[Bugs] [Bug 958781] KVM guest I/O errors with xfs backed gluster volumes

[bugzilla at redhat.com]

https://bugzilla.redhat.com/show_bug.cgi?id=958781

Niels de Vos <ndevos at redhat.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
              Flags|needinfo?(ndevos at redhat.com |
                   |)                           |



--- Comment #29 from Niels de Vos <ndevos at redhat.com> ---
(In reply to JYundt from comment #28)
> (In reply to Niels de Vos from comment #26)
> 
> > The offset could be after EOF, but O_DIRECT is used to open the file. Maybe
> > there are some alignment issues...
> 
> And FWIW, changing some of the mkfs.xfs attributes (mkfs.xfs -f -i size=512
> -s size=512) fixed my problem.
> 
> Based on this information, do you still want me to try to reproduce with
> 3.5.1?

No, that is not needed. This definitely sounds like an alignment issue.

It would be interesting to know if the errors in the kvm-guest also happen when
the filesystem of the brick is used directly to store the images.

The bricks return the same values as the output of 'stat -f /bricks/dir/....'
when a volume gets mounted (by fuse or through the api). If the blocksize
differs between bricks, I do not doubt that the result is unstable.

We would need at least the following info to understand more about this:
1. does the issue happen without gluster, but on the filesystem of the brick
2. output of xfs_info from the filesystem
3. output of dumpe2fs from the ext4 filesystem on the same storage/disk

If you can not provide the details, or have no interest in digging further, let
us know and we'll close the bug.

[ See comment #24 and comment #28 contain a fix/workaround. ]

-- 
You are receiving this mail because:
You are on the CC list for the bug.
Unsubscribe from this bug https://bugzilla.redhat.com/token.cgi?t=fLfV5pAgZI&a=cc_unsubscribe