WencongLiu commented on code in PR #23362: URL: https://github.com/apache/flink/pull/23362#discussion_r1357721747
########## docs/content/docs/dev/datastream/dataset_migration.md: ########## @@ -0,0 +1,699 @@ +--- +title: "How To Migrate From DataSet to DataStream" +weight: 302 +type: docs +bookToc: false +aliases: + - /dev/dataset_migration.html +--- +<!-- +Licensed to the Apache Software Foundation (ASF) under one +or more contributor license agreements. See the NOTICE file +distributed with this work for additional information +regarding copyright ownership. The ASF licenses this file +to you under the Apache License, Version 2.0 (the +"License"); you may not use this file except in compliance +with the License. You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, +software distributed under the License is distributed on an +"AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +KIND, either express or implied. See the License for the +specific language governing permissions and limitations +under the License. +--> + +# How to Migrate from DataSet to DataStream + +The DataSet API has been formally deprecated and will no longer receive active maintenance and support. It will be removed in the +Flink 2.0 version. Flink users are recommended to migrate from the DataSet API to the DataStream API, Table API and SQL for their +data processing requirements. + +For the most of DataSet APIs, the users can utilize the DataStream API to get the same calculation result in the batch jobs. However, +different DataSet API can be implemented by DataStream API with various difference on semantic and behavior. All DataSet APIs can be +categorized into four types: + +Category 1: These DataSet APIs can be implemented by DataStream APIs with same semantic and same calculation behavior. + +Category 2: These DataSet APIs can be implemented by DataStream APIs with different semantic but same calculation behavior. This will +make the job code more complex. + +Category 3: These DataSet APIs can be implemented by DataStream APIs with different semantic and different calculation behavior. This +will involve additional computation and shuffle costs. + +Category 4: These DataSet APIs are not supported by DataStream APIs. + +The subsequent sections will first introduce how to set the execution environment and provide detailed explanations on how to implement +each category of DataSet APIs using the DataStream APIs, highlighting the specific considerations and challenges associated with each +category. + + +## Setting the execution environment + +To execute a DataSet pipeline by DataStream API, we should first start by replacing ExecutionEnvironment with StreamExecutionEnvironment. +{{< tabs executionenv >}} +{{< tab "DataSet">}} +```java +ExecutionEnvironment env = ExecutionEnvironment.getExecutionEnvironment(); +``` +{{< /tab >}} +{{< tab "DataStream">}} +```java +StreamExecutionEnvironment executionEnvironment = StreamExecutionEnvironment.getExecutionEnvironment(); +``` +{{< /tab >}} +{{< /tabs>}} + +As the source of DataSet is always bounded, the execution mode is suggested to be set to RuntimeMode.BATCH to allow Flink to apply +additional optimizations for batch processing. +```java +StreamExecutionEnvironment executionEnvironment = StreamExecutionEnvironment.getExecutionEnvironment(); +executionEnvironment.setRuntimeMode(RuntimeExecutionMode.BATCH); +``` + +## Implement the DataSet API by DataStream + +### Category 1 + +For Category 1, the usage of the API in DataStream is almost identical to that in DataSet. This means that implementing these +DataSet APIs by the DataStream API is relatively straightforward and does not require significant modifications or complexity +in the job code. + +#### Map + +{{< tabs mapfunc >}} +{{< tab "DataSet">}} +```java +dataSet.map(new MapFunction(){ + // implement user-defined map logic +}); +``` +{{< /tab >}} +{{< tab "DataStream">}} +```java +dataStream.map(new