|
Francois JEANMOUGIN wrote:
You are talking about pure failover only right?
No no no.
I was talking about load balancing for mysql which is pretty much
undoable.
Please read the links I sent, and add this one :
http://dev.mysql.com/doc/mysql/en/mysql-cluster-overview.html
I'm talking about pure active-active cluster. This setup could be strenghted
by LVS and a Keepalived MISC checker. In the figure above, you put LVS
between Applications and SQL. About the link above, note that mysqld server
and nbdd server can be on the same machine.
François.
Ah,
This could indeed be the setup we need!
So if i get this right (i read ll three docs) you can setup a mysql
server on each of the nodes of you cluster (allowing all of the nodes to
be load balanced mysql servers)
What would be a better sollution (afaik) is to place mysql on all the
apache servers aswell, since this would would allow them to connect to
their own database instead of using a network connect, decreasing the
overhead for each query/connect.
And if one of the webserver nodes goed down, so does its mysql, so thats
not a real problem. LVS will just not redirect any request to the
machine at all.
A not clustered server would then be the nbdd server managing and
updating all of the mysql servers on the nodes itself.
And thus you could use really inexpensie hardware for the mysql server,
as you won;t need UBER speed on that machine to handle all requests from
all LVS-nodes anymore.
Il see if i can get this running on a small test setup.
You openend my eyes! :)
|
