Re: postgres processes spending most of their time in the kernel

"Jeffrey W. Baker" <jwbaker@acm.org> writes:
> strace unfortunately doesn't deref the sembuf structure in semop.

Right, but the PC would be enough to let us tell whether semop was being
called from IpcSemaphoreLock or IpcSemaphoreUnlock.

However, if the info isn't available then no use crying over it.  What
we'll need to do instead is add some debugging printouts to lwlock.c,
and hope that the extra logfile I/O calls don't change the timing enough
to hide the misbehavior.  I'll think a little about this and send you
an updated lwlock.c to try out.

            regards, tom lane