In $1904$, Prandtl introduced his famous boundary layer in order to describe the behavior of solutions of incompressible Navier Stokes equations near a boundary as the viscosity goes to $0$. His Ansatz was that the solution of Navier Stokes equations can be described as a solution of Euler equations, plus a boundary layer corrector, plus a vanishing error term in $L^\infty$ in the inviscid limit. In this talk, I will present a recent joint work with E. Grenier (ENS Lyon) which proves that for a class of  smooth solutions of Navier Stokes equations, namely for shear layer profiles which are unstable for Rayleigh equations,  this Ansatz is false if we consider solutions with Sobolev regularity, in strong contrast with the analytic case, pioneered by R.E. Caflisch and  M. Sammartino. On the other hand, we also address the classical problem of the nonlinear stability of shear layers near a boundary and prove that if a shear flow is spectrally unstable for Euler equations, then it is non linearly unstable for the Navier Stokes equations, provided the viscosity is small enough.