Palle Girgensohn <girgen@pingpong.net> writes:
> Trying all this out, I realize that on 7.4.5, I can sometimes get different
> results after `vacuum analyze' vs. a plain `analyze' (again, not exactly
> the same data, and I cannot reproduce this on the other machine with the
> data I sent you). It does not really relate to the question above, but
> perhaps you can explain how come I get different results?
No surprise. vacuum analyze produces an exact total row count, whereas
analyze can only produce an approximate total row count (since it only
samples the table rather than groveling over every row). Sometimes the
approximate count will be far enough off to affect the estimates.
regards, tom lane