Michael Teter <michael_teter(at)yahoo(dot)com> writes:
> Perhaps there's some issue with the shared memory not
> being freed quickly enough by the kernel?
Never heard of that before. There is an issue with time delays
on release of port addresses --- sometimes you may get a failure
to bind() to a port address that was just released by a previous
incarnation of the postmaster. But I haven't heard of such a
thing for shared memory or semaphores.
regards, tom lane