Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 11 Dec 2019 17:39:03 +0000
From:      Norman Gray <Norman.Gray@glasgow.ac.uk>
To:        FreeBSD questions list <freebsd-questions@freebsd.org>
Subject:   Adding to a zpool -- different redundancies and risks
Message-ID:  <6104097C-009B-4E9C-A1D8-A2D0E5FECADF@glasgow.ac.uk>

next in thread | raw e-mail | index | archive | help

Greetings.

I'd like to add a new VDEV to a pool, and I'm being warned (slightly to=20
my surprise) that there's a 'mismatched replication level'.  I'm trying=20
to get a sense of how much of a risk I'd be running by forcing this with=20
-f.

Context:

   * I currently have two raidz2 VDEVs composed of nine 5.5TB disks=20
(thus 2 x ~40TB available)
   * I'd like to add another raidz2 VDEV composed of six 12TB disks=20
(thus adding a VDEV of ~48TB, roughly the same size as the other two) --=20
this is what prompts the warning about replication level
   * The storage is a local mirror which it would be very annoying to=20
lose, but it's not holding unique copies of anything
   * I don't like using -f options unless I'm pretty damn confident I=20
know what's happening

(I wouldn't set this up in quite this way from scratch, but this is an=20
old-ish server, and a small budget windfall has allowed me to max out=20
the remaining available slots with new disks).

I can appreciate that the ideal planned setup would, in principle, be to=20
have all the VDEVs be symmetrical, in terms of size and number of disks.=20
  Is a VDEV mix merely 'not ideal', or 'not great but you'll be fine',=20
or Bad?

My mental model of what's going on suggests that, since the pool simply=20
stripes across the VDEVs, it doesn't have to care how the VDEVs=20
themselves are structured, so that a 9x5.5 raidz2 and a 6x12 raidz2=20
would be roughly equally used, and I can't see why there would be a=20
performance or a utilisation difference between the two (but I still=20
count myself as a ZFS tyro).

I can see that there would be a reliability issue if the various VDEVs=20
were different sizes of mirrored ones -- this would create different=20
amounts of resilience, and so the warning makes sense in a 'are you=20
sure?' way.  If the VDEVs were different sizes and the pool was=20
mirroring over them, then there would obviously be a utilisation issue.

Though both of [1] and [2] illustrate only mixing VDEVs of the same=20
type, [2] says merely that 'When using RAIDZ vdevs, it is also a good=20
idea to keep them at the same width and of the same type.' and=20
illustrates a 4-wide plus 8-wide raidz2 as 'not horrible'.  The forum=20
post at [3] asks essentially the same question as this email, but=20
receives a rather oblique answer.  The question at [4] gets a confident=20
answer which I don't _think_ makes complete sense.

Since [1] and [2] are both more authoritative and match my own=20
understanding, I'm inclined to believe that adding this new VDEV would=20
be less than perfect, but reasonable.  Am I deceiving myself?

Thanks for any advice you can offer,

Norman


[1]=20
https://forums.freenas.org/index.php?threads/slideshow-explaining-vdev-zpoo=
l-zil-and-l2arc-for-noobs.7775/
[2]=20
https://www.ixsystems.com/community/resources/introduction-to-zfs.111/
[3]=20
https://forums.freebsd.org/threads/zfs-mismatched-repli-levels.28226/
[4] https://serverfault.com/questions/522782/zfs-with-unsymmetric-vdevs

--=20
Norman Gray  :  https://nxg.me.uk
SUPA School of Physics and Astronomy, University of Glasgow, UK



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?6104097C-009B-4E9C-A1D8-A2D0E5FECADF>