[ https://issues.apache.org/jira/browse/FLINK-4844?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15582774#comment-15582774 ]
ASF GitHub Bot commented on FLINK-4844: --------------------------------------- Github user aljoscha commented on a diff in the pull request: https://github.com/apache/flink/pull/2648#discussion_r83664806 --- Diff: flink-core/src/main/java/org/apache/flink/api/common/functions/RuntimeContext.java --- @@ -48,7 +43,7 @@ * {@link AbstractRichFunction#getRuntimeContext()}. */ @Public -public interface RuntimeContext { +public interface RuntimeContext extends KeyedStateStore { --- End diff -- I think it would be better to not have `RuntimeContext` be a `KeyedStateStore`. In the not-so-far future `RuntimeContext` will probably provide a `KeyedStateStore` or at least use one internally to implement the state methods. Properly separating the two now seems prudent. > Partitionable Raw Keyed/Operator State > -------------------------------------- > > Key: FLINK-4844 > URL: https://issues.apache.org/jira/browse/FLINK-4844 > Project: Flink > Issue Type: New Feature > Reporter: Stefan Richter > Assignee: Stefan Richter > > Partitionable operator and keyed state are currently only available by using > backends. However, the serialization code for many operators is build around > reading/writing their state to a stream for checkpointing. We want to provide > partitionable states also through streams, so that migrating existing > operators becomes more easy. -- This message was sent by Atlassian JIRA (v6.3.4#6332)