STORM-2557 updated the arrival_rate disruptor queue metric to go off of individual tuples, and not batches of tuples like it did before. But the sojourn time is also calculated from the arrival rate and the population. But the population is computed based off of slots, not tuples. So it is now off proportionally to how many tuples are in a given slot.
We either need a way to compute the tuple population and use that (which would be a new metric), or we need to keep the old arrival rate metric around to, just for the sojourn time.
I think the first one is more likely to work out.