On Wed, 2010-10-20 at 22:20:31 +0300, Modestas Vainius wrote: > On sekmadienis 10 Spalis 2010 15:21:47 Guillem Jover wrote: > > Anyway, I actually had the changes around last July, and I'll run them > > through the release team to see what they say. I've rebased them now, > > and will polish them a bit, but they are essentially these: > > > > <http://git.hadrons.org/?p=debian/dpkg/dpkg.git;a=commitdiff;h=87740373> > > <http://git.hadrons.org/?p=debian/dpkg/dpkg.git;a=commitdiff;h=0484cd7d> > > But these do not cover fsync()s for /var/lib/dpkg/status and > /var/lib/dpkg/updates/*, do they? Those are very frequent as well and slow > the whole process down. I think you need to expand --force-unsafe-io scope.
Those fsync() have always been there, thus it cannot be part of any regression. I added few fsync()s for directories at the time, and for few other db files but nothing that amounts to much. The status file should be synced generally only once at the end of the run or when enough updates have been accumulated (currently 250), and each update file is fsync()ed once, those happen on state changes which should not be many per package and are pretty small files. If your file system is having problems keeping up with the db fsync()s then I'd urge you to complain to the file system developers, as I'd consider those buggy if they don't allow for the use of safe fsync() + rename() when they actually require it or make users suffer data loss. regards, guillem -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org