Reid Thompson <Reid(dot)Thompson(at)ateb(dot)com> writes:
> What am I missing that causes this to resort to sorting on disk?
The in-memory space required to sort N tuples can be significantly
larger than the on-disk space, because the latter representation is
optimized to be small and the in-memory representation not so much.
I haven't seen a 3X differential before, but it's not outside the realm
of reason, especially for narrow rows like these where it's all about
the overhead. I suspect if you crank work_mem up still more, you'll see
it switch over. It flips to on-disk sort when the in-memory
representation exceeds the limit ...
regards, tom lane