On Tue, Jun 30, 2020 at 12:23:28PM +0900, Masahiko Sawada wrote:
> > I propose to have a new session level GUC called
> > "enable_connectioncache"(name can be changed if it doesn't correctly
> > mean the purpose) with the default value being true which means that
> > all the remote connections are cached. If set to false, the
> > connections are not cached and so are remote sessions closed by the local backend/session at
> > the end of each remote transaction.
>
> I've not looked at your patch deeply but if this problem is talking
> only about postgres_fdw I think we should improve postgres_fdw, not
> adding a GUC to the core. It’s not that all FDW plugins use connection
> cache and postgres_fdw’s connection cache is implemented within
> postgres_fdw, I think we should focus on improving postgres_fdw. I
> also think it’s not a good design that the core manages connections to
> remote servers connected via FDW. I wonder if we can add a
> postgres_fdw option for this purpose, say keep_connection [on|off].
> That way, we can set it per server so that remote connections to the
> particular server don’t remain idle.
I thought we would add a core capability, idle_session_timeout, which
would disconnect idle sessions, and the postgres_fdw would use that. We
have already had requests for idle_session_timeout, but avoided it
because it seemed better to tell people to monitor pg_stat_activity and
terminate sessions that way, but now that postgres_fdw needs it too,
there might be enough of a requirement to add it.
--
Bruce Momjian <bruce@momjian.us> https://momjian.us
EnterpriseDB https://enterprisedb.com
The usefulness of a cup is in its emptiness, Bruce Lee