Fun in the Homelab: May 2026

  • 4 min read
  • Tags: 
  • homelab

Yet More Server Overkill

I know, I know. I didn't need another server I already had two old Poweredges. But my main one was getting so full! Turns out having terrabytes of dedicated, always-available, self-hosted storage turns you into a bit of a data hoarder. I've taken to scanning enough old, personal documents to make a paralegal proud. So like I say, my main Poweredge was getting pretty old. And 4 TB in RAID 51 was starting to feel a little bit restrictive.

On top of that, the SAF (Spouse Approval Factor)2 concerning the Server Room Where We Also Happen To Do Laundry was dropping. You see, a Dell R310 is a loud beast, to say the least. And it made that room hot. Neither of which were ingratiating to my wonderful wife.

It was time to replace that aging R310.

And, like you do when it's time to upgrade or change something, I wanted to take this opportunity to learn something new, that I had since heard good things about, or now felt "Leveled Up" enough to be able to manage. I had learned a little about Proxmox VE and decided the remote access feature would be invaluable for me.

So I did the Only Logical Thing(TM): I bought more hardware. A nice, shiny, renewed PowerEdge R530. Much quieter, and much more powerful. It's got a whopping 32 cores, split over two sockets. And the generation of Xeons it takes are apparently abundant on Ebay, so I have cheap upgrade options if I ever feel so inclined.

I was so glad to have a beefier, quieter server to work with! Buuut then it sat on a shelf for a couple weeks. I hadn't yet thought through the upgrade process. How could I migrate everything over to Proxmox without huge service disruptions?

The Great Migration

The opportunity finally presented itself a few weeks later, during a family trip to a party two hours away. I set up a VPN beforehand, and copied over as many files as I could ahead of time. On the drive, I whipped out my laptop and started shutting down services on the original server. Then, I did something incredibly bold and extremely frowned upon.

I just rsync-ed the config and data over.

I knew no databases were running. No state was being written. It was all just sitting on disk. Thanks to Docker, I had control over where application state was being written to, so I knew not to leave anything out. I organized this homelab server into /srv/config/${application_name} and /srv/config/${application_name}, so I just had to copy /srv/{config,data}.

I honestly didn't expect it to work. But I have yet to find a database issue from doing it this way.

Do I recommend this tactic? Heavens, no. Would I do this in a production environment? Absolutely not. But I include its success here as an interesting result to report.

Proxmox is Different

Honestly, the greatest challenges in migrating to Proxmox was using Proxmox itself. I had to figure out how to pass through all the many USB dongles Home Assistant makes use of. I still have some permissions weirdness on the VM running Home Assistant that I have yet to figure out.

Overall, though, Proxmox VE has given me what I had hoped for: (More) easily-managed environments for all my homelab services. I would recommend it, but not as a first server experience. I still run most of my stuff as Docker Composes, which I think are the superior way to run homelab services since they are easily backed up and don't require massive disk image storage.

Notes

1

I know, I know. RAID 5 is no longer recommended because the chances of being able to rebuild a failed RAID 5 array is less than 50%. But I'm operating on a homelab budget, not an enterprise budget. Big drives are expensive. Because of this, any chance at recoverable redundancy is better than the alternative I would have picked: RAID 0. So this is a compromise between "getting half the storage I paid for" (RAID 1 or RAID 10), and being most susceptible to drive failures (RAID 0). It's a middle-of-the-road option, to be sure.

2

Some prefer the term "Wife Approval Factor (WAF)." While, in my experience, this term is accurate more often than not, I believe it benefits us all to use inclusive language whenever feasible.