On Wed, 2002-03-20 at 05:30, Ian Hardy wrote:
> I note that my exchange of emails with Steve Lord yesterday
> "Re: XFS NFS server Oops" with regards to checking for
> fragmentation and reducing fragmentation using 'xfs_fsr' has
> generated a followup thread "Re: Fragmentation (was: XFS NFS
> server Oops)" and some interest in deframenting XFS file systems.
>
> I'd recommend that you DON'T all rush to run 'xfs_fsr', at least
> until we've got some more feedback/info from the XFS experts
> at SGI. As advised I ran 'xfs_fsr' last night, it had completed
> when I got into work in the morning, so I ran 'xfs_db' in order
> to check fragmentation levels again:
>
> # xfs_db -r /dev/md0
> xfs_db: frag -f
> Segmentation fault (core dumped)
>
> Whats more it then shutdown the filesystem (from 'messages'):
>
> Mar 20 09:30:17 blue00 kernel: xfs_force_shutdown(md(9,0),0x8) called from
> line
> 1039 of file xfs_trans.c. Return address = 0xc01e2179
> Mar 20 09:30:17 blue00 kernel: Corruption of in-memory data detected.
> Shutting
> down filesystem: md(9,0)
> Mar 20 09:30:17 blue00 kernel: Please umount the filesystem, and rectify the
> problem(s)
>
> I rebooted the server and ran 'xfs_check' on the filesystem that showed
> a number of errors, which 'xfs_repair' fixed (I'll post some more details
> shortly).
>
> While its (very) possible that the above problem was due to underlying
> problems with my filesystem, it is also possible that 'xfs_fsr' had
> some bad effect on the filesystem. Worth getting some feedback from the
> XFS experts before too many people rush to run 'xfs_fsr' !?
>
> By the way I seem to remember a lot of SGI Irix users (including me) had
> various problems with 'xfs_fsr' when SGI first introduced it (or enabled
> its default running from cron) under Irix a few years ago. I believe
> these problems where fixed, however, I still don't routinely run
> 'xfs_fsr' on my Irix servers.
>
>
> Ian
>
Sorry about that Ian, do you have the repair output you can send me?
I did run fsr on a few filesystems here before suggesting it. xfs_fsr
itself is a purely user level program, all data copying is done via
system calls, the only special thing it does is switch the extents
of two files, this should be happening atomically, and really should
not be able to corrupt a filesystem. It would however look at pretty
much all the metadata on a filesystem and find problems if they existed.
Once you recovered what did frag -f report - i.e. did fsr do any good?
Steve
--
Steve Lord voice: +1-651-683-3511
Principal Engineer, Filesystem Software email: lord@xxxxxxx
|