Neil Conway <neilc@samurai.com> writes:
> Interesting. The inconsistency you're seeing is a result of GEQO. I
> would have hoped that it would have produced a better quality plan
> more often, but apparently not. On my system, the regular query
> optimizer handily beats GEQO for this query: it produces more
> efficienty query plans 100% of the time and takes less time to do so.
> For *this* query at least, raising geqo_threshold would be a good
> idea, but that may not be true universally.
The current GEQO threshold was set some time ago; since then, the
regular optimizer has been improved while the GEQO code hasn't been
touched. It might well be time to ratchet up the threshold.
Anyone care to do some additional experiments?
regards, tom lane