[
https://issues.apache.org/jira/browse/FLINK-8158?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16271030#comment-16271030
]
ASF GitHub Bot commented on FLINK-8158:
---------------------------------------
Github user xccui commented on the issue:
https://github.com/apache/flink/pull/5094
Hi @hequn8128, as @fhueske explained, watermarks should not define the
semantics of the join. We just emit join results, in spite of whether they are
late and the join function should not care about the lateness. You can take
"holding back watermarks" as a special watermark assigner, based on which,
there could be out-of-ordered data (join results) being taken as late.
Currently, we just generates watermarks with a static delay that equals to the
larger relative offset. In the future, we should dynamically decide whether a
join result is late according to the real cache size.
> Rowtime window inner join emits late data
> -----------------------------------------
>
> Key: FLINK-8158
> URL: https://issues.apache.org/jira/browse/FLINK-8158
> Project: Flink
> Issue Type: Bug
> Components: Table API & SQL
> Reporter: Hequn Cheng
> Assignee: Hequn Cheng
> Attachments: screenshot-1xxx.png
>
>
> When executing the join, the join operator needs to make sure that no late
> data is emitted. Currently, this achieved by holding back watermarks.
> However, the window border is not handled correctly. For the sql bellow:
> {quote}
> val sqlQuery =
> """
> SELECT t2.key, t2.id, t1.id
> FROM T1 as t1 join T2 as t2 ON
> t1.key = t2.key AND
> t1.rt BETWEEN t2.rt - INTERVAL '5' SECOND AND
> t2.rt + INTERVAL '1' SECOND
> """.stripMargin
> val data1 = new mutable.MutableList[(String, String, Long)]
> // for boundary test
> data1.+=(("A", "LEFT1", 6000L))
> val data2 = new mutable.MutableList[(String, String, Long)]
> data2.+=(("A", "RIGHT1", 6000L))
> {quote}
> Join will output a watermark with timestamp 1000, but if left comes with
> another data ("A", "LEFT1", 1000L), join will output a record with timestamp
> 1000 which equals previous watermark.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
