Re: To what extent should tests rely on VACUUM ANALYZE?

29.03.2024 11:59, Alexander Lakhin wrote:
>
> But it looks like subselect is not the only test that can fail due to
> vacuum instability. I see that create_index also suffers from cranky
> ConditionalLockBufferForCleanup() (+if (rand() % 10 == 0)
> return false; ), although it placed in parallel_schedule before
> sanity_check, so this failure needs another explanation:
> -                      QUERY PLAN
> --------------------------------------------------------
> - Index Only Scan using tenk1_thous_tenthous on tenk1
> -   Index Cond: (thousand < 2)
> -   Filter: (tenthous = ANY ('{1001,3000}'::integer[]))
> -(3 rows)
> +                                      QUERY PLAN
> +--------------------------------------------------------------------------------------
> + Sort
> +   Sort Key: thousand
> +   ->  Index Only Scan using tenk1_thous_tenthous on tenk1
> +         Index Cond: ((thousand < 2) AND (tenthous = ANY ('{1001,3000}'::integer[])))
> +(4 rows)

I think that deviation can be explained by the fact that cost_index() takes
baserel->allvisfrac (derived from pg_class.relallvisible) into account for
the index-only-scan case, and I see the following difference when a test
run fails:
         relname        | relpages | reltuples | relallvisible | indisvalid | autovacuum_count | autoanalyze_count
  ----------------------+----------+-----------+---------------+------------+------------------+-------------------
- tenk1                |      345 |     10000 |           345 |            |                0 |                 0
+ tenk1                |      345 |     10000 |           305 |            |                0 |                 0

Best regards,
Alexander