I think the "standard case" would be to use something like

updated_at

as the replication key, and then handle de-duplication in the OLAP database -- OR to do an UPSERT on the updated rows.