@Nelli In case you get to "that point", I have found using Map/Reduce to import CSV files to be pretty effective. In that case, you literally "load" the CSV file from within the Map/Reduce script's "getInputData()" stage, and then allow each map() stage to process each row of data. I've found this helpful for very large imports (100 MB or less), or, for complex imports, which involve complex logic for each row import.