As observed by @imrefazekas it took some time for the cluster to be ready to receive the console call, this change helped fix it.

why this? console just connect to FIRST graph, no need to wait other graphds.
prefer a loop like: https://github.com/vesoft-inc/nebula-python/blob/master/tests/docker-compose.yaml

        for i in `seq 1 60`;do
          var=`nebula-console -addr graphd0 -port 9669 -u root -p nebula -e 'ADD HOSTS "172.28.2.1":9779,"172.28.2.2":9779,"172.28.2.3":9779;'`;
          if [[ $$? == 0 ]];then
            break;
          fi;
          sleep 1;
          echo "retry to add hosts.";
        done && tail -f /dev/null;

This is better, I'll close this, could you port your implementation here?