[
https://issues.apache.org/jira/browse/FLINK-36371?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Eason Ye updated FLINK-36371:
-----------------------------
Description:
The pipeline is Source(Kafka serialize msg to RowData type) -> FlatMap
--(keyby)–> TumbleProcessWindow -> sink.
The Source would emit two fields: UserName String, and Amount Integer.
FlatMap collects the BinaryRowData type record and emits it to the
TumbleProcessWindow operator.
KeySelector is using the UserName BinaryStringData. key by code is :
{code:java}
.keyBy(rowData -> rowData.getString(0)){code}
The window result is unexpected, the same username records arrived at
TumbleProcessWindow simultaneously, but these records were calculated in the
different windows.
When I use the below keyBy, the window result is correct.
{code:java}
.keyBy(rowData -> rowData.getString(0).toString()){code}
Reproduction:
{code:java}
env.fromSource(
kafkaSource,
WatermarkStrategy.noWatermarks(),
sourceName())
.flatMap(new FlatMapFunction<RowData, RowData>() {
public void flatMap(RowData value, Collector<RowData> out) throws
Exception {
System.out.println("is instanceof BinaryRowData =" + (value
instanceof BinaryRowData));
out.collect(value);
}
})
.keyBy((KeySelector<RowData, StringData>) value -> value.getString(0))
.window(TumblingProcessingTimeWindows.of(Time.seconds(10)))
.process(new FlinkTumbleWindowingProcessFunction())
.print(); {code}
was:
The pipeline is Source(Kafka serialize msg to RowData type) -> FlatMap
--(keyby)–> TumbleProcessWindow -> sink.
The Source would emit two fields: UserName String, and Amount Integer.
FlatMap collects the BinaryRowData type record and emits it to the
TumbleProcessWindow operator.
KeySelector is using the UserName BinaryStringData. key by code is :
{code:java}
.keyBy(rowData -> rowData.getString(0)){code}
The window result is unexpected, the same username records arrived at
TumbleProcessWindow simultaneously, but these records were calculated in the
different windows.
When I use the below keyBy, the window result is correct.
{code:java}
.keyBy(rowData -> rowData.getString(0).toString()){code}
> Field BinaryStringData as keySelector would lead the same records hash
> shuffle into different tumble windows
> ------------------------------------------------------------------------------------------------------------
>
> Key: FLINK-36371
> URL: https://issues.apache.org/jira/browse/FLINK-36371
> Project: Flink
> Issue Type: Improvement
> Components: API / Type Serialization System
> Affects Versions: 2.0.0
> Reporter: Eason Ye
> Priority: Major
>
> The pipeline is Source(Kafka serialize msg to RowData type) -> FlatMap
> --(keyby)–> TumbleProcessWindow -> sink.
> The Source would emit two fields: UserName String, and Amount Integer.
> FlatMap collects the BinaryRowData type record and emits it to the
> TumbleProcessWindow operator.
> KeySelector is using the UserName BinaryStringData. key by code is :
> {code:java}
> .keyBy(rowData -> rowData.getString(0)){code}
> The window result is unexpected, the same username records arrived at
> TumbleProcessWindow simultaneously, but these records were calculated in the
> different windows.
> When I use the below keyBy, the window result is correct.
> {code:java}
> .keyBy(rowData -> rowData.getString(0).toString()){code}
>
> Reproduction:
>
> {code:java}
> env.fromSource(
> kafkaSource,
> WatermarkStrategy.noWatermarks(),
> sourceName())
> .flatMap(new FlatMapFunction<RowData, RowData>() {
> public void flatMap(RowData value, Collector<RowData> out) throws
> Exception {
> System.out.println("is instanceof BinaryRowData =" + (value
> instanceof BinaryRowData));
> out.collect(value);
> }
> })
> .keyBy((KeySelector<RowData, StringData>) value -> value.getString(0))
> .window(TumblingProcessingTimeWindows.of(Time.seconds(10)))
> .process(new FlinkTumbleWindowingProcessFunction())
> .print(); {code}
>
>
>
>
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
