gaoyunhaii commented on issue #7368: [FLINK-10742][network] Let Netty use Flink's buffers directly in credit-based mode URL: https://github.com/apache/flink/pull/7368#issuecomment-585495323 Very thanks @zhijiangW for the review! As a whole I agree with the thought to reduce the complexity of the current model. However I think the `NettyMessageParser` is still required, since the parser has its own intermediate states, and if we push the parsing logic into the message directly, these state will be mixed with the message fields. We could keep the main thought of split the procedure for different messages directly, but move the parsing logic into the parser instead of the messages. I think in this way we could keep the benefit and do not introduce the above problem. **The only drawback compared with the initial version is that now we need to keep two message header buffers. However, since the buffer size is only dozens of bytes, it should not cause application of new chunk and thus should not affect the total direct memory used. ** I have updated the PR according to the above logic. :)
---------------------------------------------------------------- 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] With regards, Apache Git Services
