Github user eshioji commented on the pull request:
https://github.com/apache/storm/pull/568#issuecomment-107578216
@revans2 Re: pending metric, I see. I removed it thinking that it
exclusively had to do with the `pendingMessage` field, but now I realise it was
actually tracking messages inside Netty's internal buffer, too. I brought it
back; should be equivalent to the old metrics.
Re: performance, I couldn't see any performance difference on my test
cluster (5 node, processes around 60K tuples per sec) under "normal"
conditions. With `v0.9.4` it would start to go haywire after several hours and
start to consume lots of CPU and a lot of tuples start to fail. That symptom
seems to have been alleviated with the patch. However, this topology feeds off
a live data stream so admittedly it's not a strict comparison (because the
volume of data is not completely the same from run to run). Is there a standard
way you guys use to verify performance regression?
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
