ZFS Pools, Physical Disk Location and Boot Mirror

[WiteWulf]

I think I’m correct in saying that TNS and ZFS utilises disk IDs, rather than device names, for vdev/pool membership, right? So, in theory, I could shut the server down, shuffle the disks into new slots, boot up and all the vdevs, pools and datasets would be fine.

However, I note that when I run ‘zpool status’ from the command-line, my mirror-0 boot pool is listed using device names (in my case sda and sdb).

Is this because the system has to boot off /dev/sda before it’s started the ZFS service, and it only becomes mirrored (and, I guess, read/write) once it’s booted far enough to run ZFS?

Would I also be correct on thinking that, as this zpool is defined using device names, I cannot physically move those disks? (Well, I could, but it would break things)

I’ve stuck labels on those two disk caddies for now so I know what they are and where they should be located.

[dan]

@WiteWulf

So, in theory, I could shut the server down, shuffle the disks into new slots, boot up and all the vdevs, pools and datasets would be fine.

Correct.

@WiteWulf

I note that when I run ‘zpool status’ from the command-line, my mirror-0 boot pool is listed using device names (in my case sda and sdb).

Also correct. Someone made a bit of a stink about this in the release topic for SCALE 24.04; prior to that I can’t remember seeing it mentioned (and FreeNAS/TrueNAS has been using ZFS for the boot device for 10 years now).

@WiteWulf

Would I also be correct on thinking that, as this zpool is defined using device names, I cannot physically move those disks?

That would make logical sense, though I’m not sure it’s been demonstrated as yet.

[WiteWulf]

Thanks, Dan. I’ll take a “better to be safe than sorry” approach with this, then.