1) I notice that HDFS disk utilzation is not proportionate across nodes having different disk sizes. For example, is nodes 1-10 have 2 GB drives and nodes 11-15 have 3 GB drives, I would expect a larger (and therefore proportionate) usage of space on the 3 GB drives to keep the %usage around the same regardless of disk size. Instead, I observe that usage is uniform across the different sized drives causing smaller drives to to filled up faster. Is this the default behavior of HDFS? Is there a way to change this, say via Ambari, and how?
2) Also, if there were n disks per node supporting HDFS and I desire to reduce that to n-1 disks per node, I would expect this could be done dynamically. In a cluster with k-factor=3, I would think I should be able to decommission a disk on a node at the Linux level, wait for HDFS to reconcile and make the third copy elsewhere and then proceed to decommission a disk on the second node ….and so on. All this while the cluster is live. Is this a possibility? If yes, is there an easier way to do this via Ambari?