1-3 are nice if you can guarantee your data is reasonably sized, but if it gets too big for your hardware taking copies of large datasets and then doing updates

1-3 are nice if you can guarantee your data is reasonably sized, but if it gets too big for your hardware taking copies of large datasets and then doing updates on large datasets can add a lot of overhead.