How should the ETL system be designed if processing a file with a million lines fails after successful normalization of about 700 thousand lines?