@fhueske,hi, I think it's wrong. because state redistribution happens when the parallelism of the operator is changed. Conceptually, the state in the checkpoint is the concatenated list of all lists returned by the parallel transformation function instances. When restoring from a checkpoint, the list is divided into sub-lists that are assigned to each parallel function instance.
[ Full content available at: https://github.com/apache/flink/pull/6625 ] This message was relayed via gitbox.apache.org for [email protected]
