I have a cluster of about 20 datanodes. Suppose I have a need to shut about half of them off, let's say to move them across the room. I have the impression that the correct action is to stop all services on all nodes, including the Primary NameNode, then shutdown the nodes to move. Question 1) is this correct? and 2) is there risk of losing any data? (Of course I have to ask.) And question 3) is the restart procedure just to boot the nodes, then start all services on all nodes? And 4) as I don't believe the cluster's ever been rebooted, can we test this procedure by stopping and starting all services on one node at a time while leaving others running?
... View more