The stacktrace from the logs in production provided inadquate information to reproduce the problem, hence we added logging of the model variables if the exception occurs.