[
https://issues.apache.org/jira/browse/FLINK-2405?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14642597#comment-14642597
]
ASF GitHub Bot commented on FLINK-2405:
---------------------------------------
Github user tillrohrmann commented on the pull request:
https://github.com/apache/flink/pull/936#issuecomment-125175011
I don't know of a better solution than the current approach if we want to
support the `stateByKey` parameter. I agree with Aljoscha that
`ds.mapWithState({ (x, o) => (x, o) }, true)` does not look really nice and you
will always have it if you have a multi line UDF.
I was wondering whether we can get rid of the `stateByKey` parameter. Is
there any use case in which you have a `KeyedDataStream` and set `stateByKey`
to false? Otherwise we could override the `*WithState` methods in
`KeyedDataStream` and per default the `StatefulFunction` sets the `partitioned`
field to `false`?
Other than that, the PR looks really good and I like it a lot :-)
> Add stateful transformations using lambdas to the Streaming Scala API
> ---------------------------------------------------------------------
>
> Key: FLINK-2405
> URL: https://issues.apache.org/jira/browse/FLINK-2405
> Project: Flink
> Issue Type: New Feature
> Components: Scala API, Streaming
> Reporter: Gyula Fora
> Assignee: Gyula Fora
> Priority: Minor
>
> I propose to extend the Streaming Scala API methods (map, flatmap, filter
> etc) with versions that take stateful functions as lambdas. This would allow
> for a nice functional way of defining stateful transformations:
> Example:
> def mapWithState( fun: (I, Option[S]) => (O, Option[S]), stateByKey: Boolean
> = false)
> The stateByKey optional flag allows the user to partition the state by key
> when applied on a KeyedDataStream.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
