For what it's worth, the Jira for KIP-101 was created in January 2014. That has been a known potential Kafka data loss scenario for quite a while, just took some time (and evidently the findings of these new stress tests) to be prioritized as a serious problem that needed to be fixed.