The question would be whether or not Iterable would save memory. It's trivial for me to build a list out of my iterable. If I understood the code correctly, Spark takes that List and converts it to an array, so I built an ArrayList out of the iterable in the hopes that Spark would use the underlying array structure natively in the RDD. If that is the case, then no effort or memory has been wasted (at least for single node). If (when) that is not the case, if the RDD throws away my array, then indeed it would be more efficient to pass Spark an iterable and let it build up whatever internal representation that it needs.
Our bigger picture is that we have a proprietary streaming system that streams rows of data (the "row" is a java data structure we have in the form of Iterable<ProprietaryRow>. The Iterable<> may invoke other upstream Iterables which may invoke a streaming read from a database, file, etc. So far we have been careful to avoid collecting the entire stream in memory unless absolutely necessary. By experimenting with Spark and RDD, we are taking the leap of collecting the entire dataset in (potentially distributed) memory to see if it can help us parallelize and scale. -- View this message in context: http://apache-spark-user-list.1001560.n3.nabble.com/How-to-create-RDD-from-Java-in-memory-data-tp2486p2560.html Sent from the Apache Spark User List mailing list archive at Nabble.com.