http://linkedin.jira.com/browse/NOR-10 introduced the ability to back off from an unresponsive server. In my setup, I have one network server and multiple network clients, all who use the same cluster client. I see quite a few warnings of the sort
[com.linkedin.norbert.network.netty.ClientStatisticsRequestStrategy] Node Node(1,prod-scoring-101:41414,[5,10,1,6,9,2,7,3,8,4],true) has a median response time of 463.640553. The cluster response time is 11.290323. Routing requests away temporarily.
It is not clear how the cluster median response time can differ from the server response time, considering I only have one node that process all the cluster requests.