[
https://issues.apache.org/jira/browse/FLINK-21763?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Alexis Sarda-Espinosa updated FLINK-21763:
------------------------------------------
Description:
In the DataSet API, an undocumented (?) feature is using
[SplitDataProperties|[https://ci.apache.org/projects/flink/flink-docs-release-1.12/api/java/org/apache/flink/api/java/io/SplitDataProperties.html]]
to specify different grouping/partitioning/sorting properties, so if the data
is pre-sorted (e.g. when reading from a database), some operations can be
optimized. Based on this, I assume the Flink scheduler already knows about such
optimizations.
The new batch mode for the Data Stream API doesn't have a way to specify this.
I imagine something like this would be useful even for transformation
operators, not only for sources, e.g. a flat map that gets an input "batch" and
outputs new data with a different kind of grouping/sorting.
I suppose these directives would have to be ignored for the normal stream mode.
was:
In the DataSet API, an undocumented (?) feature is using
[SplitDataProperties|[https://ci.apache.org/projects/flink/flink-docs-release-1.12/api/java/org/apache/flink/api/java/io/SplitDataProperties.html]]
to specify different grouping/partitioning/sorting properties, so if the data
is pre-sorted (e.g. when reading from a database), some operations can be
optimized. Based on this, I assume the Flink scheduler already knows about such
optimizations.
The new batch mode for the Data Stream API doesn't have a way to specify this.
I imagine something like this would be useful even for transformation
operators, not only for sources, e.g. a flat map that gets an input "batch" and
outputs new data with a different kind of grouping/sorting.
I suppose these diractives would have to be ignored for the normal stream mode.
> Extend Data Stream's batch mode to handle partitioning/grouping/ordering
> ------------------------------------------------------------------------
>
> Key: FLINK-21763
> URL: https://issues.apache.org/jira/browse/FLINK-21763
> Project: Flink
> Issue Type: Improvement
> Components: API / DataStream
> Reporter: Alexis Sarda-Espinosa
> Priority: Not a Priority
> Labels: auto-deprioritized-major, auto-deprioritized-minor
>
> In the DataSet API, an undocumented (?) feature is using
> [SplitDataProperties|[https://ci.apache.org/projects/flink/flink-docs-release-1.12/api/java/org/apache/flink/api/java/io/SplitDataProperties.html]]
> to specify different grouping/partitioning/sorting properties, so if the
> data is pre-sorted (e.g. when reading from a database), some operations can
> be optimized. Based on this, I assume the Flink scheduler already knows about
> such optimizations.
> The new batch mode for the Data Stream API doesn't have a way to specify
> this. I imagine something like this would be useful even for transformation
> operators, not only for sources, e.g. a flat map that gets an input "batch"
> and outputs new data with a different kind of grouping/sorting.
> I suppose these directives would have to be ignored for the normal stream
> mode.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
