[
https://issues.apache.org/jira/browse/FLINK-26534?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
lincoln lee updated FLINK-26534:
--------------------------------
Description:
FLINK-20370 fix the wrong result when sink primary key is not the same with
query and introduced a new auto-keyby sink's primary key strategy for append
stream if the sink's parallelism differs from input stream's.
But still exists one case to be solved:
for a changelog stream, its changelog upsert key same as sink's primary key,
but sink's parallelism changed by user (via those sinks which implement the
`ParallelismProvider` interface, e.g., KafkaDynamicSink), we should fix it.
And a minor change: keyby canbe omitted when sink has single parallism (because
none partitioner will cause worse disorder)
was:
FLINK-20370 fix the wrong result when sink primary key is not the same with
query and introduced a new auto-keyby sink's primary key strategy for append
stream if the sink's parallelism differs from input stream's.
But still exists one case to be solved:
for a changelog stream, its changelog upsert key same as sink's primary key,
but sink's parallelism changed by user (via those sinks which implement the
`ParallelismProvider` interface, e.g., KafkaDynamicSink), we should fix it.
> shuffle by sink's primary key should cover the case that input changelog
> stream has a different parallelism
> -----------------------------------------------------------------------------------------------------------
>
> Key: FLINK-26534
> URL: https://issues.apache.org/jira/browse/FLINK-26534
> Project: Flink
> Issue Type: Bug
> Components: Table SQL / Planner
> Affects Versions: 1.15.0
> Reporter: lincoln lee
> Priority: Minor
>
> FLINK-20370 fix the wrong result when sink primary key is not the same with
> query and introduced a new auto-keyby sink's primary key strategy for append
> stream if the sink's parallelism differs from input stream's.
> But still exists one case to be solved:
> for a changelog stream, its changelog upsert key same as sink's primary key,
> but sink's parallelism changed by user (via those sinks which implement the
> `ParallelismProvider` interface, e.g., KafkaDynamicSink), we should fix it.
> And a minor change: keyby canbe omitted when sink has single parallism
> (because none partitioner will cause worse disorder)
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
