The config does look ok IMHO. Could you make sure that the “disks” are empty and do not hold a “single-node” cluster state? My guess (also just a guess :-)) that it is something similar to Crate DB Clustering on EC2 where the nodes where started as single nodes first.