> this is the nature of metadata journaling filesystems.
Exactly the same thing happens on XFS filesystems on IRIX 6.5.
We had a bunch of Origin systems that crashed every so often because
of some PCI bus problems, and every time this happened every file
on the system that had been updated the last seconds before the
crash would be full of, or almost full of ^@. Needless to say
this created havoc with CVS repositories and also checked-out
trees, since the CVS/Entries files would end up modified.
I first noticed this when my setiathome status files were empty
after the restarts.. *every* time. I had to install a cron job
that copied the status file every two minutes or so, the slightly
older copy would be okay so I could just copy it back after the
restart.
That was a nuisance, but when the CVS systems got wrecked it became
a little worse. Obviously we have backups, but the problem was that
you wouldn't necessarily notice right away that anything was screwed.
Hopefully SGI will fix it, or maybe they have fixed it already --
I don't know. I'm running 6.5.8 which isn't the latest (IRIX is
at 6.5.11 or 6.5.12 now I think), and it certainly wasn't fixed there.
If not, then XFS will never run on my Linux boxes, and certainly
not on the big Linux server that now holds our CVS repositories
instead of the SGI Origin system that we originally used :-)
In another posting Stephen Tweedie writes:
>Umm, no, it isn't. Ext3 would never allow that to happen in ordered
>metadata-journaling mode, and Chris Mason is already working to remove
>that window in reiserfs. It is by no means a necessary consequence of
>doing metadata-only journaling.
That's good news.
-Tor
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/