With this MR I introduced a function that organizes the output tile information in a dictionary that is later used in groups by the main program to write csv files.
I found out that managing single pickle files would be much slower than pushing big batches in csv files.
The main program is written in a way that the reusable objects (Datasource, Database connection ...) are created outside everything so in the multiprocessing they can be consulted without creating them over and over.