Nice. Thanks a lot! Similar in architecture to what I had in mind, so I’m inspired :)

A couple more clarifications, if you will! I’m asking dumb questions as that is the way I learn :D

• If your VMs need to access the data, do you then connect it via the nfs share?
• I suppose you have separate backup schemes for the data vs. the VMs?
• Does your bare metal Debian OS indeed run on the zfs pool too or does it have a separate boot disk? If on the pool, what’s that setup like? Is there a LUKS encrypted keystore partition to use with grub, or do you use the zfs boot menu? (I assume your pool is encrypted) -I’m trying to gauge how difficult this install is going to be if I want the OS on the zfs pool…

I just found out about virtiofs, and I’m piecing it together now. I haven’t done actual self hosting for long, so the conventions are a bit blurry, I’m basically piecing it together by what others seem to be doing and trying to understand why. I ended up realising I needed a much higher level discussion around this than “which fs should I use”. If you know of any resources that do NOT talk about specific technologies, but rather, principles behind them, I’d gladly bookmark!

So the changes I’m planning to my setup…

• encrypt the 2x960GB zfs pool and share it with [samba|virtiofs|nfs] from the host OS (checking later which one is the way to go)
• migrate all meaningful data (like application dbs) to reside on the pool rather than on the VM images and keep this separation of data&application layers to enable different backup schemes for them
• later / if I have the energy: try installing the host OS on the pool as well to get rid of the small SSD and make space for the HDDs.
• edit; also later: have the Pi provide a key server for unattended reboots, though if I simply leave the boot drive unencrypted and keep all the data in the pool this won’t be such an issue anyway, I can just remote in and type the passphrase for the zfs pool to get the data back online.

Uhh, from most what I have gathered from self-hosting so far, doing that is not trivial as you’d need to flush the ram contents to disk first basically. I’m starting to realize though that the same holds equally for filesystem level snapshotting. What I’m really after is making my data live on separate pass through storage that has all the fancy filesystem level stuff so I can just relax about the VM backups.