This happened to us. The problem is probably that you have your metrics replication controllers set to pull the latest versions of the images. (I think this is the default. Bad!) The current latest version needs different configuration, so your existing configuration no longer works. You probably had this problem for a long time but didn't notice until some component of the system restarted for some reason, triggering a new image pull.
We fixed this by changing the images specified in the replication controllers. For example, in rc/hawkular-metrics, we changed
While I was debugging, I restarted hawkular-cassandra, so it got upgraded, too. I don't know if it had already gotten upgraded; if yours hasn't, then you can avoid losing data. So, I had to set the :v1.2.1 tag on all three components (hawkular-cassandra, hawkular-metrics, and heapster) and also delete all data (both the data directory and the commitlog directory) on the hawkular-cassandra PV. In order to delete that data, I had to find the mountpoint on the node where the hawkular-cassandra pod was running and delete the files from the host side. Because hawkular-cassandra was failing, I was unable to use `oc rsh` to get in.