[
https://issues.apache.org/jira/browse/FLINK-3659?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15210655#comment-15210655
]
ASF GitHub Bot commented on FLINK-3659:
---------------------------------------
Github user StephanEwen commented on the pull request:
https://github.com/apache/flink/pull/1831#issuecomment-200951040
Thanks for describing this.
This has quite some big implications, as far as I can see it. The state in
the connected stream is now a "broadcast state" not partitioned, so allowing to
do that on the key/value state probably breaks some ongoing efforts, like
scaling, etc.
How about a more clean separation of these things:
1. The connected streams (fully partitioned or not)
2. Broadcast inputs, which are similar to the "side inputs" in cloud
dataflow or the broadcast variables in DataSet.
That gives us
- clean semantics, behavior that users can work with
- We do not overcomplicate the key/value abstraction
- and we can also make sure we checkpoint the broadcast state once only
(rather than on each parallel subtask).
> Allow ConnectedStreams to Be Keyed on Only One Side
> ---------------------------------------------------
>
> Key: FLINK-3659
> URL: https://issues.apache.org/jira/browse/FLINK-3659
> Project: Flink
> Issue Type: Improvement
> Components: Streaming
> Affects Versions: 1.0.0
> Reporter: Aljoscha Krettek
> Assignee: Aljoscha Krettek
>
> Currently, we only allow partitioned state when both inputs are keyed (with
> the same key type). I think a valid use case is to have only one side keyed
> and have the other side broadcast to publish some updates that are relevant
> for all keys.
> When relaxing the requirement to have only one side keyed we must still
> ensure that the same key is used if both sides are keyed.
> [~gyfora] Did you try this with what you're working on?
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
