Distributing long column data involves extra processing and a different level of transaction consistency. If a replication source row contains long columns, the row’s non-long data values are captured at the time the database is updated via our replication trigger. The replication trigger notices when a long column value is inserted or changed. The long column value is not captured at that time. Instead it is captured from the source database when the transaction in the queues is processed for distribution to the target servers.
Because the long column values are fetched after the row was captured, the fetched values may not be the same as the values that were present when the row was captured. The value sent to the target server(s) is whatever column value is present in the row at the time it is retrieved. If the row itself no longer exists, a null value is sent to and inserted into the target column.