Re: zfs on raw vs gpt

[David Brownlee <abs%absd.org@localhost>]

On Thu, 29 Jul 2021 at 21:08, Pouya Tafti <pouya+lists.netbsd%nohup.io@localhost> wrote:
>
> Hi NetBSD users,
>
> Any advice on using zfs on raw disks vs gpt partitions? I'm
> going to use entire disks for zfs and don't need
> root-on-zfs.  One advantage of using partitions seems to be
> to protect against the risk of having to replace a disk by
> a marginally smaller one.  But I'm wondering if there may
> be a reliability penalty or other cons (also using raw
> disks would be marginally less work).

Modern disks tend to be of a given size, so a set of "2TB" disks from
different manufacturers will all be the same (much less than 2 TiB)
size, so I think thats less of an issue

I started setting up using raw /dev/wdX or /dev/sdX devices, but
switched across to ZFS gpt partitions. Both work fine, but for me:

raw disk devices
+ simpler to setup
+ no additional disk devices cluttering running system utility outputs (*)
- more risk of accidental overwriting the disk ("Oh this is the empty
disk then?")
- have to export and import -f everything if devices renumber on
reboot (which can be quite scary the first time)

disks with single zfs gpt partition
- clutters your running system utility outputs with dkN devices (*)
+ somewhat safer as you can check if its a zfs disk with gpt
+ you can setup devpubd and then create the pools using /dev/wedge/nicenames
+ with a small tweak to devpubd pools can be robust if devices
renumber on reboot


*: this I find annoying)
% sysctl -n hw.disknames
ld0 wd0 dk0 sd0 dk1 sd1 dk2 sd2 dk3 sd3 dk4 sd4 dk5 sd5 dk6 sd6 dk7
dk8 dk9 dk10 sd7 dk11