/pve3-backup host storages were deleted. It still boots off its small OCZ SSD.

Since 2026-07-30 pve3 is a wake-on-LAN backup box: normally powered off, it is woken daily at 19:00 by the wol-pbs orchestrator container on CT 101 (the 13th Arcane git-sync stack, stacks/management/wol-pbs/), which runs the evening backup window and shuts the node down again once PBS is idle (minimum wake 110 minutes). If PBS is already awake at wake-time the orchestrator now continues the cycle and still shuts pve3 down (via an atomic mkdir lock); STANDBY_IF_UP=true restores the old standby behaviour. (Fixed 2026-08-12, #276.) pve3 stays in the cluster: pve1's quorum_votes was raised from 2 to 3 (total 5, quorum 3), so pve1 remains quorate on its own while pve3 sleeps.

Wake-on-LAN specifics: the magic packet targets enp6s0 (r8169). The driver clears the WoL flag at shutdown, so a pve3-wol-enable.service oneshot re-arms it, and the BIOS needs WoL enabled with ErP/EuP off. The PBS VM 102 has onboot=1 and autostarts on wake (~75 s to a responsive API). The orchestrator's SSH key is forced-command restricted on pve3 — it can only power the node off or start pbs-offsite-sync — and its PBS API access uses the read-only token api@pbs!wolpbs.

Hardware¶

Storage¶

Network¶

flowchart LR
    subgraph vmbr0[vmbr0 - 192.168.1.28/22]
        enp6s0[enp6s0<br/>1 Gbit]
    end
    vmbr0 --> GW[Gateway<br/>192.168.1.254]

LXC Containers¶

Statuses as of 2026-07-24. The cluster was rationalised 2026-07-07 — 13 stopped test/legacy guests (10 CTs + 3 VMs) were removed (see the note below). On 2026-07-20 the docker estate was consolidated onto CT 101: the Arcane manager (CT 100) and Homepage (CT 113) moved there and both LXCs were retired — stopped with onboot=0 and kept as rollbacks, not destroyed.

Virtual Machines¶

Backups¶

Since 2026-07-26 the backup topology is PBS-primary with a local vzdump floor and an incremental encrypted offsite. This realigns the earlier vzdump-only setup around the new Proxmox Backup Server (VM 102 on pve3). Since 2026-07-30 the whole PBS leg runs in an evening wake window: the wol-pbs orchestrator on CT 101 wakes pve3 at 19:00, the PBS jobs run, the offsite sync fires, and the node shuts down again once PBS is idle (minimum wake 110 minutes). The times below replaced the original 01:30 / 04:00–05:00 schedule. Full detail + tooling: proxmox/backup/ in the repo.

This gives a live 3-2-1: PBS (primary, dedup) → local vzdump (keep-last-1 floor) → encrypted Hetzner offsite.