RE: https://hachyderm.io/@badnetmask/115698872870594708

Home lab architecture update: two of the Lenovo boxes in the picture below became Talos nodes, together with the Framework Desktop. The third Lenovo became a Proxmox node.

Yep, I'm back on the Proxmox bandwagon.

Now I have a bit of everything: Talos running Kubernetes, Proxmox running VM/LXC and Raspberry Pi running Docker.

At some point the three Raspberry Pis are going away, and the services are going to be distributed between Talos and Proxmox, then they are going to become back-end compute for shenanigans (like a Home Assistant dashboard, for example).

#HomeLab #TalosLinux #Proxmox #Lenovo @homelab

Speaking about Home Assistant: I'm baffled about the fact that HAOS out of the box (no config at all) running as a VM is consuming 3.5GB of RAM, while the sum of Raspberry Pi OS + Home Assistant Docker, with thousands of entities, is barely touching 1.5GB RAM.

Is this because of the difference between x86_64 and ARM? Is that normal? 🤔

#HomeLab #HomeAssistant @homelab @homeassistant

I want to thank everyone who commented on my rant about Home Assistant VM eating so much memory. This morning I did some investigation, and I understand what's going on.

A few people who commented were right: it was all about buffers and cache. I didn't keep the VM up long enough for it to drain. After spinning up a new VM, I dropped into the shell and saw 3GB just for that.

Restarting HA didn't do anything, as it simply restarts the services. I had to actually reboot the VM. It came back using 1.4GB of RAM, which is kind of expected (still not great, but acceptable).

So yeah, I'll keep moving on migrating my Home Assistant from a Raspberry Pi to a VM on Proxmox.

Thanks hive mind.

#HomeAssistant #HomeLab @homelab @homeassistant

@badnetmask That sounds about right. I am running HAOS on a RPi, and it's requesting about 3.5GB of RAM.

I am, however, running several add-ons and have quite a few entities myself. IIRC, add-ons aren't available on HA Docker.

Also, HAOS is I believe a Debian derivative, so if you're running the HA Docker over say Alpine you'll probably have a smaller overall footprint.

@nanianmichaels
Well, I'm comparing a new/empty HAOS with a heavily used Docker on a Raspbery Pi OS (Debian based, full of integrations and entities).

@badnetmask Maybe there's some background stuff going on after the first boot, and thus requesting more RAM temporarily?

Also, how much RAM have you given the VM? You know what they say, unused RAM is wasted RAM, so maybe HAOS is just loading more stuff because there's RAM available?

@nanianmichaels
It's been running for a couple of hours, and usage doesn't go down. It has 4GB RAM allocated, but there's nothing installed yet, so it shouldn't be using that much.
@badnetmask @homelab please compare the readings of the VM with the reading inside HA settings. For me there’s a big difference. Probably all the addons but maybe there is a difference for you too
@aquisgranum @homelab @homeassistant
If you mean going into HA, Settings, Hardware, the memory usage there shows only 0.7GB. But what good does it make if the real usage on the Proxmox host is 3.5GB? I don't know where Home Assistant gets this memory information from.
@badnetmask @homelab @homeassistant maybe haos is taking reserves. But I was asking myself this over the last couple of years.

@badnetmask @homelab I can't imagine that's just because of the architecture difference. In my case, just Home Assistant is taking almost 500MB RAM.

#homeassistant #homelab

@badnetmask @homelab @homeassistant Block cache? Because a VM does it's own fs caching.
@badnetmask @homelab @homeassistant a linux VM with at least some activity will, after it’s been up for a while, always use a large portion of the RAM available. Memory that is not needed by active processes will be used for buffers/caches. So what you’re seeing is most likely not HA but the OS itself.
@badnetmask @homelab @homeassistant Measuring memory use is complicated with cache and shared libraries and such. Try using http://www.selenic.com/smem on both machines to compare.
smem memory reporting tool