ZFS already limits the amount of IO that a scrub can do. Putting
multiple pools on the same disk defeats ZFS's IO scheduler.* Scrubs are
just one example of the performance problems that will cause. I don't
think we should complicate the scrub script to accommodate this
scenario.
My suggestion is that you comment out the default scrub job in
/etc/cron.d/zfsutils-linux and replace it with something that meets your
needs. Don't change /usr/lib/zfs-linux/scrub, as that will get
overwritten on package upgrades.
For example, you might scrub the pools on different weeks with something like
this:
24 0 0-7 * * root [ $(date +\%w) -eq 0 ] && zpool list -H -o health POOL1 2>
/dev/null | grep -q ONLINE && zpool scrub POOL1
24 0 8-14 * * root [ $(date +\%w) -eq 0 ] && zpool list -H -o health POOL2 2>
/dev/null | grep -q ONLINE && zpool scrub POOL2
I'm going to boldly mark this Invalid. Others can override me,
obviously. Or, if you want to make more of a case, go for it.
* As a side note, in the general case, such a configuration also implies
that one is using partitions. This means they have the Linux IO
scheduler also in the mix, unless they're doing root-on-ZFS, in which
case zfs-initramfs is setting the noop scheduler. I assume you're doing
root-on-ZFS, since you mentioned "One pool holds OS", so that's not an
issue for you personally.
** Changed in: zfs-linux (Ubuntu)
Status: Incomplete => Invalid
--
You received this bug notification because you are a member of Kernel
Packages, which is subscribed to zfs-linux in Ubuntu.
https://bugs.launchpad.net/bugs/1731735
Title:
zfs scrub starts on all pools simultaneously
Status in zfs-linux package in Ubuntu:
Invalid
Bug description:
# Environment
Description: Ubuntu 16.04.3 LTS
Release: 16.04
Linux 4.10.0-38-generic-tuxonice #42~ppa1-Ubuntu SMP Mon Oct 30
20:21:41 UTC 2017 x86_64 x86_64 x86_64 GNU/Linux
zfsutils-linux 0.6.5.6-0ubuntu18 amd64
# Current behaviour
`/usr/lib/zfs-linux/scrub` starts `zfs scrub` on all pools at the same
time. If pools are located on the same disk - scrub performance
degrades badly.
# Proposed behaviour
* simplest one - start scrub of one pool after another scrub is finished
* advanced - detect pools which are located on the same disk and start scrubs
on them sequentially, if they are on different disks it is fine to run them in
parallel
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/zfs-linux/+bug/1731735/+subscriptions
--
Mailing list: https://launchpad.net/~kernel-packages
Post to : kernel-packages@lists.launchpad.net
Unsubscribe : https://launchpad.net/~kernel-packages
More help : https://help.launchpad.net/ListHelp
