Should be similar to what there is in Kafka.

offset.storage should be extended by providing a new Java class that implements the interface and stores and loads the offset from Redis upon startup.

As a several Deezium servers can point to a single Redis (streams) destination, a key per database should be created with the following naming convention:
offset:<server-name(from Debezium conf)>

Default flush should be on every record (currently there's a time based configuration - offset.flush.interval.ms