GraphFrames is the new graph library of Apache Spark. It supports loss-less transformations of graph representations between GraphFrames and GraphX. While introducing another Apache Spark library equivalent to an existing one, we have to explain the motivation behind this decision. The motivation is the same as it was when introducing SparkML over MLlib. The idea was to have an equivalent to GraphX, which supports DataFrames, in order to make use of the optimizations that Catalyst and Tungsten bring.

Another goal was to progress further with unification by integrating graph algorithms and graph queries and optimizing their execution.

Although GraphX supports relational queries on graph structures, it doesn't support graph queries.

Therefore, GraphFrames was implemented with the goal of unifying graph algorithms, graph queries, and DataFrames (and therefore relational queries).

This chapter will cover the following topics:

• Architecture of GraphFrames
• Performance optimizations
• Practical examples in Scala