Hi all, If we create a table with vacuum_index_cleanup = off or execute VACUUM with INDEX_CLEANUP = off, vacuum updates pg_stat_all_tables.n_dead_tup to the number of HEAPTUPLE_RECENTLY_DEAD tuples. Whereas analyze updates it to the sum of the number of HEAPTUPLE_DEAD/RECENTLY_DEAD tuples and LP_DEAD line pointers. So if the table has many LP_DEAD line pointers due to skipping index cleanup, autovacuum is triggered every time after analyze/autoanalyze. This issue seems to happen also on back branches, probably from 12 where INDEX_CLEANUP option was introduced.
I think we can have heapam_scan_analyze_next_tuple() not count LP_DEAD line pointer as lazy_scan_prune() does. Attached the patch for that. Regards, -- Masahiko Sawada EDB: https://www.enterprisedb.com/
not_count_lp_dead_by_analyze.patch
Description: Binary data
