Systems and methods are provided for re-balancing and healing of an SD-WAN in an unbalanced state and/or experiencing one or more failure states. In response to a request to connect to a new controller resulting from OMP load shedding from a first controller, the system can identify other controllers capable of handling the load requirements of the edge router. The system can incorporate the controller group preference of the edge router and select a second controller based on the identified other controllers and within the preferred controller group. If not possible, the system can temporarily assign the edge router to non-preferred controller groups and move them back to controllers in the preferred controller group once it becomes viable. The system further enhances OMP graceful restart (GR) logic to incorporate the load shedding effect and avoid unnecessary route retention that GR entails.