On Fri, Apr 3, 2020 at 11:50 AM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> (2) It only wins if a statement timeout is active, otherwise it makes
> things worse, because then you need setitimer() at statement start
> and end just to enable/disable the socket check timeout. Whereas
> if you just let a once-a-minute timeout continue to run, you don't
> incur those kernel calls.
Oh, that's a really good point. I should have thought of that.
> Anyway, the core problem with the originally-submitted patch was that
> it was totally ignorant that timeout.c had restrictions it was breaking.
> You can either fix the restrictions, or you can try to design around them,
> but you've got to be aware of what that code can and can't do today.
No disagreement there.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company