I use uuid as primary key when I plan for scale and future sharding. It's negligible performance hit of at all and I can move entire datasets from one db to another without worrying about references.