There should be a new connector option that takes the column(s) which should impact the partition, applying the same hashing strategy that's applied to the record key. E.g. given there are two tables:

PURCHASE_ORDER(ID, DATE, STATUS)
ORDER_LINE(ID, PURCHASE_ORDER_ID, ITEM, PRICE)

If there's a configuration given message.partition.columns=inventory.order_line:purchase_order_id, then the same partition selection strategy should be applied for both tables. In particular, in a distributed Kafka Streams application, the corresponding messages from both change data topics should be handled by one and the same application node.

A current work-around is to use a custom SMT, as described here.

This feature should be beneficial for table-to-table joins on non message keys, as enabled in AK 2.4 by KIP-213 ("Support non-key joining in KTable").