The problem of synchronizing the electrical angles of a power system's nodes is investigated in the distributed setting, in which only partial grid information is available at each of the aforementioned nodes. The proposed control solution takes the form of a novel two-layer regulatory architecture. The first layer ensures the stabilization of the grid's dynamics and the dynamical decoupling of its nodes, with respect to exogenous disturbance. The second layer employs predictive control strategies in order to ensure phase synchronization, while also enforcing limits on power injections. The computational cost of the proposed solution boils down to implementing a recurrence equation and to solving a linear optimization problem, at each of the grid's nodes