A system and method for performing a dynamic rollover of source and target latency in a replication environment. One or more tables of a source data store being replicated to a target data store are parked, wherein any subsequent changes for the parked tables are not replicated to the target data store. Open transactions are identified where the parked tables affect the source and target latency, wherein the open transactions are in-progress and not completed. The source and target latency are then calculated for the identified open transactions.