Github user HeartSaVioR commented on the issue:
https://github.com/apache/storm/pull/2399
Looks like the patch addresses a bit different point rather than issue
description in STORM-2793.
Below is the issue description in STORM-2793:
> Existing Storm metrics track tuple counts, but don't account for the size
of the tuple payload. If a task always gets large tuples, it can be slower
than a task that gets small tuples without showing any reason why. It would be
good to track the byte counts as well so data skew can be observed directly.
from the description I imagined that the issue would want to track the
tuples in bytes for every tasks (maybe with sampling), but this patch only
addresses tracking `transferred` tuples, which ignores local tasks.
I can still see the benefit with measuring transferred tuples, but just
like to check the patch fits origin intention for the issue.
In addition, currently we allow to put map to value in metric and let
metrics collector handles it in any way. It should be changed, but we could
resolve it along with other metrics later.
Overall looks good to me, and it would be much better if we could see the
performance impact and confirm it's small enough.
---
