Currently, the engine Dockerfile has a snippet commented to be uncommented by whom wants to use Spark. Requiring user to uncomment code in order to achieve functionality is not a good practice. Idea: create another image with different execution environment (Spark, Hadoop, Hive etc...). Maybe call it marvin-big-data
Activity