Github user roshannaik commented on the issue:
https://github.com/apache/storm/pull/2241
Based on the comments, I have the following summary of concerns raised:
1. Better handling of low throughput Topos: There appear to be two sub
issues.
- CPU usage is high
- Higher latency was noticed.
2. TVL topo: Able to run this topo with similar or better performance
compared to master. In particular concerns over CPU usage and latencies. Here I
feel the default settings are not suited for the modes its being run in. But
overall this is related to pt# 1 IMO
3. Bug in Multi worker mode prevents inter-worker communication.
4. Concern over TODOs noted in code. These seem to be giving an impression
that there is lots unfinished work. Actually these are imprtant things to do
but none of them are big tasks. The intention is to address them all soon in
this PR. There are maybe two TODOS in code for perf bottlenecks that I dont
plan to address. They were hard to find, so I felt its valuable to retain them
alongside the offending line of code. If people prefer to remoes these TODOs, I
can do that.
5. Some "real-world topology" runs as in addition to benchmark style topos.
We may find more issues as this progresses, but here is my assessment of
the known pending work:
1. Fix the high CPU usage for low throughput topos. I think this needs a
sleep strategy in the bolt. The spout already has it. Right now it does busy
polling.
2. Look into (and fix whatever is causing) the higher latencies reported
for TVL in low throughput mode
3. Fix the multi-worker mode bug
4. Fix all the TODOs noted
5. Get some more TVL runs and share results
6. Some more runs with other topos and share results
---
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 infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
