[rt-users] UseSQLForACLChecks / CurrentUserCanSee

[ktm at rice.edu]

On Fri, Jan 25, 2013 at 01:08:30AM +0400, Ruslan Zakirov wrote:
> On Fri, Jan 25, 2013 at 12:57 AM, ktm at rice.edu <ktm at rice.edu> wrote:
> > Dear RT community,
> >
> > I have been testing the option UseSQLForACLChecks which is based on the
> > function CurrentUserCanSee and the initial testing is disappointing. Here
> > are 3 representative front page loads with and without SQL-checks enabled:
> >
> > with-SQL: 16s, 13.8s, 14.5s --- 14.7s average
> > no-SQL: 2.3s, 0.99s, 0.97s --- 1.42s average
> >
> > That is a 10x slowdown and it does not get any better with reloads. In
> > some of the messages on the mailing list, it is mentioned that the performance
> > is "not bad" for some setups. Is there a reference setup that is defined that
> > produces these "not bad" timings? Alternatively, does anyone have a setup
> > that is working well or that has an appropriate cache layer setup?
> >
> 
> Hello Ken,
> 
> You probably read description of the option, right? I always suspected
> that there
> will be instances for which performance degrades dramatically because
> of the option. So far we had one case when option works good for core
> RT and behaves badly in a proprietary extension. For most of our
> customers it works great.
> 
> A few changes have been made since the option was introduced to fix
> ACL problems and performance issues, so as always you should mention
> RT version.
> 
> We would love to see more details on your case. The option will be
> turned on by default in RT 4.2.
> 
> > Regards,
> > Ken
> 
> -- 
> Best regards, Ruslan.
> 

Hi Ruslan,

We are running version 3.8.13 with a PostgreSQL 9.2 backend. I will
try to look into it further and see if I can locate the cause of the
pathological performance issues.

Regards,
Ken