W
After the success of the first node replacement, I'm now also replacing the other one, with the same M920q Tiny ThinkCentre. After running load tests (read stitching some big minimaps) for a day, I did the final reimaging yesterday.
Also installing an active P1 splitter so HA can run on both these nodes. This has been a very long-term item on the list. (For science!)
Timberborn added HTTP support in their latest version. Both sending and receiving requests. Just a URL, no body or anything. But that is enough to have it signal my @homeassistant instance about whether I have enough logs or not in stock. On a trigger, it sends a GET request to HA, and HA announces it through a voice satellite:
Gave it some more work, and blew out of my 3 credits way later than expected:
It did leave a bit of a mess tho, should harden my testcontainer set up:
Ow crap it's already done:
It's funny because this was supposed to be a mainly handsoff sessions while I game with my spouse. (Waiting for midnight.) But here I am approving commands to run that are identical 
Ever since the last cluster reinstall a few weeks ago Home Assistant has been running slower. Since it's been running on a Raspberry Pi 4, I decided to get a refurbished ThinkCentre to replace the node. Currently doing some testing, so far so good:
Hah it spiked due to Renovate kicking off some PRs and thus GitHub Action runners starting on it 
With this being the first Raspberry Pi 5 of the cluster, and since those are notorious for running hotter than the Pi 4. Which I did extensive testing on last year: https://blog.wyrihaximus.net/2024/12/building-a-kubernetes-homelab-with-raspberry-pies-and-lego-nodes-enclosure/. The decision was made to add the official active cooler with heatsink to this node. And it shows, slightly higher clock speeds, 10 to 20 degrees lower than the other nodes. But also only 1 out of 4 cores is really busy. So we'll see during more tests tomorrow.
First real test is underway! One of the things running on the cluster is some image conversion and stitching code. This simple #PHP service consumes from #RabbitMQ and then runs all the CPU heavy work in a thread. On the Pi4 it would blow out of memory (8GB) but on the Pi 5 it now has enough to do the job (16GB) . The cool part is that the memory usage graph really shows off the thread usage because normally PHP processes don't return claimed memory to the kernel.
