digoal(at)126(dot)com writes:
> i see postgresql optimizer don't add column's indexCorrelation in index scan
> and then shold not choose optimal scannode.
When I try this test case, I find that the planner prefers a plain
seqscan, then a bitmap scan, and last an indexscan; and the cost estimates
are not too out of line with reality. I suspect you've changed the
planner's cost parameters to some non-default settings that don't really
square very well with your environment.
regards, tom lane