Bruno Baguette <bruno.baguette@gmail.com> writes:
> Le 13/11/08 14:31, Tom Lane a �crit :
>> 16.511 * 818 = 13505.998, so this is all but about 100 msec of the
>> runtime. Can't tell if there's any easy way to improve it. In
>> pre-8.4 releases trying to convert the EXISTS into an IN might help.
> Can you explain why a IN is fastest than an EXISTS subquery ?
The planner is smarter about IN than EXISTS --- it can usually convert
the former into a join plan instead of a subplan. (This situation will
improve in 8.4.)
> Do you think I can improve again the performance of that query ?
You've still got a subplan in there, not quite sure why. Anyway,
increasing work_mem might get it to change to a hashed subplan,
which'd likely be faster.
regards, tom lane