I am trying to measure latency of each bolt in a topology. The latency numbers given by Storm is not enough as we want to calculate percentiles. In my current setup, I measure latency of the bolt by measuring time it takes to complete the execute method including call to emit. The assumption with above method is that, the collector's emit will return immediately without calling next bolt instances execute method even if the same executor is shared between current bolt instance and next bolt instance in the topology.