[jira] [Commented] (SPARK-38212) Streaming dropDuplicates should remove out-of-watermark keys

["Apache Spark (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/SPARK-38212?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17492386#comment-17492386 ] 

Apache Spark commented on SPARK-38212:
--------------------------------------

User 'viirya' has created a pull request for this issue:
https://github.com/apache/spark/pull/35521

> Streaming dropDuplicates should remove out-of-watermark keys
> ------------------------------------------------------------
>
>                 Key: SPARK-38212
>                 URL: https://issues.apache.org/jira/browse/SPARK-38212
>             Project: Spark
>          Issue Type: Improvement
>          Components: Structured Streaming
>    Affects Versions: 3.3.0
>            Reporter: L. C. Hsieh
>            Priority: Major
>
> Currently streaming dropDuplicates simply stores all states forever if no watermark-attached key exists.
> For streaming stateful operator, this seems counterintuitive behavior if we consider other stateful operations such as streaming joining.
> It also means the number of state will grow up infinitely as we observe in real application. But it doesn't make sense to add a time column into dedup keys because we don't really deduplicate rows by [key, time] but only by [key].
> More reasonable streaming dedup seems to remove out-of-watermark states if the input data (not key) has watermark. This will be a behavior change, for streaming queries with dedup operation.



--
This message was sent by Atlassian Jira
(v8.20.1#820001)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org