DomainKey-Signature: a=rsa-sha1; q=dns; c=nofws; s=s1024; d=yahoo.com;
	h=Message-ID:Date:From:Subject:To:MIME-Version:Content-Type;
	b=ydh0EGfYc+Ub/m6PSD/eWV4F48FEbyxxq0AOESNZai+TxYUGGB6I64WAd6+cBKexGyBkEQYuNLbPcabh6y9SXn26MTDQxhAKXHSgvMCR1xF65jIpLFB8X8/5hfzuoDP0EgG6IeqeQvTZ1Rz0eULMBc5ZX7SHy+tLOb1mQZJmrF0=
	;
Message-ID: <20060321145839.68211.qmail@web50301.mail.yahoo.com>
Date: Tue, 21 Mar 2006 06:58:39 -0800 (PST)
From: <ogjunk-pgjedan@yahoo.com>
Subject: Referential integrity broken (8.0.3), sub-select help
To: pgsql-sql@postgresql.org
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii

Hello,

I've got 2 tables, "url" (U), and "bookmark" (B), with "bookmark" pointing to "url" via FK.
Somehow I ended up with some rows in B referencing non-existent rows in U.
This sounds super strange and dangerous to me, and it's not clear to me how/why PG let this happen.
I'm using 8.0.3.

Here are the table references I just mentioned:

Table "bookmark":
     id                  SERIAL
                         CONSTRAINT pk_bookmark_id PRIMARY KEY
 
 Table "url":
    url_id              INTEGER
                         CONSTRAINT fk_bookmark_id REFERENCES bookmark(id)


Problem #1: Strange that PG allowed this to happen.  Maybe my DDL above allows this to happen and needs to be tightened?  I thought the above would ensure referential integrity, but maybe I need to specify something else?

Problem #2: I'd like to find all rows in B that point to non-existent rows in U.  I can do it with the following sub-select, I believe, but it's rather inefficient (EXPLAIN shows both tables would be sequentially scanned):

  SELECT * FROM bookmark WHERE id NOT IN (SELECT b.id FROM bookmark b, url u WHERE b.url_id=u.id);

Is there a more efficient way to get the rows from "bookmark"?

Thanks,
Otis