Becky,

I am not sure we ever spent time on a test case that could trigger the
problem outside of our environment. We found that the mtime field on some
files was being updated when it should not have been. That led us to
discovering that the pinode_flags were not getting cleared when a new inode
was allocated, so there is a bit of random luck involved.

A pvfs2_inode structure has to be allocated and used in a way that the
pinode_flags indicates the mtime needs to be updated. Once that structure is
returned to the cache, it has to be allocated again for an operation that
doesn't want the mtime field updated. We did not see this occur
epidemically, but it was often enough to be noticed by users and caused some
issues with processes that check that field.

I know that is not terribly helpful for testing the changes. We have been
using them since they were committed over a year ago, and it has effectively
resolved out issue without any perceived side-effects. I imagine as core a
component as the inode cache is, a good round of LTP tests and any other
regression tests you have lying around would quickly identify any problems
with the changes.

Bart.


On Fri, Jun 10, 2011 at 4:09 PM, Becky Ligon <[email protected]> wrote:

> Bart:
>
> I am ready to update OrangeFS with the changes below.  Can you tell me the
> best way to test this?
>
> Becky
> --
> Becky Ligon
> HPC Admin Staff
> PVFS/OrangeFS Developer
> Clemson University/Omnibond.com OrangeFS Support
> 864-650-4065
>
> > Can someone please check on a patch that was accepted April 14, 2010
> > regarding allocation and initialization of inode structures from the slab
> > cache? The final email in the chain with applicable details can be found
> > here:
> >
> >
> http://www.beowulf-underground.org/pipermail/pvfs2-users/2010-April/003097.html
> >
> > A link to the changeset can be found here:
> >
> > http://www.pvfs.org/fisheye/changelog/PVFS?cs=MAIN:pcarns:20100414181405
> >
> > It looks like the Orange-Branch may have been created prior to these
> > changes
> > being committed.
> >
> > Thanks,
> > Bart.
> > _______________________________________________
> > Pvfs2-developers mailing list
> > [email protected]
> > http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers
> >
>
>
>
_______________________________________________
Pvfs2-developers mailing list
[email protected]
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers

Reply via email to