Oh btw, one final thing you can try just so we’re 100% sure:
Don’t do parallelism. Just stream the CSV file records and insert them one by one and see if you get 10M records.
BTW the amount above is mind-blowingly different… you expect 10M records but get almost 110M? Something is super weird, man, you must have overlooked something in the code.
I say make a public GitHub project with all your code and include a link to where the big CSV file can be downloaded (outside GitHub) so people can try and help you.
