Path: utzoo!attcan!uunet!tut.cis.ohio-state.edu!ucbvax!pasteur!eden!mao
From: mao@eden (Mike Olson)
Newsgroups: comp.databases
Subject: Re: Postgres (was Re: Public Domain Databases)
Keywords: db data base database postgres relational
Message-ID: <27803@pasteur.Berkeley.EDU>
Date: 14 Sep 90 14:14:10 GMT
References: <556@mcspdx.pdx.csd.mot.com> <27728@pasteur.Berkeley.EDU> <1990Sep14.065746.5711@tukki.jyu.fi>
Sender: news@pasteur.Berkeley.EDU
Reply-To: mao@postgres.Berkeley.EDU (Mike Olson)
Organization: University of California, Berkeley
Lines: 43


In <1990Sep14.065746.5711@tukki.jyu.fi>, sakkinen@jytko.jyu.fi (Markku
Sakkinen) writes

> One could say that Postgres is _almost_ relational and _almost_ an OODB,
> but not fully either.
>
> ...
>
> Why not really object-oriented: One of the most important requirements
> of OO systems in general and OODB's in particular is strong object
> identity. The tuple identifiers of Postgres fall short because they
> are unique only within a relation (perhaps they can also be reused?).

if you have observed this behavior in the system, you have observed a
bug.  the intent of oid's (the name we use for tuple identifiers) is
that they be unique within a database, not just a relation.

we are aware that v2.0 (the most-recently released version of the system)
can exhibit the behavior you've described not just across relations, but
also within relations.  the problem was that oid's were a function of
time, rather than some something deterministic.  the problems with this
approach are too gory to go into.

in any case, the oid allocation code now manages oids in the same way
that transaction ids are allocated.  we have a distinguished system
relation subject to locking and access controls so that uniqueness is
guaranteed.  a side effect of this design is that no oid is ever reused
within a database.

i don't know when this code will be going out; send me email if you're
interested, or contact the postgres@postgres mailing list.

> This is not to deny that Postgres is a highly interesting system.
> I have just suggested to a student to do her Master's thesis on it.

great!  we have a few of those here, as you might imagine.  we are
interested in any research results she might get from the system.

					mike olson
					postgres research group
					uc berkeley
					mao@postgres.berkeley.edu