This is intresting. The node that runs Cloudera Manager, Hue, HIve, Ozzie, reports a Clock Offset even with ntp running correctly. This node is not a Datanode or a Namenode. In this situation ALL services are reported as failed ("Bad") health except HDFS. I'm still able to browse the HDFS filesystem, run example MapReduce jobs and other services reported as "Bad" such as the Hue service, still work. Once I disable the Clock Offset ckeck on this node all serviced return to a good state within a minute. I can duplicate this behavor by enabling the check.
I'd like to see the actual output from the ntpq check, but fail to find anything in the logs that indicate errors.