[
https://issues.apache.org/jira/browse/KAFKA-18196?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17926669#comment-17926669
]
A. Sophie Blee-Goldman commented on KAFKA-18196:
------------------------------------------------
I'm actually wondering if it even is a problem to turn this on for an existing
app...you already have two identical state stores so why would it be a problem
to just drop the copy? I guess this is only safe is you've named your
operators since removing the second store means we won't burn an index, and
this could impact the naming of downstream processors/stores/topics?
That's the only reason you can't turn this on, right? Though IIUC it would be
an incompatible change to disable it since one side would be missing data
(depends on the implementation details I suppose, without the optimization do
we actually produce a result for each side and get double output?)
> Reuse window store for stream-stream self-join
> ----------------------------------------------
>
> Key: KAFKA-18196
> URL: https://issues.apache.org/jira/browse/KAFKA-18196
> Project: Kafka
> Issue Type: Improvement
> Components: streams
> Reporter: A. Sophie Blee-Goldman
> Assignee: Kuan Po Tseng
> Priority: Major
>
> We can optimize the stream-stream self-join by using only a single copy of
> the state store. Right now this operator gets compiled down into two separate
> window stores for each side, even though both sides are identical.
> See https://github.com/apache/kafka/pull/18111#discussion_r1876844961
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
