Issue with Public LAN in Windows – Exchange Clustering
Yesterday I began a demo of a Windows Server 2003 Cluster running an Exchange 2003 virtual server. This is a 2 node active/passive cluster. Everything was fine until I got to the part where I was showing hardware failure and the resulting failover performance. I unplugged the public LAN ethernet cable from the active cluster node. To my dismay and embarrasment the cluster administrator locked up and failover never happened. I was sure I had tested this particular scenario before during the initial setup. After thinking about this a little more I probably tested this with the default cluster during setup, and not after Exchange was up and running. Why would you want to disconnect the public LAN cable right? Well, when tested I discovered that Exchange in a cluster is very sensitive (as it usually is) to losing its public LAN interface. Anytime Exchange (clustered or not) loses connection to AD and DNS, it can hang. Well my problem was cuased by the Exchange 2003 services hanging once public LAN connectivity was lost. I found that the cluster was trying to stop the Exchange services but they were hanging. The cluster can’t failover to the second node until the core Exchange services are stopped and Exchange cluster resources are marked as offline. So because the services were hanging, my cluster would not fail over to the second node.
I only discovered this limitation through testing and found very little about the problem online. I tried my handy experts-exchange account, Microsoft Technet newsgroups, etc. The closest I could find was a comment from a Technet MCP that losing the public LAN on a cluster was “Very very bad”.
So the workaround for this issue is this. Add a third network interface to each node in the cluster. You should already have one for a heartbeet network and one for a public LAN network. So add a third interface on the public LAN same setup as the other public interface, just a different IP. This way, if something happens to any single public LAN interface your cluster won’t have to failover and it won’t cause a problem. Of course you should connect the ethernet cable on this backup public interface to a physically separate switch to add to the redundancy.
Posted on May 17, 2007, in Professional/Tech and tagged cable, DNS, exchange, experts, handy, IP, LAN, love, man, Microsoft, network, newsgroups, server, sing, tech, time, windows. Bookmark the permalink. Comments Off on Issue with Public LAN in Windows – Exchange Clustering.