On Mon, Jan 7, 2019 at 10:29 PM Bartosz Polnik <bartoszpolnik@gmail.com> wrote:
>
> > Can you confirm if there are concurrent changes being made to table_b
> > perhaps while the query is running?
>
> I confirm. There are no concurrent changes being made and I'm the only client connecting to that db.
>
> > Do you still get the variation with an otherwise idle database with no
> > open transactions that's just received a complete VACUUM?
>
> I executed VACUUM (FULL, ANALYZE) on all tables from the query, but I'm still getting different results.
>
> With your script, I'm always getting the same output, so it must be something else.
What do you get if you run it with SET force_parallel_mode = on?
--
Thomas Munro
http://www.enterprisedb.com