On Tuesday 25 March 2008, David Chinner wrote:
> That points to I/O not completing (not an XFS problem at all), or
> the filesystem freeze is just taking a long time to run (as it has
> to sync everything to disk). Given that this is a snapshot target,
> writing new blocks will take quite some time. Is the system still
> making writeback progress when in this state, or is it really hung?
That is real deadlock, operations which normally takes few minutes,
hangs for few hours until I restart machine.
This bug is very strange because I can't catch it with more verbose
debug and tracing options,
I did more tests without debugging options and looks that version from
sgi cvs (2.6.25-rc3) works well. I did also test with two patches you suggest,
applied (with some troubles) for 2.6.24.2, but finally things hung after
2 days - better than without patches, as you expected.
So, looks things are fixed in sgi cvs repository. Don't know if these are
xfs or other subsystem (device-mapper?) fixes, anyway I can't reproduce
bug on this version, that's good news, I think.
>
> Cheers,
>
> Dave.
Thanks
Stanislaw Gruszka
|