Alvaro Herrera <alvherre@commandprompt.com> writes:
> What I don't understand is what you mean with it "obliterating" the
> stats for a table.
The point seems to be that if there is no pg_statistic data for these
tables, we fall back to default estimates of the selectivity of the
WHERE clauses, and those produce rowcount estimates that are pretty
far away from the truth. As soon as some pg_statistic entries
exist, we obtain better rowcount estimates, and that changes the join
plan.
regards, tom lane