dajac opened a new pull request #10318: URL: https://github.com/apache/kafka/pull/10318
The incremental FetchSessionCache sessions deprioritizes partitions where a response is returned. This may happen if log metadata such as log start offset, hwm, etc is returned, or if data for that partition is returned. When a fetch response fills to maxBytes, data may not be returned for partitions even if the fetch offset is lower than the fetch upper bound. However, the fetch response will still contain updates to metadata such as hwm if that metadata has changed. This can lead to degenerate behavior where a partition's hwm or log start offset is updated resulting in the next fetch being unnecessarily skipped for that partition. At first this appeared to be worse, as hwm updates occur frequently, but starvation should result in hwm movement becoming blocked, allowing a fetch to go through and then becoming unstuck. However, it'll still require one more fetch request than necessary to do so. Consumers may be affected more than replica fetchers, however they often remove partitions with fetched data from the next fetch request and this may be helping prevent starvation. ### Committer Checklist (excluded from commit message) - [ ] Verify design and implementation - [ ] Verify test coverage and CI build status - [ ] Verify documentation (including upgrade notes) ---------------------------------------------------------------- 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: us...@infra.apache.org
