On Wed, May 19, 2010 at 1:32 PM, Alex Dean <alex@crackpot.org> wrote:

On May 19, 2010, at 2:47 PM, keith smith wrote:



Hi Plug,

I am considering combining the sites on two servers and creating a load balanced configuration.  One server will be in one data center and another will be in a different data center.  This is to ensure if one data center goes off line that these site will still be available via the Internet.  This is also a consideration for backing up our data and content.  Each server is configured with a RAID 1 disk set.

I have never taken on such a responsibility so I am not sure of any gotchas or anything I should be considering.

Our data center guy will be doing most of the work, which will help.

Any suggestions or concerns I should have with two servers in a load balanced arrangement with each server in a different location?

You're entering a world of pain. :)

HA is cool, but is no panacea.  If you haven't actually experienced downtime due to your server crashing or your datacenter losing connectivity, I recommend thinking long and hard about it.  Don't solve a problem you don't have.  The downtime created from unneeded failovers will likely exceed the actual/real downtime caused by either a server or datacenter being offline.  Managing the cluster itself (as distinct from the services provided by the cluster) needs to be accounted for as an expense/responsibility.

I don't want to sound overly pessimistic.  I've set up quite a few HA clusters, and actually enjoy it most of the time.  But it WILL cause you headaches in the middle of the night which you wouldn't have had if you only had a single server.

Leave yourself lots of time to set up a development/test cluster, and abuse it in many ways.  Pull out network cables, kill the switch, yank out power cables, etc.  Do this with real hardware, not VMs.

When the cluster nodes lose contact with each other, both will decide to become primary.  This is a split brain.  This can happen when the switch in-between them gets busy and starts dropping pings.  Now, you can always recover from such things.  I'm just recommending you become very familiar with these issues before going live with this setup.

http://clusterlabs.org/wiki/Main_Page
http://people.l inbit.com/~florian/heartbeat-users-guide/

Let me/us know if you have specific questions once you start setting things up.  Good luck!

alex
---------------------------------------------------
PLUG-discuss mailing list - PLUG-discuss@lists.plug.phoenix.az.us
To subscribe, unsubscribe, or to change your mail settings:
http://lists.PLUG.phoenix.az.us/mailman/listinfo/plug-discuss


Hi Kev!

You could also easily use round robin DNS to hit one server and then the other in a load balanced way.  This would not add any polling for uptime between servers, and if one goes down the other always answers.  It works....

You are not going to be able to easily use any of the ARP/RARP based cluster products with one node off on another network, BTW.  A coyote point LB is also not going to work, neither is drbd.  Of course you could hang a nice VLAN between data centers?

I doubt that the bandwidth costs between these servers (and the Load Balancer) is going to be worth any warm and fuzzy "insurance" for uptime.

In actuality you could do well with two built identical and a backup dd iso of a configured server, with off site data backup.  I can build or restore a server in 2 hours.

An Ultra Monkey Cluster is also not going to work without a VLAN accross two  datacenters.  If you are using session cookies, you might have some issues with clustering, without a layer 3.5 load balancer. 

Have you considered using a VMware server for failover?  Many people do this noa
--
Office: (480)307-8707
AT&T: (503)754-4452