Shame on me. It's a partial index - *where is not null.* Put the* is not null *predicate in place and planner always goes for index. (tested with thousands of IN entries) CTE version always goes for index too even *without **is not null , *which led to a slight confusion.
Thanks Tom, Michael, Best Rado On Wed, Jul 15, 2020 at 1:06 AM Tom Lane <t...@sss.pgh.pa.us> wrote: > Radoslav Nedyalkov <rnedyal...@gmail.com> writes: > > Ah, I could have messed up the examples I gave. Row numbers are > different. > > Once again the plans , sorry about that. > > Given that it works at 100 entries and not 101, I can't escape the > suspicion that you're being burnt by predtest.c's MAX_SAOP_ARRAY_SIZE > limit. However, that only affects the planner's willingness to make > constraint proofs involving the large IN clause, and nothing you've > mentioned here explains why such a proof would be needed. Is there > something you're not telling us about this table's schema? (I'm > wondering if the index is partial, for instance, though one would > think that the CTE form of the query wouldn't work either if so.) > > regards, tom lane >
