We are attempting to upgrade from Cloudera Manager 4.7 / CDH 4.8.0 to Cloudera Manager 5.1 / CDH 5.1.0. While our worker nodes have local disk for tmp, swap, and HDFS, we boot these nodes disklessly using a common NFS root and oneSIS image based on RHEL 6.5. The cloudera agent now appears to have a UUID in /var/lib/cloudera-scm-agent/uuid that is (should be) unique to each node. When the file is populated with the UUID generated on the image install node, and all the nodes are booted, the cluster is very confused. When the file is empty or missing, the agent fails to start. How can we get the agent to populate this file on boot, so that we can run CM5 / CDH5 on diskless worker nodes? Thanks for reading and responding to this.
... View more