The authors analyze Hadoop workloads from three different research clusters from a user-centric perspective. The goal is to better understand data scientists' use of the system and how well the use of the system matches its design. Their analysis suggests that Hadoop usage is still in its adolescence. They see underuse of Hadoop features, extensions, and tools. They see significant diversity in resource usage and application styles, including some interactive and iterative workloads, motivating new tools in the ecosystem. They also observe significant opportunities for optimizations of these workloads.