>>> On Fri, 18 Nov 2005 12:10:19 -0600, Eric Sandeen
>>> <sandeen@xxxxxxx> said:
[ ... ]
>>> Oh, repair on a 300T filesystem -wil-l be painful anywhere,
>>> I think, unfortunately.
>> hm, painful yes, but hopefully not impossible? otherwise if
>> sth. goes wrong on a 300TB fs the only way to fix it would be
>> restoring from backup, no matter how tiny the corruption
>> might be (and xfs_repair could fix it easliy if only the
>> volume was not so big)....
> The time & memory requirements for repair on a filesystem of
> this size are currently extremely large... there have been
> some rules of thumb for time/memory requirements on this list
> before, but I don't have them offhand...
As a handy pointer I summarized the issue a bit, with links
to past discussions, in a recent posting, Nov. 10th.:
BTW, I was scanning recently the XFS mailing list archives (and
got depressed by the sheer percentage of fools trying rather
ambitious things), and 'xfs_check'/'xfs_repair' time/space taken
is a common ''surprise''.
My impression is that the most common questions are:
* What about XFS and RH EL 3/4?
* What about files filled with zeroes?
* What about XFS, lots of other stuff, and 4K kernel stacks?
* How much RAM do 'xfs_check'/'xfs_repair' take?
* I got RAM and plenty of swap space, 'xfs_check'/'xfs_repair'
still fail on 32 bit system on filesystems larger than 2TiB?
* What about XFS, small lowmem and lots and lots of cached
pages with kernel 2.4?
* What about XFS and SCSI HAs that don't support READ/WRITE16?
I suppose that someone could put these in a little web page and
then put the URL in the footer of every email sent out by the
mailing list processor.