Hey so im working on setting up a new server cluster for a client of mine.

Was wondering how you guys handle HA while still having the ability to have multiple nodes active at the same time with writes and reads going to all of them.

What I was thinking was to have a NFS backend that all the web/db nodes connect to to get their data thus allowing then to act separately whole sharing a backend. Of course there would be some sort of transaction tracking for the db so that it wouldn't get screwed up but its a overview of what I was thinking. The other part of all of this is being cost effective.

Right now I have them running on 2 2GB plans through linode one being the web and second being the database. While its working great and have not had any serious problems the client is planning on ramping up his advertising and we will probably need some more power behind his site, so I figured instead of building up I would start building out.

Any tips tricks, ideas etc will be appreciated.

NFS would still be the single point of failure. MySQL and php? Same data center?

NFS would be setup on a DRBD w/ heartbeat for node failover.

Yes same datacenter but I may do a heartbeat failover if the client desires that much redundancy. Basically the goal is to be able to take down a node and upgrade it or w/e and not take downtime to do it.

The website is run on Wordpress (not my call unfortunately but I was tasked with running the servers) so PHP and MySQL

What about Nodebalancer? http://library.linode.com/nodebalancers/getting-started


For MySQL, there are few different types of replication if you need that.

TechSnap had a feedback question that chatted about this sort of setup:

http://www.jupiterbroadcasting.com/1681 ... chsnap-44/