Another thing is to plan for a solution that can expand. Don't use a data type that requires massive table rebuild times if you ever increase its size. If adding a column or expanding a field takes O(n) time, don't expect n to be small forever. If an operation takes greater than O(n) time, realize that you'll pay that price eventually.
Basically, establish soft upper bounds.
And as a P.S.: don't expect anything sanitized by your application layer to be an invariant. Your database and the rules enforced by it are the only real invariants. If an unexpected character in a name field will blow up your application, you should fix it in the database (varyingly difficul with many RDBMS solutions) or treat it as user input and sanitize/scrub it at the application layer (more common with NoSQL solutions). Ideally, do both!
Basically, establish soft upper bounds.
And as a P.S.: don't expect anything sanitized by your application layer to be an invariant. Your database and the rules enforced by it are the only real invariants. If an unexpected character in a name field will blow up your application, you should fix it in the database (varyingly difficul with many RDBMS solutions) or treat it as user input and sanitize/scrub it at the application layer (more common with NoSQL solutions). Ideally, do both!