My understanding is that the s3 input fetches every object from the bucket and compares the last_modified metadata with the sincedb. If watch_for_new_files is set it will do that over and over, which is why you might want one of the backup options and the delete option so that once an object is processed it is no longer fetched over and over again.