Github user xccui commented on the issue: https://github.com/apache/flink/pull/4625 Hi @fhueske, the PR has been updated. Temporarily, I keep the logic for dealing with the late data, as well as the fine-grained cache. For the late data semantics problem, I think we need to rethink it and make a final decision (maybe we should consult others). For the cache optimization problem, I want to leave it a future work.
---