Lagrange multipliers and constrained optimization

How do you maximize a multivariable function when there's some constraint on the inputs that you're allowed? This is the kind of problem that comes up all the time in practice, so knowing how to approach it is incredibly useful.