it seems kafka need tuning in order to get rid of the issues..

1. Reduce poll.interval.ms

2. reduce log.mining.sleep.time.millis

   From Database side you can guarantee undo retention..

3. SELECT tablespace_name, retention FROM dba_tablespaces WHERE contents='UNDO';

4. ALTER TABLESPACE UNDOTBS1 RETENTION GUARANTEE;

   Hope it fix the issue