This is a good approach. I wonder if using a UUID-aware data type (like PostgreSQL's UUID type) would improve performance without making the second column necessary?