Nice work. Indeed, we still can find some open performance issues related to disk reads in the Kafka community. Although there are some solutions, they are not very elegant. Like https://issues.apache.org/jira/browse/KAFKA-7504