Developing Client Applications for Clusters

These are recommendations for developing client applications that operate effectively on Microsoft® Windows NT® Clustering Service clusters:

• Most server applications do not need to be modified for use with Clustering Service.

  However, transaction-based applications (for example, database servers such as SQL Server) usually require additional modification so that if the server fails, loss of transactional integrity is prevented and transactions are rolled back. Design client applications with database recovery and error checking in mind.

• During failover, client applications lose their connection to SQL Server and must reconnect to continue processing.

  If the client and server have a common state (for example, open cursors, session variables, or data in tempdb), failover is not transparent to the client. In these cases, design an application to inform the user that the connection was lost or reset, or to reestablish its connection to the server automatically.

• The error-checking process is similar to developing a client database application for a cluster.

  When the cluster begins failover, the client program receives an error on the database connection. The error encountered depends on what the client program is attempting to do at that time.

• When a virtual database server fails, a connection link failed error is returned to the waiting client.

  The database on the failed server of the cluster is shut down and restarted on the remaining node, where it recovers the databases. If a client attempts to connect to the server while a database is being recovered, the client receives a waiting for database recovery error and should retry after a short pause.