- - <loh.law@hotmail.com> writes:
> The weird thing is that before I updated my server the query was about 5 times faster.
> I've googled and I think the problem lies with the under-estimation of the query planner about the number of rows in
thenested table.I will be trying the 'set enable_seqscan = false' solution to see if that'll improve.
You evidently already do have that turned off. I'd suggest reverting
that change (ie, allow seqscan) and instead increase work_mem enough
so that the hash join can work without spilling to disk. This query
is a perfect example of where indexes do not help, and trying to force
them to be used makes things slower not faster.
regards, tom lane