On Fri, Sep 27, 2019 at 3:52 PM Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> n_live_tup/n_dead_tup should not be thought to be better than
> approximations. Most operations adjust them only incrementally,
> and messages to the stats collector can get dropped entirely
> under heavy load, causing deltas to go missing. There are
> also race conditions involved in some update scenarios.
>
Are you suggesting tha autovacuum made free space available so that
pg_stat_user_tables lags behind? Because I don't see last_autovacuum
changing in the mantime.
Luca