Github user senorcarbone commented on the pull request: https://github.com/apache/flink/pull/643#issuecomment-98757999 I like the interfaces, they are quite precise. To sum up (and see if I get this right), with `Checkpointed` we simply guarantee that an operator will not get invoked while state is persisted unless it implements `CheckpointedAsynchronously`. In that case the operator should instead send a copy of its state. Regarding reconfiguration with arbitrary partitioning, a scheme that comes into my mind right now is to let functions to lazily request state on demand (without priorly injecting it) when they need to access it since we can't possible preallocate it. The first time they will try to access specific keys for example, only the associated state with these keys will be retrieved from the persistent recovery checkpoint. This way we let the reconfigured partitioners to decide on how state will be eventually allocated to instances.
--- 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. ---