Bruno Wolff III <bruno@wolff.to> writes:
> After doing an initdb I got the expected plan.
Hm. I'm not sure what happened there --- I don't recall that we made
any initdb-needing changes in the past week or so. (If we did, we
should have forced initdb by incrementing catversion, but sometimes
people forget to do that.) The only change I can think of that's
related at all is that the outer query's "group by foo order by foo desc"
should now only require one sort step not two (which is probably why
my test went for the Sort/GroupAggregate plan not the HashAgg/Sort
plan you showed). But that shouldn't have affected the plan for the
inner SELECT DISTINCT query, AFAICS. Odd.
Probably not worth spending time on though.
regards, tom lane