divijvaidya commented on PR #13312:
URL: https://github.com/apache/kafka/pull/13312#issuecomment-1536165039
Thank you @showuon and @ijuma for your time and patience in this review
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to
divijvaidya commented on PR #13312:
URL: https://github.com/apache/kafka/pull/13312#issuecomment-1523106081
@ijuma this is ready for your review. All failing tests are unrelated.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to
divijvaidya commented on PR #13312:
URL: https://github.com/apache/kafka/pull/13312#issuecomment-1515034487
@ijuma I tried end to end benchmarks w/ and w/o this change but I didn't
observe any noticeable change in end to end latency or throughput (because the
bottle neck in end to end
divijvaidya commented on PR #13312:
URL: https://github.com/apache/kafka/pull/13312#issuecomment-1508353737
@ijuma @mimaison please take a look at this PR. I would ideally like to have
this part of 3.5. Note that existing tests associated with ByteUtil conversion
pass and I have added new
divijvaidya commented on PR #13312:
URL: https://github.com/apache/kafka/pull/13312#issuecomment-1506707381
Rebased from trunk to fix the unrelated integration test failures.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub
divijvaidya commented on PR #13312:
URL: https://github.com/apache/kafka/pull/13312#issuecomment-1495898660
> There is a hollow VarInt implementation here that is similarly unrolled on
the `write` path but is I think a little easier to read!
>
>
divijvaidya commented on PR #13312:
URL: https://github.com/apache/kafka/pull/13312#issuecomment-1448306949
Tagging a few folks who may be interested in reviewing this: @ijuma
@jasonk000 @hachikuji. Kindly review when you get an opportunity.
--
This is an automated message from the
divijvaidya commented on PR #13312:
URL: https://github.com/apache/kafka/pull/13312#issuecomment-1448299260
> It is interesting that unrolling the loop while keeping the same
underlying logic yields this increase in throughput for writing varints
Yes, indeed. In non hotspot code