[ 
https://issues.apache.org/jira/browse/KAFKA-12242?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17288646#comment-17288646
 ] 

John Roesler commented on KAFKA-12242:
--------------------------------------

Thanks for raising this, [~guozhang] , I have seen several people getting 
tripped up by this issue.

 

Just to throw it out there, another approach to solve this would be to begin to 
go down the road I proposed here: 
[https://cwiki.apache.org/confluence/display/KAFKA/Kafka+Streams+DSL+Grammar]

Specifically, what I have in mind is that attaching serdes to a processor node 
is logically independent from requesting materialization of that node's view. 
This ticket highlights the awkwardness of using a cross-cutting "Materialized" 
config at all. In contrast, if it were just "MapValuesParameters" with 
independent settings for specifying serdes and for requesting materialization, 
it wouldn't be an issue.

I definitely don't insist on the proposed grammar, but wanted to document the 
relationship with that proposal.

> Decouple state store materialization enforcement from name/serde provider
> -------------------------------------------------------------------------
>
>                 Key: KAFKA-12242
>                 URL: https://issues.apache.org/jira/browse/KAFKA-12242
>             Project: Kafka
>          Issue Type: Improvement
>          Components: streams
>            Reporter: Guozhang Wang
>            Priority: Major
>              Labels: needs-kip
>
> Many users of Streams would want the following: let the Streams runtime to 
> decide whether or not to materialize a state store; AND if it decides to do 
> so, use the store name / serdes I provided ahead of time, if not, then 
> nothing happens (the provided store name and serdes can just be dropped).
> However, Streams today take `Materialized` as an indicator to enforce the 
> materialization. We should think of a way for users to optionally decouple 
> materialization enforcement from name/serde provider.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

Reply via email to