Re: Index on System Table

That's awesome, thanks! Yeah, I doubt I'll do that to our production
database, but maybe I'll try it on a copy sometime down the line.
Adjusting the cost for pg_table_is_visible is working well enough so
far.

Cody Cutrer

On Wed, Mar 21, 2012 at 12:17 PM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> Cody Cutrer <cody(at)instructure(dot)com> writes:
>> On Tue, Mar 20, 2012 at 6:06 PM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>>> There's not really support for adding indexes to system catalogs
>>> on-the-fly.  I think it would work (barring concurrency issues)
>>> for most catalogs, but pg_class has special limitations due to
>>> the "relmapping" infrastructure.  It's not something I'd particularly
>>> care to try on a production database.
>
> BTW, I experimented with that a little bit and found that the relmapper
> is not really the stumbling block, at least not after applying this
> one-line patch:
> http://git.postgresql.org/gitweb/?p=postgresql.git;a=commitdiff;h=f70f095c9096d5e2689e8d79172b37b57a84e51b
>
> It seemed to work for me to do this:
>
>        stop postmaster
>        start a standalone backend with -c allow_system_table_mods=1
>        create index pg_class_relnamespace_index on pg_class(relnamespace);
>        stop standalone backend, restart postmaster
>
> There are a lot of gotchas here, notably that the session in which you
> create the index won't know it's there (so in this case, a reindex on
> pg_class would likely be advisable afterwards).  I still think you'd be
> nuts to try it on a production database, but ...
>
>                        regards, tom lane