Date: Tue, 23 Jan 2007 14:35:48 +0100
From: Tobias Brox <tobias@nordicbet.com>
To: Chad Wagner <chad.wagner@gmail.com>
Cc: Tobias Brox <tobias@nordicbet.com>,
	pgsql-performance@postgresql.org
Subject: Re: extract(field from timestamp) vs date dimension
Message-ID: <20070123133548.GA17420@oppetid.no>
References: <20070123124937.GA13122@oppetid.no>
	<81961ff50701230524y14aa883alfe4dab8d1dd5cf10@mail.gmail.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <81961ff50701230524y14aa883alfe4dab8d1dd5cf10@mail.gmail.com>
Organization: Group Nordicbet
User-Agent: Mutt/1.5.9i

[Chad Wagner - Tue at 08:24:34AM -0500]
> I guess go with your gut, but at some point the expressions are going to be
> too complicated to maintain, and inefficient.

The layout of my system is quite flexible, so it should eventually be
fairly trivial to throw in a date dimension at a later stage.

> Calendar tables are very very common, because traditional date functions
> simply can't define business logic (especially things like month end close,
> quarter end close, and year end close) that doesn't have any repeating
> patterns (every 4th friday, 1st monday in the quarter, etc).  Sure you can
> stuff it into a function, but it just isn't as maintainable as a table.

So far I haven't been bothered with anything more complex than "clean"
weeks, months, quarters, etc.

I suppose the strongest argument for introducing date dimensions already
now is that I probably will benefit from having conform and
well-designed dimensions when I will be introducing more data marts.  As
for now I have only one fact table and some few dimensions in the
system.