I bought a car that comes with a “free” 300k/30 year warranty, but only if to do oil changes every 4k miles or 3 months. Maybe this guy has something similar?

For me, I may try And keep it up for a bit, but driving to one particular dealer every 3 months just to get a ridiculous warranty that will probably never actually pay out isn’t worth it.

Also of note - if you’re using docker (and Linux), make sure the user is/group id match across everything to eliminate any permissions issues.

Not really, but I can give you my reasons for doing so. Know that you’ll need some shared storage (NFS, CIFS, etc) to take full advantage of the cluster.

1. Zero downtime for patching. Taking systems offline to update Proxmox sucks, especially if the upgrade fails for some reason. A cluster means I can evacuate one host, upgrade it, and move on to the next with no downtime for the hosted VMs.
2. Critical service resiliency. I have a couple of critical systems in my home lab that, if they unexpectedly go down, will make for a very bad day. For instance, my entire home network (and lab) is configured to use a PowerDNS cluster for DNS. I can put the master PowerDNS server on one host and the slave on a second host - if I have a hardware failure, I won’t lose DNS. I have a similar setup for my Kubernetes cluster’s worker nodes.
3. Experimentation. A cluster gives me a larger shared pool of CPU/Memory than my single host could offer. This means I can spin up new VMs, LXC containers, etc and just play with new software and services. Heck that’s how I got started with my Kubernetes cluster - I had some spare capacity so I found a blog post that talked about Kubes on LXC containers and I spun it up.

I hope that helps give some reasons for doing a cluster, and apologies for not replying immediately. I’m happy to share more about my homelab/answer other questions about my setup.

Those are beasts! My homelab has three of them in a Proxmox cluster. I love that for not a ton of extra money you can throw in a PCIe expansion slot and the power consumption for all three is less than my second hand Dell Tower server.

Sorry, I wasn’t clear - I use PowerDNS so that I can more easily deploy services that can be resolved by my internal networks (deployed via Kubernetes or Terraform). In my case, the secondary PowerDNS server does regular zone transfers from the primary in order to ensure it has a copy of all A, PTR, CNAME, etc records.

But PowerDNS (and all DNS servers really), can either be authoritative resolvers or recursors. In my case, the PDNS servers are authoritative for my homelab zone/domain and they perform recursive lookups (with caching) for non-authoritative domains like google.com, infosec.pub, etc. By pointing my PDNS servers to PiHole for recursive lookups, I ensure that I have ad blocking while still allowing for my automation to handle the homelab records.

This is overkill.

I have a dedicated raspberry pi for pihole, then two VMs running PowerDNS in Master/Slave mode. The PDNS servers use the Pihole as their primary recursive lookup, followed by some other Internet privacy DNS server that I can’t recall right now.

If I need to do maintenance on the pihole, power DNS can fall back to the internet DNS server. If I need to do updates on the PowerDNS cluster, I can do it one at a time to reduce the outage window.

EDIT: I should have phrased the first sentence: “My setup is overkill” rather than “This is overkill” - the Op is asking a very valid question and the passive phrasing of my post’s first sentence could be taken multiple ways.

I put my Plex media server to work doing Ollama - it has a GPU for transcoding that’s not awful for simple LLMs.

Hosting on the public web isn’t too crazy - start with port forwarding on standard ports (443 for sale/web) and add in a dynamic DNS address.

More than likely your residential ISP doesn’t change your IP that often, but Dynamic DNS solves that problem before it hits. I use Cloudflare, but mostly because I’m lazy and haven’t moved off of them after their most recent sketch behavior.