Convex vs Non-Convex Landscapes

How the shape of an optimization landscape decides whether gradient descent finds the global optimum or gets stuck.