Date: Fri, 18 Jan 2008 10:48:02 -0500
From: "D'Arcy J.M. Cain" <darcy@druid.net>
To: Gerardo Herzig <gherzig@fmed.uba.ar>
Cc: pgsql-sql@postgresql.org
Subject: Re: transaction and triggers
Message-Id: <20080118104802.181ca4a8.darcy@druid.net>
In-Reply-To: <4790C2B4.1070103@fmed.uba.ar>
References: <4790B50A.5050101@fmed.uba.ar>
	<92869e660801180642i1fdd417r3d3b9399c5e5ab98@mail.gmail.com>
	<4790C2B4.1070103@fmed.uba.ar>
Mime-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit

On Fri, 18 Jan 2008 12:16:04 -0300
Gerardo Herzig <gherzig@fmed.uba.ar> wrote:
> Right.  But  today, that trigger do some other work, wich includes 
> writing some files to disk, so there is my problem. Crap, i guess i will 
> have to review the main logic.

I built a replication system that syncs up dozens of systems in a
multi-master environment spanning multiple continents in almost
real-time and it works flawlessly so don't give up hope.  It is
doable.  I can't give you the code because it was written under
contract and it was based heavily on our specific business requirements
but I can give you a few pointers.

You have discovered the basic problem of trying to replicate in full
real time.  You'll probably have to give up on that.  Instead, focus on
making updates to the local database.  Create a replication table or
tables that you update with triggers.  Basically this needs to be a log
of every change to the database in a structured way.

Once you have the replication table(s) you can create external programs
that connect to the master and update the slave.  In the slave you can
track the last ID that completed.  Do the insert/update/delete in a
transaction so that you have a guarantee that your database is up to
date to a very specific point.  Note that you can have multiple slaves
in this scenario and, in fact, the slaves can have slaves using the
exact same scheme giving you a hierarchy.

If you need multi-master you just need to have another process to feed
your local changes up to the master.  This is not just a matter of
making the master a slave though.  If you do that you get into a
feedback loop.

Also, if you need multi-master, you have to think about your
sequencing.  If you need unique IDs on some tables you will have to
think about setting up ranges of sequences based on server or have a
central sequence server.  We used a combination of both as well as
specifying that certain tables could only be inserted to on one
system.  Of course, this system doesn't need to be the same as the top
of the hierarchy and, in fact, different tables can have different
generator systems.

Hope this gets you started.  There's still lots of gotchas on the way.

-- 
D'Arcy J.M. Cain <darcy@druid.net>         |  Democracy is three wolves
http://www.druid.net/darcy/                |  and a sheep voting on
+1 416 425 1212     (DoD#0082)    (eNTP)   |  what's for dinner.