Maybe my expactations where too strict, since near-real-time is not too bad.
However for my business case, finishing the replacation process before triggering the next sql request on the secondary db connection was crucical, so I implemented a logic, that checked the shadow tables, that none data is left for replication.