Le jeu. 30 déc. 2021 à 17:25, Tom Lane <t...@sss.pgh.pa.us> a écrit :
> Japin Li <japi...@hotmail.com> writes: > > On Thu, 30 Dec 2021 at 18:53, Guillaume Lelarge <guilla...@lelarge.info> > wrote: > >> pg_dump works in a single transaction, so it's already dealt with > >> idle_in_transaction_timeout. Though I guess setting both would work too. > > > Attached fix this, please consider reveiew it. Thanks. > > This seems rather pointless to me. The idle-session timeout is only > activated in PostgresMain's input loop, so it will never be reached > in autovacuum or other background workers. (The same is true for > idle_in_transaction_session_timeout, so the fact that somebody made > autovacuum.c clear that looks like cargo-cult programming from here, > not useful code.) And as for pg_dump, how would it ever trigger the > timeout? It's not going to sit there thinking, especially not > outside a transaction. > > Agreed. It makes more sense. So no need for the patch. Thanks to both. -- Guillaume.