The kudu-mapreduce integration has not been improved in a couple of years even though there are known outstanding issues and feature gaps. Additionally similar functionality and capabilities now exist via Spark, Hive, Impala, and NiFi integrations. We should document the deprecation of the kudu-mapreduce integration to notify users and so that we can remove it if maintenance becomes an issue.
Even if eventually removed, users can still use the jars published in older versions as needed.
Additionally if a Hadoop InputFormat and OutputFormat is required, a much more modern and maintained version is available in the Apache Hive project: https://github.com/apache/hive/tree/master/kudu-handler/src/java/org/apache/hadoop/hive/kudu