junrao commented on PR #17700: URL: https://github.com/apache/kafka/pull/17700#issuecomment-2576037114
@kirktrue : Thanks for the experimental results. What's the reason for the high eviction rate at the beginning of all tests? Regarding the options. The fetch session is designed for the case that a client will consume all previously fetched data before issuing the next fetch request. Option 4 fits this design the best and doesn't unnecessarily increase the fetch request size. It does lose the ability to prefetch a partition when another partition from the same node has unconsumed data in the client. However, I am not sure that's super important since It still allows for parallelism between the client consuming the last fetched data and the server serving the next request. Could you do some consumer latency test to see if there is any noticeable impact with option 4? -- 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: jira-unsubscr...@kafka.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
