digoal(at)126(dot)com writes:
> But when use BERNOULLI sample method, it's return rows not the same as
> reltuples*factor.
> Can we add filter to exact return rows equal to reltuples*factor ?
No. It is neither practical nor required by the standard that the number
of returned rows be exactly N*S. If we tried to do that, rows near the
end of the table would end up having a different probability of being
selected than rows near the front.
regards, tom lane