I keep looking at PBS... I'm the fence if I just want to copy the data that's on the VMs (Doing that for Semaphore actively, configs are on Github, Semaphore can be rebuilt from scratch and have its data restored) over to the NAS or do I want to do full VM backups. In my case, I'm still working on getting the cattle going instead of having pets around. So with cattle, I just need to get the living data thrown to a backup point, and when one of the cattle dies or needs an OS upgrade (Which Debian is due to go EOL at the beginning of Q3), I can bring a VM up with the required config and Semaphore build, then restore the backup, and I'm back in the game. But then there's the whole "backup the VM" idea, which, to me, turns this back into a pet situation, not cattle. Because in that way you're treating the VM as it's own thing with its own quirks, and not a perfect restore from an IaC situation. That said, it's easier to restore the VM as a thing to itself from the last VM backup.... Of these two, one requires tending the herd (Configure the VM to do its own backups to the NAS for each type of cattle, or in other words, define what content in the VM needs to be backup specifically for that VMs job in life), or just backup the whole thing and let drift happen if it happens..... I guess a hybrid or use both processes could also be done. Indicate the critical machines (pfSense and Semaphore) to be done with PBS while everything else is done via Semaphore and internal backups..... #FirstWorldProblems

@Jose Pelaez NUCs for $50 even.... Swarm of Proxmox nodes inbound..... >:]

@Brandon Lee I don't need "hope", I know it's going to happen, bubble or not. We got through Covid, no one was making equipment, people stayed home and upped their rigs because the occasionally used system was now their primary rig, so demand went up while supply went down which made prices go insane. We got through earthquakes and flooding, which made manufacturing literally STOPPED. It's going to take time to "heal" from this stupidity, but it'll happen. But I'll be really straight up. I am so glad that I got my 5070 a year or so back, same with my new AM5 build (Previous life was my AI and game server setup). I was tempted to get an elcheapo GPU for another lesser-AI build, but decided to put money into retro equipment for the next couple months. I get more joy out of that and less gray hairs.

I'm maintaining size and scope unless this 10+year old hardware decides to give up the ghost. If I do downsize, it'll be just to power off one or two of the nodes after I migrate the VMs off to the to-be primary machine. chances are it'd be temporary as I prefer redundancy versus having money I guess?

I use VMWare at work, ProxMox at home. VMWare for close to 20 years since I started where I'm working at. I have (had?) paperwork from VMWare for training in all things VMWare from 15 years ago, and all that. I'm pretty good with it. ProxMox at home, I've been using it for about the same amount of time, just not on the Enterprise grade hardware I have now, but, on desktop PCs and such. Your comment in your blog: "But here is the important part. It is 100 percent functional for everything I want to do. I kind of had an epiphany after looking back at this past year. I care far less about polish than I thought I did, and far more about the capabilities and day-to-day management." For me, function over form every day, every time, all the time, no exceptions. If you can't get to where you need to be fast, quickly, and efficiently, then it's not worth the effort. I don't care about flashy modern looks, I care about consistency, getting to the place that's familiar from 20 years of experience, and get in and out as fast as I can. Take me back to DOS TUI, I don't care, so long that it's a few keyboard clicks away, I'm happy.

1> Firewall - I don't trust ISP provided modems to act as a router. I don't trust companies that make routers to be resilient enough to last more than a year due to growth of services, bandwidth, and networking limitations (Memory constraints). I use pfSense, put the ISPs modem into bridge mode, pfSense gets an IP from the modem and acts as a router/dns/dhcp to start. 2> DHCP - I have to know what is on my network, which is why I spent SO MUCH TIME writing the tooling to get IPAM running. DHCP will start out in a default config to get whomever I need out on the network, but its intent is to take over whatever pfSense had to offer. [3, 4, and 5, I may swap around] 3> IPAM - Need something to control the DHCP/DNS servers 4> Semaphore - For Ansible builds, Github controlled configuration for almost everything 5> DNS - PiHole is my go to. I may try out the other one mentioned for funsies. 6> haProxy - To get external web requests into my network for my personal use 7> KASM - When I need to get onto my network without a VPN (Because work won't let me install VPN software to get me home when "needed") 8> ChangeDetection - Because I like knowing when certain websites change "something", be it product, pricing, etc. Discord notifies me of those changes. I have it poke the remote sites periodically, typically once a day, some are weekly, so I'm not lambasted with minute by minute changes. 9> Pulse - Monitor the above 8 (and more) "things" I'll put on my network. Note, no AI garbage. i have an n8n instance, but with the answers and fights I have with my LLM on my 5070, not worth the fight. I may look at writing my own logic cores for it to function when problems are documented, but even then I wouldn't trust it. Notification is enough. If I cannot trust the answers I get from an LLM, or its results, I'll have nothing to do with it for automation purposes, period. I'll go to Claude or ChatGPT to get answers, sure, but that's me ACTIVELY looking at its answers and deciding the validity then and there.

"When you combine Ollama with OpenWebUI as part of your home lab services, it gives you a way to totally duplicate what you can do with something like ChatGPT and control everything locally. It works with the CPU only, albeit very slowly, but if you add a GPU with GPU passthrough, you will drastically improve your experience." I highly, HIGHLY disagree with this. There are no GPUs on the market that a home labber is going to be able to afford that's going to do what the big boys do. Period. I run a 5070 (As I said) and the "conversation" side of it is horrible, no matter the model. Sure you're going to get an answer faster on a GPU compared to a CPU, but the slop is seriously real. Granted, fine, learning about it could be something important to know about, but saying "totally duplicate what you can do with something like ChatGPT" is far, far from any experience I've had local vs OpenAI/Anthropic.