On Mon, 2024-03-11 at 12:36 +0000, Tefft, Michael J wrote:
> I have been struggling to set n_distinct on a few table columns, and confirm that my changes have actually been
accepted.
>
> I have a 400-million row table with 81 partitions. PostgreSQL version is 14.11.
> Column p_id has 13 million distinct values but pg_stats says n_distinct is only 82k.
> Column pi_id has 18 million distinct values but pg_stats says n_distinct is only 29k.
> Column i_id has 70k distinct values but pg_stats says n_distinct is only 1800.
>
> I tried:
> alter table xxx alter column p_id set (n_distinct=-0.033 );
> alter table xxx alter column pi_id set (n_distinct=-0.045 );
> alter table xxx alter column i_id set (n_distinct=-0.0002 );
>
> All ran without error, but pg_stats shows no changes. I tried:
>
> alter table xxx alter column p_id set (n_distinct=13118955 );
> alter table xxx alter column pi_id set (n_distinct=18059179 );
> alter table xxx alter column i_id set (n_distinct=69911 );
>
> Again, all ran without error, but pg_stats shows no changes.
>
> I saw somewhere (but can’t locate the reference today) that setting n_distinct takes effect after the next ANALYZE. I
triedANALYZE xxx and VACUUM ANALYZE xxx but the settings never appeared to take effect.
>
> I tried increasing “statistics” on the columns to 1000 and running ANALYZE; this produced somewhat-higher n_distinct
valuesbut still far from accurate and my manually-set values still did not appear.
>
> How can I get these values to take effect?
If it is a partitioned table, set "n_distinct_inherited" on the column,
then run ANALYZE.
Yours,
Laurenz Albe