> > I'd say this whole mechanism is unnecessary; we should just build
> > the data on-the-fly the way it's done in write_irels(), and eliminate
> > all the file reading and writing code in init_irels and write_irels.
> > The only thing it could possibly be doing for us is saving some backend
> > startup time, but I'm not able to measure any difference when I delete
> > the init file.
> >
> > I'll work on that tomorrow, unless I hear squawks of outrage from
> > someone who remembers what this code was for.
>
> Hmm. If you can get it to work without the file, great, or you could
> just delete the file when vacuum is performed, so the next backend
> recreates the file. That would work too.
One other cache thing I want to do is enable index lookups on cache
failure for tables like pg_operator, that didn't use index lookups in
the cache because you didn't have multi-key system tables.
If you are looking for a single row, does anyone know when it is faster
to do a sequential scan, and when it is faster to do an index lookup
into the table.
-- Bruce Momjian | http://www.op.net/~candle maillist@candle.pha.pa.us | (610)
853-3000+ If your life is a hard drive, | 830 Blythe Avenue + Christ can be your backup. | Drexel Hill,
Pennsylvania19026
