Multiple times per week, one of my clients loses access to their SQL Server cluster. This is a problem for me because this very large company has a wide variety of applications and languages (Progress 4GL, .Net, Java) interacting with the cluster, and the only servers and apps that lose access are the ColdFusion ones. They have CF 8 Enterprise (fully patched, some standalone, some JRun) on 5 servers, running Windows 2000, 2003, and 2008. All of them experience the issue, but never at the same time. Other applications running on the same servers do not have the issue - it is only the CF across the entire enterprise that periodically becomes unable to see the SQL cluster. (I emphasize this to shortcut the inevitable replies of "check your SQL Server settings".)
The error message received is "Error Executing Database Query. [Macromedia][SQLServer JDBC Driver][Macromedia][SQLServer JDBC Driver]The requested instance is either invalid or not running." (Yes, the instance is up and running, and CF was talking to it 5 seconds ago.) Once this occurs, all queries through that DSN fail, though other DSNs connecting to the same instance on the same cluster may still function. Restarting the CF App Server service does not resolve the problem - only restarting the whole server does. Flushing DNS also does not help. Using the IP instead of name does not help.
Further complicating the issue is the bizarre results of Wireshark debugging during these periods of disconnectedness.
- When queries are made to an unclustered server, CF properly sends the request over the network.
- When queries are made to the cluster's default instance (i.e. no instance name is specified), CF properly sends the request over the network.
- When queries are made to an instance on the cluster, CF sends no data over the network at all.
At that point we have reached the end of my technical ability. I'm a Web programmer, I don't have the ability to debug a network driver. Somehow, when an instance name is specified on a cluster, the CF SQL Server driver just occasionally gets stuck in a state where it will not even try to connect to the server anymore.
The client is not happy. They specifically blame CF for the issue. That it happens across 5 different servers running 3 different OS's, that all the other languages and apps never have the problem, and of course the Wireshark results make it basically impossible to argue otherwise. Surely there's some parameter I can specify in the DSN to stop the issue? Or magic hotfix I've never seen? Would really love to hear anything anyone has to suggest on the matter.
Nick Walters
Federated Solutions Group