AHeise commented on pull request #16990: URL: https://github.com/apache/flink/pull/16990#issuecomment-906360736
> I've left one more comment. > > Do you know/suspect the reason behind the remaining 7-10% regression? I'm not sure if we should be merging this and then trying to resolve the remaining problems or if maybe we should revert the feature and fix it? The more commits we are adding on top of the current version, the more problematic it will be to investigate additional problems/revert this change if it will come to it. Baseline is around 32k and we now come back to 30.5k. That's 5% regression for me. Master is currently at 27.2k or 15%. So it's already a huge improvement. I'm also not convinced that we get it to 0%; we simply do more stuff on the hotpath to provide a better user experience. Ultimately, when there is actual I/O involved (we are talking about a source), it will not be noticeable. Current benchmark implies that new sources are 2x faster than old sources, which we cannot measure at all on actual sources. There is a clear limitation how far you get with synth benchmarks. On the other hand, it saved my ass now for this topic twice already, so don't get me wrong: I love them. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
