When a graph fails to execute, the server process has to be restarted in some instances. We would like to investigate if this can be prevented, or if the model can be re-executed with diferent configuration.