> I don't think planner should do things like creating an index. But it
> might hint at doing it in the logs.
> There was a discussion around that sort of feature on -hackers not so
> long time ago. I don't remember what the conclusion was, but probably
> that it just isn't worth wasting planner's cycles when you execute a
> query.
Yes, that would be a much better solution, definitely. Something like
'watch seq-scan nodes and if the number of matching rows is much smaller
than the total number, put a HINT into log.'
I was thinking about doing something like this for correlated columns
(when the number of matching rows is severely underestimated). But that's
far in the future.
Tomas