X-Original-To: pgsql-performance@postgresql.org Received: from localhost (postgresql.org [64.49.215.8]) by postgresql.org (Postfix) with ESMTP id E9834475AF8 for ; Wed, 26 Feb 2003 11:09:47 -0500 (EST) Received: from postgresql.org ([64.49.215.8]) by localhost (postgresql.org [64.49.215.8:10024]) (amavisd-new) with SMTP id 82222-01 for ; Wed, 26 Feb 2003 11:09:37 -0500 (EST) Received: from sss.pgh.pa.us (unknown [192.204.191.242]) by postgresql.org (Postfix) with ESMTP id 6B4DB475AFA for ; Wed, 26 Feb 2003 11:02:31 -0500 (EST) Received: from sss2.sss.pgh.pa.us (tgl@localhost [127.0.0.1]) by sss.pgh.pa.us (8.12.6/8.12.6) with ESMTP id h1QG2V5u029928; Wed, 26 Feb 2003 11:02:31 -0500 (EST) To: Richard Huxton Cc: daniel alvarez , pgsql-performance@postgresql.org Subject: Re: OIDs as keys In-reply-to: <200302261358.53730.dev@archonet.com> References: <1859.1046264679@www36.gmx.net> <200302261358.53730.dev@archonet.com> Comments: In-reply-to Richard Huxton message dated "Wed, 26 Feb 2003 13:58:53 +0000" Date: Wed, 26 Feb 2003 11:02:30 -0500 Message-ID: <29927.1046275350@sss.pgh.pa.us> From: Tom Lane X-Virus-Scanned: by amavisd-new X-Archive-Number: 200302/213 X-Sequence-Number: 1269 Richard Huxton writes: >> How unique are oids as of version 7.3 of postgres ? > OIDs are unique per object (table) I believe, no more so. Even then, you should only assume uniqueness if you put a unique index on the table's OID column to enforce it. (The system catalogs that use OID all have such indexes.) Without that, you might have duplicates after the OID counter wraps around. >> Will the oid type be extended so that oids can be larger than 4 bytes (if >> this is still correct for 7.3) and do not rotate in large systems? > Strikes me as unlikely, though I'm not a developer. I tend to agree. At one point that was a live possibility, but now we're more likely to change the default for user tables to WITHOUT OIDS and declare the problem solved. Making OIDs 8 bytes looks like too much of a performance hit for non-64-bit machines. (Not to mention machines that haven't got "long long" at all; right now the only thing that doesn't work for them is type int8, and I'd like it to stay that way, at least for a few more years.) regards, tom lane