Github user StephanEwen commented on the pull request: https://github.com/apache/flink/pull/1527#issuecomment-181922703 @wangyangjun We actually merged all the changes concerning the state abstraction. To make this window join work seamlessly on Flink's state backends (memory, or key/value stores, managed memory, ...) you would need to implement it against the key/value state. That means that whenever you store data in the operator, the data should go into the partitioned state that you can obtain from the `AbstractStreamOperator` or the `RuntimeContext`. I think that for this window operator, the `ListState` is a good choice, where you can add values to a key and retrieve the list as a whole once the windows are evaluated. Please write back if you need some more pointers on the state abstraction.
--- If your project is set up for it, you can reply to this email and have your reply appear on GitHub as well. If your project does not have this feature enabled and wishes so, or if the feature is enabled but not working, please contact infrastructure at infrastruct...@apache.org or file a JIRA ticket with INFRA. ---