[
https://issues.apache.org/jira/browse/FLINK-8578?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16650352#comment-16650352
]
Hequn Cheng commented on FLINK-8578:
------------------------------------
[~fhueske] Thanks for your suggestions. You are right. I think we need to
specify a field to order records. Even we restrict upsert mode to be same with
table mode, i.e, proc-time to proc-time and row-time to row-time. There is a
chance that records with different order share a same rowtime. The order field
can be an auto-increment id. It's type can also be a String.
We haven't discussed about how to handle time attributes after the upsert
conversion. I think we need to materialize the proc-time field or we can't
perform a correct retraction.
> Implement rowtime DataStream to Table upsert conversion.
> --------------------------------------------------------
>
> Key: FLINK-8578
> URL: https://issues.apache.org/jira/browse/FLINK-8578
> Project: Flink
> Issue Type: Sub-task
> Components: Table API & SQL
> Reporter: Hequn Cheng
> Assignee: Hequn Cheng
> Priority: Major
>
> Flink-8577 implements upsert from stream under proctime. This task is going
> to solve the order problem introduce by proctime. As proposed by Fabian in
> FLINK-8545, it would be good to be able to declare a time attribute that
> decides whether an upsert is performed or not.
> {code:java}
> Table table = tEnv.upsertFromStream(input, 'a, 'b.rowtime.upsertOrder, 'c.key)
> {code}
> This is a good way to solve the order problem using rowtime. And an idea
> comes to my mind that we can even remove the `.upsertOrder`, because the
> rowtime attribute can only be defined once in a table schema. Removing
> `.upsertOrder` also makes it easier to design api for TableSource and sql,
> i.e, we don't need to add another new feature for the api.
> Any suggestions are welcomed!
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
