Kris Jurka <books@ejurka.com> writes:
> What happens if you add a sleep 5 call in between the createdb calls?
Even a fractional-second sleep would likely do it. I have traced
through scenarios like this in the past, and AFAICT the exiting backend
is not doing anything wrong or even using an unreasonable amount of
CPU time before it quits. It simply isn't getting scheduled to run
because for one reason or another the kernel gives time slices to the
foreground clients and the new backend first.
regards, tom lane