"Jim C. Nasby" <jnasby(at)pervasive(dot)com> writes:
> What's with the second workMem setting?
That's the second sort bucket for recently-dead tuples (which still have
to be indexed, but shouldn't be fed into the unique-checking sort).
We assume there won't be a lot of them, hence don't use
maintenance_work_mem to size that sort.
regards, tom lane