On Tue, May 22, 2012 at 03:52:14PM +0900, Jun'ichi Nomura wrote:
<SNIP>
Thank you for the comments.
Unfortunately, I've seen it only a few times in a year...
I wonder whether those functions using umask are called
from threads (or signal handlers).
If they do, it would certainly cause such a problem
in the same way as the following reproducer.
<SNIP>
Hi,
We've also ran into this same issue on 3 different clusters this year,
on a couple of occasions. I can also confirm that we've not been able
to reproduce it; it's just something we've reacted to when it did hit
us.
My hunch was that it was use of hardlinks, but your analysis seems
more logical.
Thanks.
Kind Regards,
Dave Walker