Manually Initiate a Failover in Cluster Administrator
This test is also performed from Cluster Administrator. Select any of the resources found in the SQL Server Group resource group (not the group itself), right-click on it, and select Initiate Failure. Because the cluster service always tries to recover up to three times from a failover, if it can, you will have to select this option four times before a test failover is initiated.
As above, after the first failover, check for any problems, then failback using the same procedure. Then check again for problems.
Manually Failover Nodes by Turning Them Off
Turn off the active node. Once this happens, watch the failover in Cluster Administrator and the clients. As before, check for any problems. Next, turn on the node and wait until it boots back up successfully. Then turn off the now current active node by turning it off hard. And again, watch the failover in Cluster Administrator and the clients, and check for problems. Turn the node back on when done.
Manually Failover Nodes by Breaking the Public Network Connections
Unplug the public network connection from the active node. This will cause a failover to a passive node, which you can watch in Cluster Administrator and the clients. Check for any problems. Now, plug the public network connection back into the server. And unplug the public network connection from the now active node. This will cause a failover to the current passive node, which you can watch in Cluster Administrator. And again, watch the failover in Cluster Administrator and the clients, and check for problems. Once the testing is complete, plug the network connection back into the server.
Manually Failover Nodes by Breaking the Shared Array Connection
From the active node, remove the shared array connection. This will cause a failover, which you can watch in Cluster Administrator and the clients. Check for any problems. Next, reconnect the broken connection from the now active node, and remove the shared array connection. Watch the failover in Cluster Administrator. And again, watch the failover in Cluster Administrator and the clients, and check for problems. When done, reconnect the broken connection.
If you pass all of these tests the first time, it would almost be a miracle. But I do believe in miracles. If you run into problems, you have to figure them out.
Ready for Production
Once you have successfully tested your production SQL Server 2005 cluster, you are ready to go into production. If you have time, you might want to consider running the cluster in test mode for a while, “playing” with it to learn more about how it works under various conditions. But even if you don’t have any extra time to “play” with your cluster, it should be ready to go into production. Now pat yourself on the back for a job well done.
]]>