Using hadoop/spark for <2gb of data seems like a terrible idea.

When all you have is a hammer everything starts to look like a nail.