[
https://issues.apache.org/jira/browse/FLINK-38201?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18013666#comment-18013666
]
Xuyang Zhong commented on FLINK-38201:
--------------------------------------
Hi, [~twalthr] I have some doubts about whether we can remove the sink
materializer from the retract sink without loss. Assuming we have an upstream
node with the upsert key `b`, writing to a sink where the primary key is column
`a`, and the sink is a retract sink.
If two subtasks on the upstream node produce the following data: subtask1 sends
`+I(a1, b1)` and `-U(a1, b1)`; subtask2 sends `+U(a1, b2)`. What will be the
final result?
In fact, there will be several possible sequences of data that the sink can
receive:
* If the sink receives `+I(a1, b1)`, `-U(a1, b1)`, and `+U(a1, b2)` in that
order, then regardless of whether there is a sink materializer node, we will
ultimately get `a1, b2`.
* If the sink receives `+I(a1, b1)`, `+U(a1, b2)`, and `-U(a1, b1)` in that
order, and if there is a sink materializer, the result will be `a1, b2`. But
what is the current result after removing the sink materializer?
* If the sink receives `+U(a1, b2)`, `+I(a1, b1)`, and `-U(a1, b1)` in that
order, and if there is a sink materializer, the result will be `a1, b2`. But
what is the current result after removing the sink materializer?
> SinkUpsertMaterializer should not be inserted for retract sinks
> ---------------------------------------------------------------
>
> Key: FLINK-38201
> URL: https://issues.apache.org/jira/browse/FLINK-38201
> Project: Flink
> Issue Type: Bug
> Components: Table SQL / Planner
> Affects Versions: 2.1.0
> Reporter: Timo Walther
> Assignee: Timo Walther
> Priority: Major
> Labels: pull-request-available
> Fix For: 2.2.0
>
>
> Currently, the logic in FlinkChangelogModeInferenceProgram for enabling
> upsert materialize does not distinguish between retract and upsert. Thus,
> SinkUpsertMaterializer is added for retract sinks. The resulting changelog is
> incorrect as it misses -U. Retract is flexible enough to simply be passed to
> the sink without further changes.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
