Connect CDC detects duplicate keys and accepts only one record for a certain key, rejecting all others. Select Skip Records as the Error Mode. In this mode, only one record is allowed per key and duplicates are rejected. All other rows in the batch are processed. This has been implemented for the Oracle loader, MS SQL Server loader, and Teradata loader.
When using the MS SQL Server loader, Connect CDC detects duplicate keys by using an index that enforces uniqueness. Thus, it ignores a row with a duplicate key but loads all other rows in the same batch or transaction. The index must be defined with the IGNORE_DUP_KEY option.