Using /_cluster/settings -> cluster.routing.allocation.exclude works for only one node?

And btw, for those running in env's that have nodes restart from time to time, make sure its persistent (like solution says).