On Thu, Mar 17, 2005 at 08:14:13PM -0600, James Foris wrote:
>
> Here is a script (developed by Rich Coe) that demonstrates the file
> corruption problem found on our systems.
>
Just a quick update here, eecause some people are concerned.
I've been running James' script without hitting the problem
on a CVS kernel with the same MD setup. Doesn't mean there
isn't a problem, of course, but seems non-trivial to hit it
at least.
Discussing with James off-list, he's also found that using
the -onoalign mount option completely removes the problem
at his end, and he even sees better performance with that,
oddly enough. So, if you are seeing the same problem, you
might find that workaround useful.
James also sees the problem go away with debugging enabled,
which suggests some kind of small race. I'll keep looking
into it as time permits. Actually, James, could you double
check that the problem persists with the CVS 2.6.12-based
kernel, please? Couple of XFS fixes on the IO path there,
perhaps they'll help (perhaps thats why I'm not seeing it).
cheers.
--
Nathan
|