Failure and Redundancy
If you have redundant copies of a partitioned table, you can lose servers without loss of data or interruption of service. When a server fails, GemFire XD automatically re-routes any operations that were trying to write to the failed member to the surviving members.
GemFire XD also attempts to re-route failed read operations to another server if possible. If a read operation returns only a single row, then transparent failover is always possible. However, if an operation returns multiple rows and the application has consumed one or more rows, then GemFire XD cannot fail over if a server involved in the query happens goes offline before all the results have been consumed; in this case the application receives a SQLException with SQLState X0Z01. All applications should account for the possibility of receiving such an exception, and should manually retry the query if such a failure occurs..
Read operations are also retried if a server is unavailable when a query is performed. In this figure, M1 is reading table values W and Y. It reads W directly from its local copy and attempts to read Y from M3, which is currently offline. In this case, the read is automatically retried in another available member that holds a redundant copy of the table data.