On Tue, Jun 08, 2004 at 10:39:29AM +0300, Ivan Danchev wrote:
> I mean that sutiaction usualy is litle bit different
> 
> 1 .I REALLY whant to know how my client application must communicate
> with this "cluster" ??? directly to master DB ? in this case if master 
> crashes, my client application (J2EE web application for example) will lost 
> it's DB connection. ?

Yes, you talk to the master.  Unless you're doing DDL; there's a
special interface for that.

> 2.Who "decide" to "promote" some slave to master ??? - is this some other
> process that I have to run ? or I have to manage this logic in MY CLIENT 
> APPLICATION ?

You do.  If your master fails, you probably shouldn't promote a slave
automatically, because there's no way to know whether the slave is
really up to date.  In other words, buy good hardware for your
master.  Once you get the breaking master up and limping, you can
switch over.  Alternatively, you can make the decision to abandon any
data that might have been committed on the master but which never
made it to the slaves, or something like that.  This is true for
every async replication system in the world: if your data is
important, you'd better have plenty of managers sign the papers
before you abandon some data.

> 3.To what datasource my client application should connect to ?

That probably depends on your application.

> 4.Who will take care after master DB fails, to switch Client
> Application
> connection to newly promoted to master slave.

You will.  You'll have to configure your client.

A

-- 
Andrew Sullivan  | ajs at crankycanuck.ca
The fact that technology doesn't work is no bar to success in the marketplace.
		--Philip Greenspun