Greg Stark <gsstark@mit.edu> writes:
> On Thu, Jul 16, 2009 at 4:32 PM, Tom Lane<tgl@sss.pgh.pa.us> wrote:
>> So maybe a redesign of the equivalence-class joinclause mechanism is in
>> order. �Still, this is unlikely to fix the fundamental issue that the
>> time for large join problems grows nonlinearly.
> Perhaps it's GEQO's fault that it's using these functions
> inappropriately, calling them often to calculate these answers
> whenever it needs them instead of looking once for join clauses and
> then optimizing based on the results. But I've never actually looked
> at geqo, mabe that's inherent in the design?
geqo isn't doing anything the regular planner wouldn't do under similar
conditions. It might well be that better caching is the answer to this
particular problem, but I don't have time to look closer today.
regards, tom lane