Xfs_Repair

[Summers, James B. II]

Hello All,

Looks like I am beginning to see another lun about to fail that is one of my storage targets.

I am seeing:

kernel: XFS (dm-3): metadata I/O error: block 0x5b81f0 ("xfs_trans_read_buf") error 5 buf count 4096

in messages and then it recommends that I unmount and run xfs_repair. I am not seeing any errors in any of the fhgfs log files, so I am not sure if this could be a red herring from the kernel / multipath or not? Since I am not seeing any errors in the fhgfs logs I do not believe I need to try fhgfs-fsck.

But it seems like running xfs_repair may create problems for the fhgfs filesystem if it removes / fixes things in it’s metadata.

Ideas / Suggestions?

Thanks
Jim Summers
University of Oklahoma
jsum...@ou.edu

[Christian Mohrbacher]

Hi,

Am 19.08.2014 um 17:22 schrieb Summers, James B. II:
> Hello All,
>
> Looks like I am beginning to see another lun about to fail that is one of my storage targets.
>
> I am seeing:
>
> kernel: XFS (dm-3): metadata I/O error: block 0x5b81f0 ("xfs_trans_read_buf") error 5 buf count 4096
>
> in messages and then it recommends that I unmount and run xfs_repair. I am not seeing any errors in any of the fhgfs log files, so I am not sure if this could be a red herring from the kernel / multipath or not? Since I am not seeing any errors in the fhgfs logs I do not believe I need to try fhgfs-fsck.

not for now. It wouldn't make sense at all, if the underlying filesystem
is still corrupt.

> But it seems like running xfs_repair may create problems for the fhgfs filesystem if it removes / fixes things in it's metadata.

well, corrupt/faulty information on the filesystem may also cause
problems. So in my opinion repairing the XFS is the best you can do. If
data isn't corrupt xfs_repair shouldn't remove it. And if it is corrupt,
we had a problem with it whatsoever.

Regards,
Christian