Simon Riggs <simon(at)2ndquadrant(dot)com> writes:
> Any holes in that thinking?
Only that it's about five times more complicated than is currently known
to be necessary ;-). How about we just implement the dynamic spill to
disk first, and not bother with the other stuff until we see problems in
the field? Saying "we have to do all this" is a good recipe for not
getting any of it done.
regards, tom lane