Are you using Spark-MR3 in production, or interested in using it in production? We are planning to release Spark 3.2.1 in the release of MR3 1.5, but we are not aware of any use case of Spark-MR3 in production.
Spark-MR3 differentiates itself from vanilla Spark when it comes to running multiple Spark drivers concurrently. That is, you need just a single Spark-MR3 cluster to support multiple Spark drivers, which can improve resource utilization significantly.
Cheers,
--- Sungwoo