On Jan 7, 2009, at 12:30 PM, Tom Lane wrote:
Kirk Strauser <
kirk@strauser.com> writes:
First, shmget is asking for a lot less than shmmax - why is it failing?
Check to see if things work as expected when you have shmmax and shmall
set to a shade less than 2GB and fail when they are a shade more. If
so, it would seem there's a signed-integer-overflow bug somewhere in the
kernel's handling of shmem requests ... which would be a reportable
kernel bug.
Actually, it seems that if shmget is trying to get more than 2GB, it will fail regardless of shmmax. When I backed shared_buffers down to 1980MB, PostgreSQL was able to start as long as shmmax is >= 2GB (which I verified with 2GB+1MB, 3GB, and 8GB). I'm off to file a report now.