Hello Thank you for the KIP.
Could you please add the following items to the KIP to make it easier to review. 1. Please add some more details on "Especially for Kafka Stream, we have seen many cases when users got bit by the current behavior.". What were the scenarios that impacted users of Kafka and what was the impact? This would help us understand the motivation for this change. 2. What are the side effects of this approach and why are they an acceptable trade-off? e.g. the solution increases the record size on the disk etc. 3. What is the reason for which we rejected the alternative of increasing the retention time? (probably covered by answer to 1 above) 4. How is max event time calculated on broker startup? Is it persisted somewhere or is it kept in-memory and updated with every record? 5. How do we guarantee "keep at max" since data eviction is asynchronous in nature and may not provide "at max" guarantee. It can provide "at least" guarantee. -- Divij Vaidya On Wed, Sep 28, 2022 at 3:47 PM Николай Ижиков <nizhi...@apache.org> wrote: > Hello! > > Please, share your feedback? > Is proposed retention policy can be useful for Kafka users? > > > 21 сент. 2022 г., в 13:02, Николай Ижиков <nizhikov....@gmail.com> > написал(а): > > > > Hello. > > > > I want to start discussion of KIP-870 [1] [2]. > > > > The goal of this KIP is to provide new retention policy purely based on > record event time. > > > > [1] > https://cwiki.apache.org/confluence/display/KAFKA/KIP-870%3A+Retention+policy+based+on+record+event+time > > [2] https://issues.apache.org/jira/browse/KAFKA-13866 > > > >
