@bbejeck That's a good question. Personally I think we should make strict guarantees that if `ProcessorContext.commit()` is called, we make sure the state is committed before the next record for this task is going to be processed, because some application logic may rely on that (think: I did some external changes, and hence I do want to make sure the state is committed right aware before processing the next record). And hence the implementation of `maybeCommitPerUserRequested` in stream thread actually always check for `commitRequest` and `commitNeeded` while ignoring the commit interval.
Could you point out which part of the doc indicates that `ProcessorContext.commit()` is more of a suggestion? Maybe we do need to update it. [ Full content available at: https://github.com/apache/kafka/pull/5428 ] This message was relayed via gitbox.apache.org for [email protected]