MapFunction(){ + // implement user-defined map logic +}); +``` +{{< /tab >}} +{{< /tabs>}} + + +#### FlatMap + +{{< tabs flatmapfunc >}} +{{< tab "DataSet">}} +```java +dataSet.flatMap(new FlatMapFunction(){ + // implement user-defined flatmap logic +}); +``` +{{< /tab >}} +{{< tab "DataStream">}} +```java +dataStream.flatMap(new FlatMapFunction(){ + // implement user-defined flatmap logic +}); +``` +{{< /tab >}} +{{< /tabs>}} + +#### Filter + +{{< tabs filterfunc >}} +{{< tab "DataSet">}} +```java +dataSet.filter(new FilterFunction(){ + // implement user-defined filter logic +}); +``` +{{< /tab >}} +{{< tab "DataStream">}} +```java +dataStream.filter(new FilterFunction(){ + // implement user-defined filter logic +}); +``` +{{< /tab >}} +{{< /tabs>}} + +#### Union + +{{< tabs unionfunc >}} +{{< tab "DataSet">}} +```java +dataSet1.union(dataSet2); +``` +{{< /tab >}} +{{< tab "DataStream">}} +```java +dataStream1.union(dataStream2); +``` +{{< /tab >}} +{{< /tabs>}} + + +#### Rebalance + +{{< tabs rebalancefunc >}} +{{< tab "DataSet">}} +```java +dataSet.rebalance(); +``` +{{< /tab >}} +{{< tab "DataStream">}} +```java +dataStream.rebalance(); +``` +{{< /tab >}} +{{< /tabs>}} + +#### Reduce on Grouped DataSet + +{{< tabs reducegroupfunc >}} +{{< tab "DataSet">}} +```java +DataSet<Tuple2<String, Integer>> dataSet = // [...] +dataSet.groupBy(value -> value.f0) + .reduce(new ReduceFunction(){ + // implement user-defined reduce logic + }); +``` +{{< /tab >}} +{{< tab "DataStream">}} +```java +DataStream<Tuple2<String, Integer>> dataStream = // [...] +dataStream.keyBy(value -> value.f0) + .reduce(new ReduceFunction(){ + // implement user-defined reduce logic + }); +``` +{{< /tab >}} +{{< /tabs>}} + +#### Aggregate on Grouped DataSet + +{{< tabs aggregategroupfunc >}} +{{< tab "DataSet">}} +```java +DataSet<Tuple2<String, Integer>> dataSet = // [...] +// compute sum of the second field +dataSet.groupBy(value -> value.f0).aggregate(SUM, 1); +// compute min of the second field +dataSet.groupBy(value -> value.f0).aggregate(MIN, 1); +// compute max of the second field +dataSet.groupBy(value -> value.f0).aggregate(MAX, 1); +``` +{{< /tab >}} +{{< tab "DataStream">}} +```java +DataStream<Tuple2<String, Integer>> dataStream = // [...] +// compute sum of the second field +dataStream.keyBy(value -> value.f0).sum(1); +// compute min of the second field +dataStream.keyBy(value -> value.f0).min(1); +// compute max of the second field +dataStream.keyBy(value -> value.f0).max(1); +``` +{{< /tab >}} +{{< /tabs>}} + + +### Category 2 + +For category 2, these DataSet APIs can be implemented by DataStream APIs with different semantic but same calculation behavior. +The developers need to adapt their code to accommodate these variations, which introduces additional complexity. + +#### Project + +{{< tabs projectfunc>}} +{{< tab "DataSet">}} +```java +DataSet<Tuple3<Integer, Double, String>> dataSet = // [...] +dataSet.project(2,0); +``` +{{< /tab >}} +{{< tab "DataStream">}} +```java +DataStream<Tuple3<Integer, Double, String>> dataStream = // [...] +dataStream.map(value -> Tuple2.of(value.f2, value.f0)); +``` +{{< /tab >}} +{{< /tabs>}} + +#### Distinct + +{{< tabs distinctfunc>}} +{{< tab "DataSet">}} +```java +DataSet<Integer> dataSet = // [...] +dataSet.distinct(); +``` +{{< /tab >}} +{{< tab "DataStream">}} +```java +DataStream<Integer> dataStream = // [...] +DataStream<Integer> output = dataStream + .keyBy(value -> value) + .reduce((value1, value2) -> value1); +``` +{{< /tab >}} +{{< /tabs>}} + +#### Hash-Partition + +{{< tabs hashpartitionfunc>}} +{{< tab "DataSet">}} +```java +DataSet<Tuple2<String, Integer>> dataSet = // [...] +dataSet.partitionByHash(value -> value.f0); +``` +{{< /tab >}} +{{< tab "DataStream">}} +```java +DataStream<Tuple2<String, Integer>> dataStream = // [...] +// partition by the hashcode of key +dataStream.partitionCustom((key, numSubpartition) -> key.hashCode() % numSubpartition, value -> value.f0); +``` +{{< /tab >}} +{{< /tabs>}} + +#### Reduce on Full DataSet + +If developers want to compute data of full datastream, GlobalWindow could be used to collect all records of datastream. +However, a special trigger is also required to trigger the computation of GlobalWindow at the end of its inputs. Here is an example +code snippet of the trigger. + +```java +public class EOFTrigger extends Trigger<Object, GlobalWindow> { + + private boolean hasRegistered; + + @Override + public TriggerResult onElement( + Object element, long timestamp, GlobalWindow window, TriggerContext ctx) { + if (!hasRegistered) { + ctx.registerEventTimeTimer(Long.MAX_VALUE); + hasRegistered = true; + } + return TriggerResult.CONTINUE; + } + + @Override + public TriggerResult onEventTime(long time, GlobalWindow window, TriggerContext ctx) { + return TriggerResult.FIRE; + } + + @Override + public TriggerResult onProcessingTime(long time, GlobalWindow window, TriggerContext ctx) { + return TriggerResult.CONTINUE; + } + + @Override + public void clear(GlobalWindow window, TriggerContext ctx) throws Exception {} +} +``` +Then the reduce operation on full datastream could be performed by EOFTrigger. + +{{< tabs reducefullfunc>}} +{{< tab "DataSet">}} +```java +DataSet<String> dataSet = // [...] +dataSet.reduce(new ReduceFunction(){ + // implement user-defined reduce logic + }); +``` +{{< /tab >}} +{{< tab "DataStream">}} +```java +DataStream<String> dataStream = // [...] +dataStream.windowAll(GlobalWindows.create()).trigger(new EOFTrigger()) + .reduce(new ReduceFunction(){ + // implement user-defined reduce logic + }); +``` +{{< /tab >}} +{{< /tabs>}} + + +#### Aggregate on Full DataSet + +The aggregate on full datastream could also be performed by EOFTrigger. + +{{< tabs aggregatefullfunc>}} +{{< tab "DataSet">}} +```java +DataSet<Tuple2<Integer, Integer>> dataSet = // [...] +// compute sum of the second field +dataSet.aggregate(SUM, 1); +// compute min of the second field +dataSet.aggregate(MIN, 1); +// compute max of the second field +dataSet.aggregate(MAX, 1); +``` +{{< /tab >}} +{{< tab "DataStream">}} +```java +DataStream<Tuple2<Integer, Integer>> dataStream = // [...] +// compute sum of the second field +dataStream.windowAll(GlobalWindows.create()).trigger(new EOFTrigger()).sum(1); +// compute min of the second field +dataStream.windowAll(GlobalWindows.create()).trigger(new EOFTrigger()).min(1); +// compute max of the second field +dataStream.windowAll(GlobalWindows.create()).trigger(new EOFTrigger()).max(1); +``` +{{< /tab >}} +{{< /tabs>}} + + +#### GroupReduce on Full DataSet + +The grpup reduce on full datastream could also be performed by EOFTrigger. + +{{< tabs groupreducefullfunc>}} +{{< tab "DataSet">}} +```java +DataSet<Integer> dataSet = // [...] +dataSet.reduceGroup(new GroupReduceFunction(){ + // implement user-defined group reduce logic + }); +``` +{{< /tab >}} +{{< tab "DataStream">}} +```java +DataStream<Integer> dataStream = // [...] +// assign a same timestamp to all records +assignSameTimestamp(dataStream); +dataStream.windowAll(GlobalWindows.create()).trigger(new EOFTrigger()) + .apply(new WindowFunction(){ + // implement user-defined group reduce logic + }); +``` +{{< /tab >}} +{{< /tabs>}} + +### Category 3 + +For category 3, these DataSet APIs can be implemented by DataStream APIs with different semantic and different calculation behavior. Additional +calculation steps will be added. + +To collect records from each key, the DataStream assigns a same timestamp to all records and uses a fixed-length time window to gather them. +However, this introduces additional computational costs due to timestamp processing. + +Here is an example code snippet. The function assignSameTimestamp is used to explain the detailed behavior and will be utilized in the +subsequent sections: +```java +// assign a same timestamp to all records +void assignSameTimestamp(DataStream<Tuple2<String, Integer>> dataStream) { + dataStream.assignTimestampsAndWatermarks( + WatermarkStrategy.