You cannot use asynchronous replication with synchronous durability. With asynchronous replication, after a transaction commits, data is replicated from the master to the replica partitions, with a possible delay. A crash of the master partitions would cause the replica partitions to be promoted to masters, resulting in possible data loss due to asynchronous replication. If the transaction were to have synchronous durability, it would commit on the replicas only after the logs on the replicas were saved; yet if the masters failed, data could still be lost when failing over to the replicas. So taking the time to save the logs on the replicas has no expected benefit. Hence, it is not allowed.