You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by "Alexander Smirnov (Jira)" <ji...@apache.org> on 2023/04/05 07:41:00 UTC

[jira] [Commented] (FLINK-17337) Send UPDATE messages instead of INSERT and DELETE in streaming join operator

    [ https://issues.apache.org/jira/browse/FLINK-17337?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17708727#comment-17708727 ] 

Alexander Smirnov commented on FLINK-17337:
-------------------------------------------

Hi [~jark]! I've found a [bug|https://issues.apache.org/jira/browse/FLINK-31729]  with the current logic of LEFT/RIGHT/FULL OUTER JOIN (sometimes UPDATE_BEFORE can still be emitted). I decided to implement ticket FLINK-17337 right away, because with this optimization done, bug won't be present anymore. I've create a [PR|[https://github.com/apache/flink/pull/22351].] Can you take a look at it?

> Send UPDATE messages instead of INSERT and DELETE in streaming join operator
> ----------------------------------------------------------------------------
>
>                 Key: FLINK-17337
>                 URL: https://issues.apache.org/jira/browse/FLINK-17337
>             Project: Flink
>          Issue Type: Improvement
>          Components: Table SQL / Runtime
>            Reporter: Jark Wu
>            Priority: Minor
>              Labels: auto-deprioritized-major, pull-request-available
>             Fix For: 1.18.0
>
>
> Currently, streaming join operator always send INSERT and DELETE messages for simplification if it's not inner join. However, we can send UPDATE_BEFORE and UPDATE_AFTER messages instead of INSERT and DELETE. For example, when we recieve right record "b", then we can send {{UB[a, null]}} and {{UA[a,b]}} instead of {{D[a,null]}}, {{I[a,b]}}. This is an optimization, because UB can be omitted in some cases to reduce IO cost and computation. 



--
This message was sent by Atlassian Jira
(v8.20.10#820010)