[Pkg-zfsonlinux-devel] Bug#951080: Bug#951080: zfsutils-linux: Please disable shipped crontab zfs scrub script by default

[Richard Laager]

On 2/10/20 4:04 PM, Witold Baryluk wrote:
> The current package, significant deviats from the upstream, and is a
> rather arbitrary decision by the maintainer, not a user or zfs
> developers.

I wouldn't characterize this as a significant deviation from upstream.
But yeah, we should ship a scrub cron job upstream too.
> This decision also can negatively impact system performance with multiple
> pools being scrubbed at the same time, or just affect performance in
> general, i.e. on laptops when running on battery, or when pool is mounted
> from device on iSCSI share.
Which of these examples is your situation? There may be other issues or
changes to consider.

Can you share your zpool status output?

Scrubs are (supposed to be) throttled to reduce their impact. Are you
experiencing problems only after upgrading to 0.8.x? If so, the new
sequential scrub code may be a factor here. Overall, it is much more
efficient, but perhaps it is creating too much load for you.

It may be necessary to customize this script or the timings to your
particular needs. That doesn't make it a bad default.

The point of the script is to ensure that people's pools _are_ being
scrubbed. Otherwise, they may not find out about errors until it is too
late (i.e. they have data loss). ZFS scrubbing is analogous to mdadm's
check functionality, and the mdadm package ships a similar script on a
similar schedule. (The particular scrubbing time for ZFS was
specifically selected to be a different week than mdadm just in case the
user has both in use.)

-- 
Richard