I'm making join between Parquet DB stored on S3

but it's seems that anyway Spark try to read all the data as we not see better performance when changing the queries.

I need to continue to investigate this point because it's not yet clear.