I'm not sure what they mean here. Nobody writes map reduce programs in Java anymore, running it as a classic Hadoop MapReduce job. People use languages such as

I'm not sure what they mean here. Nobody writes map reduce programs in Java anymore, running it as a classic Hadoop MapReduce job. People use languages such as Spark, Hive or Pig to run their queries against data that is usually in hdfs (part of Hadoop), and the cluster resources are distributed to jobs from yarn (which is also part of Hadoop). So the question to use Spark or Hadoop seems odd to me. It's like comparing apples and oranges, no?