Andrew Gierth <rhodiumtoad(at)postgresql(dot)org> writes:
> Avoid having vacuum set reltuples to 0 on non-empty relations in the
> presence of page pins, which leads to serious estimation errors in the
> planner.
Hm, buildfarm results suggest this test is not entirely stable:
https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=crake&dt=2017-03-17%2000%3A47%3A57
I have not looked very closely, but I'm suspicious that the test case
depends on no autovacuum transactions running concurrently with it.
Disabling autovac on the table itself is not enough to control whether
global xmin is being held back by some other autovac transaction
somewhere.
regards, tom lane