wsry edited a comment on pull request #15199: URL: https://github.com/apache/flink/pull/15199#issuecomment-805025446
@StephanEwen Thanks for the review and comments. **About Lazy Initialization** I agree that initializing the ExecutorService eagerly is better if it already handles lazy thread spawning and termination. **About Buffer Request Batching** I have already tested the default configuration with tpcds benchmark. Compared with my previous implementation which uses exclusive buffers and exclusive thread per result partition, there is no performance regression. Besides, batch reading is important for IO performance especially on HDD. **Buffer Request Batching** is a simple way to guarantee that the reader can always get enough buffers for sequential read, otherwise, all readers may get a few buffers and read in parallel which can be bad for performance. We may can implement a method which accepts a "size" parameter and can allocate any size batching for the requester. However, we do not need such a method now and implementing it can introduce extra complexity. I think we can implement it when we need it. -- 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. For queries about this service, please contact Infrastructure at: [email protected]
