A long Monday: an on-site network test, three faults at home, and a lot of work on the leads tool.
Hotel IPTV test on site
I sent the hotel's IT provider a short list for the VLAN 220 test. Core switch 1 port 22 had to be untagged on 220 only, and port 44 needed 220 added as tagged. Both were done by 12:32, and VLAN 50, the hotel's own TV network, kept the same querier as before.
I patched the IPTV switch's GE1/0/24 into port 22 and brought it up with a five minute commit timer. It had no link until I reseated both ends, then it came up at 1G with no hotel TV addresses leaking in.
Room 11 had no link at all. My Mac on the room jack showed the cable was bad. With that fixed, the Mac pulled an address from the IPTV server across the whole path.
The set-top box then got an address but stayed black. Its IGMP join never reached the IPTV switch, and it never heard a query. The core could see the querier, so the loss was in the UniFi switch between them: VLAN 220 wasn't in UniFi's IGMP snooping list, so UniFi dropped its multicast. The provider added it at about 13:53 and a live channel started a minute later at roughly 4.2 Mbps. Channel changes swapped groups cleanly. I'd ranked that fix as a later item, but it was needed first.
I stopped at 13:59. The double power points we asked for aren't fitted, so the room went back to the hotel TV on its old port.
pve02 froze, and the smart plugs
pve02 hung on disk I/O at about 02:15. Writes stopped, I/O wait hit 80 to 96% and load reached 27. Two minutes later it left both cluster networks, and HA moved its guests to pve01 and pve03 within about 80 seconds. Memory, temperatures and the network were normal. The hardware watchdog is off in the BIOS on every node, so it never recovered. The other reboots this month trace to the Oct 3 power loss, the planned shutdowns on the 5th and 7th, and updates on the 13th.
The Meross plugs were a separate fault. 13 of 15 mss315 plugs froze when the gateway went down on the 17th at 20:56 for the drive pull. They stayed on Wi-Fi with no IP address. Unplugging five fixed those, but the wine fridge plug came back with its outlet off. Kicking a plug off Wi-Fi from UniFi brought it back in 10 to 40 seconds with the outlet still on, so I did the rest that way, including the plugs under pve02, pve03 and pve04. The nodes stayed up.
Sim rig USB adapter
The sim rig lost its network 20 seconds to three minutes after I plugged in a USB 2.5GbE adapter. Turning off power saving changed nothing. Windows was using Microsoft's 2015 driver. I installed Realtek's through Windows Update and the link held for 24 minutes of checking. I added the adapter's MAC to the UniFi admin allowlist, which now has 35 entries.
Camera recorder move
I restored a fresh Protect-only backup from the gateway onto the UNVR Instant, and all five cameras came across with their settings. It sits at x.x.10.30 on the camera VLAN, with one new firewall rule for my PC, the server and the Home VLAN to its web port, and one for SMB to the NAS for archiving. The SkyHawk drive reads clean: 0 bad sectors, 47 C, about 4.9 years of power-on time. Protect estimates 14 days of retention, against 4 on the gateway.
Home Assistant refused to move its integration to a different recorder (wrong_nvr), so I removed it and added a new one. I checked 174 entities against a snapshot and the camera IDs, names and areas matched. The controls only returned once that account had full management rights.
Leads tool for a venue client
After deciding on the 18th to build my own, I got the tool most of the way there. It starts from open government data: 336 organisations within 10 km, with 234 corporate groups after merging branches. 206 have a phone or email, each kept with its source page and quote, and a script re-opens every page to check the quote.
The tool is one command with 729 tests passing. Writes are a dry run by default, back up first, skip the formula and staff columns, and refuse a workbook Excel has re-saved in the newer dropdown format. A clean Windows 11 VM on pve04 (VM 100) installed and ran it as a standard user, including locked-file handling. A probe pack on my PC answered the Excel questions, and tonight I started writing the code that patches the file's XML to keep the dropdowns.
Network audit
I audited my own VLANs. The management VLAN is the whole lab: hypervisors, NAS, my PC, every guest and the tunnel connectors. Home and management share a firewall zone behind a 36-entry MAC allowlist, and the Proxmox firewall is off on every node. The staged runbook adds local DNS names first, builds new beside old, and drains each node before touching it. Nothing has changed yet.
Still open
- Confirm the core switch config is saved. One check said it wasn't.
- Test a cold-booted set-top box without the IGMP override, and check Fast Leave before a second room.
- Find the pve02 stall cause in its own logs.
- Raise the IoT DHCP lease from 10 minutes to 24 hours.
- Set up the NAS archive share and check the Apple TV streams locally.
- Finish the tool build and run the full Windows trial.