Re: [RFC] Removing "magic" oids

On Fri, Jul 19, 2019 at 10:12:57AM -0700, Andres Freund wrote:
> On 2019-07-07 10:00:35 -0700, Noah Misch wrote:
> > +# Test concurrent OID generation via pg_enum_oid_index. This indirectly
> > +# exercises LWLock and spinlock concurrency.
> > +my $labels = join ',', map { "'l$_'" } 1 .. 1000;
> > pgbench(
> > '--no-vacuum --client=5 --protocol=prepared --transactions=25',
> > 0,
> > [qr{processed: 125/125}],
> > [qr{^$}],
> > - 'concurrent insert workload',
> > + 'concurrent OID generation',
> > {
> > '001_pgbench_concurrent_insert' =>
> > - 'INSERT INTO insert_tbl SELECT FROM generate_series(1,1000);'
> > + "CREATE TYPE pg_temp.e AS ENUM ($labels); DROP TYPE pg_temp.e;"
> > });
>
> Hm, perhaps we should just do something stupid an insert into a catalog
> table, determining the oid to insert with pg_nextoid? That ought to be a
> lot faster and thus more "stress testing" than going through a full
> blown DDL statement? But perhaps that's just too ugly.

I expect the pg_nextoid strategy could have sufficed. The ENUM strategy
wastes some time parsing 1000 label names, discarding odd-numbered OIDs, and
dropping the type. The pg_nextoid strategy wastes time by performing the
insertion loop in the executor instead of dedicated C code of
EnumValuesCreate(). Hard to say how to weight those factors.