XFS - issues with writes using sync

[Amit Sahrawat]

On Thu, Jan 20, 2011 at 12:11 PM, Dave Chinner <david at fromorbit.com> wrote:

> On Thu, Jan 20, 2011 at 11:37:44AM +0530, Amit Sahrawat wrote:
> > Hi,
> >
> > I will try to find out the cause for this.
> > Meanwhile, just a small request/suggestion - in the past this type of
> > testcases have helped us in finding many problems in XFS.
> > Can something like this be added to xfstests? This might help.
>
> Definitely.  We're always looking for more people to add tests that
> expose problems to xfstests. :) We try to keep individual test
> runtime to as little as possible - under 5 minutes for the auto
> group, under 15s for the quick group, but by the looks of it the
> test you are running doesn't take that long to run.
>


> >> Yes, the test case we are using is not taking much time this time to
> cause issue - around 3-4 minutes.
>


> FWIW, there are already tests that cause worst case filesystem
> fragmentation as part of their test setup (e.g. test 042) but the
> coverage of such issues could definitely be improved. Also, the way
> we generate fragmented filesystems - by writing files and then
> removing a subset - could be greatly sped up by preallocation and
> hole punching. There's no need to write data when we could just use
> unwritten extents to do the same thing...
>
>


>  >> We basically fragment by doing following steps:
>
 Full the disk with same size files (in this case it is 16k)
Then, randomly remove files from the disk.
Doing so, allows us to create files as per our requirement - complete
control when we need - single extent, multiple extent files, Btree format
file, single leaf file, multiple leaf file - It allows us ease to use XFS.
I hope you are getting what I am trying to say.

But, in this case - it starts causing issue at the very first stage - just
copying 16k file to make disk full.

Thanks,
Amit Sahrawat
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://oss.sgi.com/pipermail/xfs/attachments/20110120/398a6a8d/attachment.htm>