Re: 2.6.39.1: (2.6.35+ probably): delaylog bug
From: Dave Chinner <david@fromorbit.com>
Date: 2011-07-04 00:47:33
On Sun, Jul 03, 2011 at 05:24:09PM -0400, Justin Piszcz wrote:
Hello, I help to support a system (similar to mine) with a 3ware 9750-24i4e and when you rm -rf a directory with too many files, the system will act lagged, lock up for periods of time and drop connections (including samba shares) to and from the host. Its so bad it cannot even talk to the UPS for example (ups: communications lost, e.g. from nut). This is pretty bad and I've remounted my XFS FS w/out delaylog but for the xfstests package, it really needs to test with a few million files in a directory or two, and you should easily be able to reproduce the problem.
http://oss.sgi.com/archives/xfs/2011-06/msg00259.html FYI, the problem is not easy to reproduce - it requires a specific set of conditions to exist with interrupt configuration (i.e. all on the same CPU that the log IO completions are running) to expose the workqueue priority inversion that the filesystem level issue triggers.
When the delaylog option is removed, the files are removed slowly, but there are no errors, SSH/samba connections do NOT drop, etc, please look into this, thanks.
http://oss.sgi.com/archives/xfs/2011-06/msg00463.html http://oss.sgi.com/archives/xfs/2011-06/msg00473.html Cheers, Dave. -- Dave Chinner david@fromorbit.com _______________________________________________ xfs mailing list xfs@oss.sgi.com http://oss.sgi.com/mailman/listinfo/xfs