Tez UI is subtly broken on HA clusters

[karth295]

We run the app timeline server on all masters, and point clients to $HOSTNAME to find the timeline server. This means that a) each Tez UI server only reads metrics from the timeline server on localhost, and the RM only logs metrics from the timeline server on localhost.

Does this mean that the Tez UI on a node only has history for drivers that ran on that node? Or the hive-server2 that ran queries? (Not sure what the effect is yet -- just wanted to document this).