Matthew Wakeling <matthew(at)flymine(dot)org> writes:
> Ah, now I see it - I re-analysed, and found entries in pg_stats where
> tablename is the name of the index. Now the query plans correctly and has
> the right estimates. So, one needs to analyse AFTER creating indexes -
> didn't know that.
Yes, for functional indexes it's helpful to do that. Doesn't matter
for plain-old-plain-old indexes.
regards, tom lane