Spark SQL does the same thing, slower than Impala (does not even have join reordering) but probably more popular.