[Gluster-users] crazy quota numbers

Manikandan Selvaganesh manikandancs333 at gmail.com
Tue Feb 14 17:09:43 UTC 2017


Hi,

In case if you do not want to do a quota enable/disable(which is a cleaner
operation)
you can try this:

Just a point to add, to what Sanoj has mentioned in the previous mail,
if you have identified the lowest level directory in the tree where the
accounting
has messed up, you can set "dirty" xattr flag on the directory. By setting
a "dirty" flag
marker will forcefully recalculate(add the sum of children and account to
parent) the
quota values. You can do this by setting "dirty" flag on (1) a gluster
mount which will
set on all the bricks or (2) you can do the same on a individual directory
where
quota accounting has gone bad:

setfattr -n trusted.glusterfs.quota.dirty -v 1 <directory>

--
Thanks & Regards,
Manikandan Selvaganesan.
(@Manikandan Selvaganesh on Web)

On Tue, Feb 14, 2017 at 11:42 AM, Sanoj Unnikrishnan <sunnikri at redhat.com>
wrote:

> Hi lejeczek,
>
> We have seen this issue before. https://bugzilla.redhat.com/sh
> ow_bug.cgi?id=1421933
> The problem here is that the accounting xattr in the backend becomes
> FFFFF...
> This happens likely due to an underflow (value goes below 0) on some
> operation like rm/unlink/truncate/mv. We would be adding defensive checks
> to prevent this, but that will not solve the accounting issue.
>
> This is a race and we do not have a consistent reproducer for it yet. *Could
> you describe the operations that were being done on the volume when u ran
> into this?*
>
> Due to the way quota works, if xattr of any directory goes wrong they get
> bubbled up to all its ancestors. So, it would also help if we can identify
> the lowest directory in the tree where the xattr has gone wrong and then
> try to find the operations on that particular directory.
> *Could you also share the xattr values on all bricks: find . | xargs
> getfattr -d -m. -e hex *
>> To recover from this you could do a quota disable and enable.
>
> Thanks and Regards,
> Sanoj
>
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://lists.gluster.org/mailman/listinfo/gluster-users
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.gluster.org/pipermail/gluster-users/attachments/20170214/b6b145ab/attachment.html>


More information about the Gluster-users mailing list