Re: Why isn't my table auto-analyzed/vacuumed?

On Thu, Oct 30, 2025 at 11:55 AM Dimitrios Apostolou <jimis(at)gmx(dot)net> wrote:

> Hello list,
>
> I have a table that is constantly growing, and it's not being
> vacuumed/analyzed. I think my problem is rather common, but how to even
> debug it if "nothing works"?
>
> I've already set log_autovacuum_min_duration = 0 but the table is never
> mentioned in my logs, grep'ing for "vacuum".
>
> I have run ANALYZE manually once but nothing automatic.
> Here is more info:
>
> > SELECT * FROM pg_stat_user_tables WHERE relname =
> 'test_runs_summarized_per_function' \gx
> -[ RECORD 1 ]-------+----------------------------------
> relid | 780653
> schemaname | public
> relname | test_runs_summarized_per_function
> seq_scan | 32
> last_seq_scan | 2025-10-19 10:31:08.289922+00
> seq_tup_read | 26484817584
> idx_scan | 4554128
> last_idx_scan | 2025-10-10 22:02:50.987532+00
> idx_tup_fetch | 7418587674
> n_tup_ins | 921064234
> n_tup_upd | 0
> n_tup_del | 0
> n_tup_hot_upd | 0
> n_tup_newpage_upd | 0
> n_live_tup | 6484485348
> n_dead_tup | 0
> n_mod_since_analyze | 423101205
> n_ins_since_vacuum | 921064234
> last_vacuum |
> last_autovacuum |
> last_analyze | 2025-09-30 18:24:47.550543+00
> last_autoanalyze |
> vacuum_count | 0
> autovacuum_count | 0
> analyze_count | 1
> autoanalyze_count | 0
>
> > SELECT reltuples FROM pg_class WHERE relname =
> 'test_runs_summarized_per_function' \gx
> -[ RECORD 1 ]-----------
> reltuples | 6.061923e+09
>
> > SELECT name,setting FROM pg_settings WHERE name ILIKE '%factor%' ;
> name | setting
> ---------------------------------------+---------
> autovacuum_analyze_scale_factor | 0.1
>

0.1 means 10%.

> autovacuum_vacuum_insert_scale_factor | 0.2
> autovacuum_vacuum_scale_factor | 0.2
> recursive_worktable_factor | 10
>

n_mod_since_analyze=423101205
n_live_tup=6484485348

n_mod_since_analyze/n_live_tup = 6.5%

> How can I get more info from postgres on the autovacuum logic?
>

I would:
1) manually VACUUM ANALYZE the table,
2) drop the three autovacuum_*_scale_factor values down to 0.03 (i.e. 3%),
3) reload the conf file,
4) add "(1.0*n_mod_since_analyze/n_live_tup)::decimal(6,3)" to the
pg_stat_user_tables query, and
4) closely monitor pg_stat_user_tables WHERE relname =
'test_runs_summarized_per_function'.

https://www.percona.com/blog/importance-of-postgresql-vacuum-tuning-and-custom-scheduled-vacuum-job/
helped me a lot. It also validated my cron job that does "manual" ANALYZE
& VACUUM on tables that autovacuum isn't picking up, even though it seems
like it should.

--
Death to <Redacted>, and butter sauce.
Don't boil me, I'm still alive.
<Redacted> lobster!