<Tuple2<String, Integer>>forMonotonousTimestamps() + .withTimestampAssigner((event, timestamp) -> 0)); +} +``` + +To collect records from each subtask, every record needs to be assigned a unique subtask ID and grouped accordingly within the window. +This additional step of assigning the subtask ID and performing a groupby operation introduces shuffle costs. Here is an example code +snippet showing how to assign a subtask ID to each record. The function assignSubtaskID is used to explain the detailed behavior and will +be utilized in the subsequent sections: +```java +// assign subtask ID to all records +DataStream<Tuple2<String, Integer>> assignSubtaskID(DataStream<Integer> dataStream) { + return dataStream.map(new RichMapFunction<Integer, Tuple2<String, Integer>>() { + @Override + public Tuple2<String, Integer> map(Integer value) { + return Tuple2.of(String.valueOf(getRuntimeContext().getIndexOfThisSubtask()), value); + } + }); +} +``` + +#### MapPartition/SortPartition + +{{< tabs mapsortpartitionfunc>}} +{{< tab "DataSet">}} +```java +DataSet<Integer> dataSet = // [...] +// MapPartition +dataSet.mapPartition(new MapPartitionFunction(){ + // implement user-defined map partition logic + }); +// SortPartition +dataSet.sortPartition(0, Order.ASCENDING); +dataSet.sortPartition(0, Order.DESCENDING) +``` +{{< /tab >}} +{{< tab "DataStream">}} +```java +DataStream<Integer> dataStream = // [...] +// assign subtask ID to all records +DataStream<Tuple2<String, Integer>> dataStream1 = assignSubtaskID(dataStream); +// assign a same timestamp to all records +assignSameTimestamp(dataStream1); +dataStream1.keyBy(value -> value.f0) + .window(TumblingEventTimeWindows.of(Time.seconds(1))) + .apply(new WindowFunction(){ + // implement user-defined map partition or sort partition logic + }); +``` +{{< /tab >}} +{{< /tabs>}} + +#### GroupReduce on Grouped DataSet + +{{< tabs groupreducefunc>}} +{{< tab "DataSet">}} +```java +DataSet<Tuple2<Integer, String>> dataSet = // [...] +dataSet.groupBy(value -> value.f0) + .reduceGroup(new GroupReduceFunction(){ + // implement user-defined group reduce logic + }); +``` +{{< /tab >}} +{{< tab "DataStream">}} +```java +DataStream<Tuple2<String, Integer>> dataStream = // [...] +// assign a same timestamp to all records +assignSameTimestamp(dataStream); +dataStream.keyBy(value -> value.f0) + .window(TumblingEventTimeWindows.of(Time.seconds(1))) + .apply(new WindowFunction(){ + // implement user-defined group reduce logic + }); +``` +{{< /tab >}} +{{< /tabs>}} + +#### Join + +{{< tabs joinfunc>}} +{{< tab "DataSet">}} +```java +DataSet<Tuple2<String, Integer>> dataSet1 = // [...] +DataSet<Tuple2<String, Integer>> dataSet2 = // [...] +dataSet1.join(dataSet2) + .where(data -> data.f0) + .equalTo(data -> data.f0) + .with(new JoinFunction(){ + // implement user-defined join logic + }); +``` +{{< /tab >}} +{{< tab "DataStream">}} +```java +DataStream<Tuple2<String, Integer>> dataStream1 = // [...] +DataStream<Tuple2<String, Integer>> dataStream2 = // [...] +// assign a same timestamp to all records +assignSameTimestamp(dataStream1); +assignSameTimestamp(dataStream2); +dataStream1.join(dataStream2) + .where(value -> value.f0) + .equalTo(value -> value.f0) + .window(TumblingEventTimeWindows.of(Time.seconds(1))) + .apply(new JoinFunction(){ + // implement user-defined join logic + }); +``` +{{< /tab >}} +{{< /tabs>}} + +The Join operator can be efficiently implemented using the Table object in the Table API. The Table object allows seamless +conversion from a DataStream. Once the join computation on the Table is completed, the Table object can be converted back to +a DataStream. Here is an example code snippet: Review Comment: The examples that utilize the Table API has been removed. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@flink.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
