Pushing Data-Induced Predicates Through Joins in Big-Data Clusters: Extended Version