High Availability Lamp cluster

[vincentthe]

Hi there,

I'm in the situation that I need a new HA-cluster in operation, but I want to limit the needed resources to a minimum.
After reading the series on the high availability nfs/apache/mysql clusters + loadbalancer I have a couple of questions, I wonder if anyone could give me a hand in this.

I'm going to run a php/mysql web-application with several interactive elements (= heavy DB usage) and some data-hogs (movies on demand). At this moment I think that fail-over is more important than pure performance so the setup I have in mind at this moment consists of:
2x loadbalancer
2x webserver
2x db-server + nfs server

My questions:
- Is this the logical combination or is it better to combine webserver & nfs server on 1 box?
- The MySQL clustering solution has loadbalancer 1 as the management node. Shouldn't it be installed on loadbalancer 2 as well? It seems to me that the approach from the tutorial still creates a single point of failure. What happens to the Cluster if LB 1 crashes and DB1 crashes after that?

Thanks in advance!

[vincentthe]

Could it be an option to purchase 3(?) faster machines and use Virtualization and Xen to obtain the same goal? It would save on co-location costs...

[falko]

I'd use MySQL Master-Master replication ( http://www.howtoforge.com/mysql5_mas...on_debian_etch ) instead of a MySQL cluster. A MySQL cluster holds all data in memory, so if there's a power failure, your data is lost.

If you want to save costs, you can get just two physical servers and set up Xen VMs on them.

[vincentthe]

From the tutorial it seems to me that downtime from one of the master requires quite a lot of manual intervention to get the system up again? In short, how is disaster recovery managed in this case?
Is it still possible to use mysql load balancing in this setup?

[falko]

Originally Posted by vincentthe

From the tutorial it seems to me that downtime from one of the master requires quite a lot of manual intervention to get the system up again?

Once the systems gets up again, it should catch up with the other master again, provided that the relevant logs still exist on the other master (look at the expire_log_days setting in my.cnf). You can check with the

SHOW SLAVE STATUS;

command if both systems are in sync. If they aren't, you must set up replication again as shown in the tutorial.

Originally Posted by vincentthe

Is it still possible to use mysql load balancing in this setup?

You can put a load balancer in front of the MySQL server, e.g. HAProxy.