In this post, Facebook describes how they used Apache Spark to redesign a large-scale, complex (100+ stage) pipeline originally written in HQL over Hive.
