On Wed, Aug 29, 2018 at 2:09 AM Robert Haas <robertmh...@gmail.com> wrote: > On Mon, Aug 27, 2018 at 10:12 PM, Andres Freund <and...@anarazel.de> wrote: > > Huh? Oids between, say, 1 and FirstNormalObjectId, are vastly more > > common than the rest. And even after that, individual tables get large > > clusters of sequential values to the global oid counter. > > Sure, but if you get a large cluster of sequential values, a straight > mod-N bucket mapping works just fine. I think the bigger problem is > that you might get a large cluster of values separated by exactly a > power of 2. For instance, say you have one serial column and one > index: > > rhaas=# create table a (x serial primary key); > CREATE TABLE > rhaas=# create table b (x serial primary key); > CREATE TABLE > rhaas=# select 'a'::regclass::oid, 'b'::regclass::oid; > oid | oid > -------+------- > 16422 | 16430 > (1 row) > > If you have a lot of tables like that, bad things are going to happen > to your hash table.
Right. I suppose that might happen accidentally when creating a lot of partitions. Advance the OID generator by some prime number after every CREATE TABLE? /me ducks -- Thomas Munro http://www.enterprisedb.com
