Hi, On 2025-03-24 02:26:35 +0900, Fujii Masao wrote: > > With autovacuum = off, all of these fluctuations go away. > > So you are suggesting disabling autovacuum during the postgres_fdw regression > test?
I don't think that's a good fix in this case. > Just my idea to stabilize the test with "RETURNING *" is to use WITH, like > this: > > WITH tmp AS (UPDATE ... RETURNING *) SELECT * FROM tmp ORDER BY ... +1 On 2025-03-23 16:00:00 +0200, Alexander Lakhin wrote: > Interestingly enough, with "log_autovacuum_min_duration = 0" added to > TEMP_CONFIG, I can't see "automatic vacuum/analyze" messages related > to ft2/ "S 1"."T 1", so autovacuum somehow affects contents of this table > indirectly. > > With autovacuum = off, all of these fluctuations go away. Hm. Is it possible that autovacuum started but didn't finish? FWIW, I observed some oddities around autovacuum=0 recently, but didn't have time to track the source down yet. E.g. running pg_trgm's tests against a server with autovacuum=0 results in test failures due to plan changes, despite no AVs getting launched. Greetings, Andres
