From: Chris Browne <cbbrowne@acm.org>
Subject: Re: trigger for TRUNCATE?
Date: Thu, 10 Jan 2008 15:37:54 -0500
Organization: cbbrowne Computing Inc
Lines: 52
Message-ID: <601w8pxvq5.fsf@dba2.int.libertyrms.com>
References: <4783BEA9.3020507@fmed.uba.ar>
	<60tzloyr0u.fsf@dba2.int.libertyrms.com>
	<162867790801081338n237497efpecffaa255253274d@mail.gmail.com>
	<47864A6B.3070902@fmed.uba.ar>
	<20080110171027.GK6465@alvh.no-ip.org>
	<8003.1199986734@sss.pgh.pa.us>
Mime-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Transfer-Encoding: 8bit
User-Agent: Gnus/5.1008 (Gnus v5.10.8) XEmacs/21.4.21 (linux)
Cancel-Lock: sha1:y59F1KI6MfijzNoH4B9dTkLFDP0=
To: pgsql-sql@postgresql.org

tgl@sss.pgh.pa.us (Tom Lane) writes:
> Alvaro Herrera <alvherre@commandprompt.com> writes:
>> Gerardo Herzig escribió:
>>> Yes, the TRUNCATE statement is not sql ansi, maybe is a more low level 
>>> thing than i think.
>
>> TRUNCATE currently does not fire triggers, but that doesn't mean it's
>> impossible to do it.  I think it would be fairly easy to add support
>> for that.
>
> The entire point of TRUNCATE is to not do a table scan, so making it
> fire per-row triggers seems pretty misguided to me.
>
> We could maybe make it fire per-statement ON DELETE triggers, but
> there's a future-proofing pitfall in that: someday it'd be nice
> for statement-level triggers to have access to the set of deleted rows,
> and then you'd be stuck either scanning the table or having TRUNCATE
> act differently from plain DELETE.
>
> My feeling is that if you want to know what was deleted, you shouldn't
> use TRUNCATE.

No, what would be nice to have is NOT per-row triggering, but rather
simply the ability to run a stored function ON TRUNCATE.

This would be useful for Slony-I:

 - On replica nodes, we might add a trigger:
 create trigger t_trunc before truncate on my_table for each statement execute _sl_cluster.deny_truncate();
   which would raise the error: "Slony-I: Cannot TRUNCATE on subscriber node!"

 - On the "master" we might add a trigger:
 create trigger t_trunc before truncate on my_table for each statement execute _sl_cluster.createEvent('sl_cluster', 'TRUNCATE_TABLE', 14);
   which would generate a 'TRUNCATE_TABLE' event that would tell other nodes to truncate table #14, that is, my_table.

For the case where people want to track "COUNT(*)" on a table using
triggers, TRUNCATE presently throws that off.  With a truncate
trigger, we might implement the following:

 create trigger t_trunc before truncate on my_table for each statement execute purge_table('public', 'my_table');

 create or replace function purge_table (text,text) returns null as $$
    delete from count_summary_table where nspname = $1 and tabname = $2
 $$ language sql;

That's three use cases, so far, none of which expect to have access to
the data that is being truncated.
-- 
"cbbrowne","@","acm.org"
http://linuxfinances.info/info/rdbms.html
Security-wise, NT is a server with a "Kick me" sign taped to it.
-- Peter Gutmann in the Scary Devil Monastery