It usually gives us a chuckle when we find some natural and seemingly easy data science question is NP-hard. For instance we have written that variable pruning is NP-hard when one insists on finding a minimal sized set of variables (and also why there are no obvious methods for exact large permutation tests).
In this note we show that finding a minimal set of columns that form a primary key in a database is also NP-hard.