Jeff Janes wrote: > The attached patch forces there to be at least one element in MCE, keeping > the one element with the highest predicted frequency if the MCE would > otherwise be empty. Then any other element queried for is assumed to be no > more common than this most common element.
Hmm, what happens if a common-but-not-an-MCE element is pruned out of the array when a bucket is filled? I imagine it's going to mis-estimate the selectivity (though I imagine the effect is going to be pretty benign anyway, I mean it's still going to be better than stock 0.5%.) > I'd also briefly considered just having the part of the code that pulls the > stats out of pg_stats interpret a MCE array as meaning that nothing is more > frequent than the threshold, but that would mean that that part of the code > needs to know about how the threshold is chosen, which just seems wrong. I wonder if we shouldn't add a separate stats STATISTIC_KIND for this, instead ot trying to transfer knowledge. Given how simple this patch is, I am tempted to apply it anyway. It needs a few additional comment to explain what is going on, though. -- Álvaro Herrera http://www.2ndQuadrant.com/ PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services -- Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-hackers