Github user roshannaik commented on the issue:
https://github.com/apache/storm/pull/2241
@revans2 , Much thanks for spending your cycles on this kind of deep
testing. Its really valuable to build confidence and achieve the stability we
desire.
I see one key issue with this latest test run. Please disable
max.spout.pending for 2306.It is not needed anymore and will be removed shortly
in my updates to this PR. Noted in design doc. I think your values of 1000 &
1500 are constraining its throughput. You can set the max.spout.pending to
whatever suits best for the master.
- What were the values used for **topology.producer.batch.size** &
**topology.flush.tuple.freq.millis** in the 2306 runs ? For low throughput
testing, batchSize of 1 is the right setting for 2306.
Suggestion...Since ACKer is a known bottleneck, I think ... 2 acker, 1
spout 1 split bolt 1 count might be a more suitable version of 'flavor A' for a
4 core machine.
The Throughput vs CPU is worth re-measuring once I provide the fix for the
busy polling.
I am in agreement with all that you said in that post about user's thought
process and TVL's use case. For 2306, I find TVL unsuitable for gauging
throughput limits due to the spout throttling. But it appears, that it is not
your goal at the moment.
---
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.
---
