This basically wraps a named timer which (through code not shown here) creates a timer metric and queues it to be sent to Riemann. There's nothing magical here except the point I was making about being able to nREPL to the server. If we are trying to live-debug a performance issue, I can connect to the server and wrap the code I'm interested in a with-timer. We'll immediately get live metrics into Riemann and then into elastic search for those pretty charts I showed in kibana. When we're done, I can remove the with-timer.
I find this brilliant. It doesn't require making a lot of toggle-able metrics in advance. It doesn't require any tooling fancier than nREPL. And, it doesn't require any server downtime.