Re: Cache invalidation after authentication (on-the-fly role creation)

On Wed, Jul 4, 2018 at 12:10 PM, Andres Freund <andres@anarazel.de> wrote:
> On 2018-07-03 19:44:21 -0400, Alvaro Herrera wrote:
>> On 2018-Jul-04, Thomas Munro wrote:
>> >                 PerformAuthentication(MyProcPort);
>> > +               AcceptInvalidationMessages();
>> >                 InitializeSessionUserId(username, useroid);
>> >
>> > Any objections?
>>
>> Is there a measurable performance overhead to this change?
>
> I can't see it being relevant here. We accept inval message in plenty
> other places, and in comparison to session startup it should be just
> about inmeasurable.

Yeah, using "pgbench -c 8 -j 8 -T 60 --connect -S -M prepared
postgres" I wasn't able to measure a significant difference on my
laptop.  The performance was equally terrible, at around 940 TPS +/-
10 including connection time.  Adding to open commitfest.

-- 
Thomas Munro
http://www.enterprisedb.com