Date: Thu, 11 Jan 2007 15:01:22 -0600
From: "Jim C. Nasby" <jim@nasby.net>
To: Simon Riggs <simon@2ndquadrant.com>
Cc: Steven Flatt <steven.flatt@gmail.com>,
	Colin Taylor <colin.taylor@gmail.com>,
	pgsql-performance@postgresql.org, pgsql-hackers@postgresql.org
Subject: Re: [HACKERS] table partioning performance
Message-ID: <20070111210122.GR36267@nasby.net>
References: <84d3067d0701062037k3229c9ddjbde73b1fd79ee209@mail.gmail.com>
	<357fa7590701081202p726c049ej61390dd1816d1369@mail.gmail.com>
	<1168346933.3951.276.camel@silverbirch.site>
	<357fa7590701101300m2812545k1487ca58e97970ed@mail.gmail.com>
	<1168517751.3951.538.camel@silverbirch.site>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <1168517751.3951.538.camel@silverbirch.site>
User-Agent: Mutt/1.5.13 (2006-08-11)

On Thu, Jan 11, 2007 at 12:15:50PM +0000, Simon Riggs wrote:
> On Wed, 2007-01-10 at 16:00 -0500, Steven Flatt wrote:
> > On 1/9/07, Simon Riggs <simon@2ndquadrant.com> wrote: 
> >         If you are doing date range partitioning it should be fairly
> >         simple to
> >         load data into the latest table directly. That was the way I
> >         originally 
> >         intended for it to be used. The rules approach isn't something
> >         I'd
> >         recommend as a bulk loading option and its a lot more complex
> >         anyway.
> > The problem we have with blindly loading all data into the latest
> > table is that some data (< 5%, possibly even much less) is actually
> > delivered "late" and belongs in earlier partitions.  So we still
> > needed the ability to send data to an arbitrary partition.
> 
> Yes, understand the problem.
> 
> COPY is always going to be faster than INSERTs anyhow and COPY doesn't
> allow views, nor utilise rules. You can set up a client-side program to
> pre-qualify the data and feed it to multiple simultaneous COPY commands,
> as the best current way to handle this.
> 
> --
> Next section aimed at pgsql-hackers, relates directly to above:

I'm wondering if you see any issues with COPYing into a partitioned
table that's using triggers instead of rules to direct data to the
appropriate tables?

BTW, I think improved copy error handling would be great, and might
perform better than triggers, once we have it...
-- 
Jim Nasby                                            jim@nasby.net
EnterpriseDB      http://enterprisedb.com      512.569.9461 (cell)