Re: Further thoughts about warning for costly FK checks

Fabien COELHO wrote:
>
> > > Or maybe some new "ADVICE" or "ODDITY" level next to "DEBUG NOTICE
> > > WARNING ERROR PANIC...", that would be hidden by default and triggered
> > > by an option?
> >
> > But that doesn't really solve the problem, which is that there are
> > conditions that it's difficult to test for on-the-fly while a schema is
> > being constructed.
>
> Ok, I was off my shoes again, sorry.
>
> > A pg_advisor tool would look at the completed schema and would not have
> > that problem.
>
> Ok.
>
> As a more interesting contribution to the issue, I would suggest that such
> checks could be triggered by "ANALYSE".
>
> 1/ it is the first thing to be done on performance problems.
>
> 2/ it is not done too often.
>
> 3/ checking this stuff seems to belong to the analyse phase.
> warnings about costly foreign key checks could also be repeated
> on these occasions.
>
> Also, because of the opposition by some DBA, these checks could be disable
> by some options, but I would suggest the option to be on by default.

I was thinking of a GUC variable called PERFORMANCE_HINTS, which would
throw a message if a lookup from the primary to the foreign key didn't
have an index.

--
Bruce Momjian | http://candle.pha.pa.us
pgman(at)candle(dot)pha(dot)pa(dot)us | (610) 359-1001
+ If your life is a hard drive, | 13 Roberts Road
+ Christ can be your backup. | Newtown Square, Pennsylvania 19073