Using Fenics with Petsc TS

I am using Fenics together with Petsc TS for timestepping.
This works in serial. If I am using mpirun -N 2 the solution is only calculated on one partition. Which is probably related to the way I access the vectors.
At the moment I use


and pass this vector x to the Petsc routines. Afterwards I write it back with


What is the right way to interface the vectors in parallel?
I am using the discontinous Garlerkin Method, it seems the problem is solved on both partitions, but the domain are not coupled (ghost_mode is set to shared_facet)
Hower solving my steady state problem with dolfin PETScSNESSolver, both partitions are coupled.