[
https://issues.apache.org/jira/browse/FLINK-9422?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16493571#comment-16493571
]
Fabian Hueske commented on FLINK-9422:
--------------------------------------
As for most relational operators, the user should not need to worry about the
implementation.
If both inputs have the same time attribute, the first implementation is
chosen, otherwise the second. If users want to have more confidence about the
deduplication, they should configure more a more conservative watermark
strategy instead of waiting for two watermarks.
If a user really wants to use the non-watermarked implementation, they can cast
the time attribute (with WMs) into a regular timestamp.
> Dedicated operator for UNION on streaming tables with time attributes
> ---------------------------------------------------------------------
>
> Key: FLINK-9422
> URL: https://issues.apache.org/jira/browse/FLINK-9422
> Project: Flink
> Issue Type: Improvement
> Components: Table API & SQL
> Reporter: Fabian Hueske
> Assignee: Ruidong Li
> Priority: Minor
>
> We can implement a dedicated operator for a {{UNION}} operator on tables with
> time attributes. Currently, {{UNION}} is translated into a {{UNION ALL}} and
> a subsequent {{GROUP BY}} on all attributes without aggregation functions.
> The state of the grouping operator is only clean up using state retention
> timers.
> The dedicated operator would leverage the monotonicity property of the time
> attribute and watermarks to automatically clean up its state.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
