[Discuss] Debian 11 -> 12

[richard.pieri at gmail.com (Rich Pieri)]

On Thu, 30 May 2024 10:15:18 -0700
Kent Borg <kentborg at borg.org> wrote:

> I would amend that: Any new deployment?that is conventional (from
> ZFS's perspective) and can afford the necessary expertise.

While ZFS's *syntax* is different from say LVM + ext4, in *practice*
it's quite simpler:

zpool create tank /dev/sda /dev/sdb...
zfs create tank/mydata

vs

gdisk /dev/sda, create partition; repeat with /dev/sdb...
pvcreate /dev/sda1; repeat with /dev/sdb1...
vgcreate volumegroup /dev/sda /dev/sdb...
lvcreate -L size volumegroup logicalvol
mkfs.ext4 /dev/mapper/volumegroup-logicalvol
mkdir /mountpoint
edit fstab
mount /mountpoint

Want to change a mountpoint?

zfs set mountpoint=/path/to/mydata zpool/mydata

and now mydata is mounted as /path/to/mydata instead of /zpool/mydata.

Want compression? Simple:

zfs set compression=on pool/dataset

Can't do that with the commonly used filesystems (ext2 has a simple
compression mechanism but it's clunky and to my knowledge was never
forward ported to ext3 and ext4). That's the default algorithm there by
the way, but there are others which balance performance and compression.

Want encryption on your dataset?

zfs create -o encryption=on -o keylocation=prompt \
-o keyformat=passphrase pool/dataset

for a simple example, and you'd import this after reboot with

zpool import pool -l

'-l' tells zpool to request encryption keys for encrypted datasets.
Doing anything even vaguely similar with LVM + anything else, or just
anything else, requires mucking around with cryptsetup and loopback
devices.


> matter of taste, I found it ornery. And it flat out *crashed* when I
> tried to do the same stuff on a Raspberry PI 4. I was certainly doing

And as I have noted in the past, SD cards are inherently flakey.

Your Pi itself might be flakey. Or overheating. Or power management is
set wrong. Or insufficient power. Or any number of possible root causes
which aren't ZFS itself.


> As far as I can tell ZFS is a specialized tool, with impressive 
> features, but rough edges. It is not a smoothly crafted, general
> purpose package suited to a general audience.

I almost agree on a technicality: ZFS was not designed for a "general
audience". It was designed to be the last word -- or at least the last
letter, "Z" -- in enterprise scalability and performance. But it just so
happens to be really good at smaller scales, too. Better than almost
anything else I've used, but I have a fondness for Digital's AdvFS for
OSF/1 aka Tru64 Unix and there may be nostalgia goggles in the way.

-- 
\m/ (--) \m/