Github user revans2 commented on the issue:
https://github.com/apache/storm/pull/2203
The new overhead for metrics_v2_replace_meters_to_counters_with_sampling
looks like it is in an acceptable range for me. Right about a 1% difference
for latency. (0.97% to 1.29%) and right about a 1.8% to 1.9% difference for
CPU usage. It is also good to see that even if we turn off the sub sampling
there is not too much of an increase in the overhead.
I would be +1 for merging this in (from a performance perspective still
need to look at the code in more depth), assuming that we can get similar
results for 2.x and with a plan to have the old API deprecated in 2.x and
completely removed in a 3.x release.
We also need to make sure that we have the naming convention figured out or
some kind of tag support so that we can properly do rollups for different types
of metrics.
---
