There's nothing built in to measure the time a tuple sits in the disruptor queues, or the latency between emission and reception currently. You could approximate it somewhat by attaching a timestamp at emit time and measuring when the tuple is received in the bolt. You could have the same tuples go to another bolt that does nothing but measure latency to reduce the effects of real work causing queuing.
Michael Rose (@Xorlev <https://twitter.com/xorlev>) Senior Platform Engineer, FullContact <http://www.fullcontact.com/> [email protected] On Fri, Nov 14, 2014 at 10:44 AM, Nick Katsipoulakis <[email protected]> wrote: > Hello, > > I was trying to use the built-in Metrics of Storm to calculate average > latency for each bolt. I define latency as the difference between the time > that a tuple is emitted from a bolt, and the time this tuple was inserted > in the Wait Queue of that bolt. Is there any built-in metric to give me > this difference? If no, is there any way that I can calculate it? > > Thank you, > Nick > >
