Flink 如何分流資料

ipoo發表於2020-06-11

場景

獲取流資料的時候,通常需要根據所需把流拆分出其他多個流,根據不同的流再去作相應的處理。

舉個例子:建立一個商品實時流,商品有季節標籤,需要對不同標籤的商品做統計處理,這個時候就需要把商品資料流根據季節標籤分流。

分流方式

使用Filter分流
使用Split分流
使用Side Output分流

如何分流

先模擬一個實時的資料流

import lombok.Data;
@Data
public class Product {
    public Integer id;
    public String seasonType;
}

自定義Source

import common.Product;
import org.apache.flink.streaming.api.functions.source.SourceFunction;

import java.util.ArrayList;
import java.util.Random;

public class ProductStremingSource implements SourceFunction<Product> {
    private boolean isRunning = true;

    @Override
    public void run(SourceContext<Product> ctx) throws Exception {
        while (isRunning){
            // 每一秒鐘產生一條資料
            Product product = generateProduct();
            ctx.collect(product);
            Thread.sleep(1000);
        }
    }

    private Product generateProduct(){
        int i = new Random().nextInt(100);
        ArrayList<String> list = new ArrayList();
        list.add("spring");
        list.add("summer");
        list.add("autumn");
        list.add("winter");
        Product product = new Product();
        product.setSeasonType(list.get(new Random().nextInt(4)));
        product.setId(i);
        return product;
    }
    @Override
    public void cancel() {

    }
}

輸出:

使用Filter分流

使用 filter 運算元根據資料的欄位進行過濾。

import common.Product;
import org.apache.flink.streaming.api.datastream.DataStreamSource;
import org.apache.flink.streaming.api.datastream.SingleOutputStreamOperator;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
import source.ProductStremingSource;

public class OutputStremingDemo {
    public static void main(String[] args) throws Exception {
        StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();

        DataStreamSource<Product> source = env.addSource(new ProductStremingSource());

        // 使用Filter分流
        SingleOutputStreamOperator<Product> spring = source.filter(product -> "spring".equals(product.getSeasonType()));
        SingleOutputStreamOperator<Product> summer = source.filter(product -> "summer".equals(product.getSeasonType()));
        SingleOutputStreamOperator<Product> autumn  = source.filter(product -> "autumn".equals(product.getSeasonType()));
        SingleOutputStreamOperator<Product> winter  = source.filter(product -> "winter".equals(product.getSeasonType()));
        source.print();
        winter.printToErr();

        env.execute("output");
    }
}

結果輸出(紅色為季節標籤是winter的分流輸出):

使用Split分流

重寫OutputSelector內部類的select()方法,根據資料所需要分流的型別反正不同的標籤下,返回SplitStream,通過SplitStream的select()方法去選擇相應的資料流。

只分流一次是沒有問題的，但是不能使用它來做連續的分流。

SplitStream已經標記過時了

public class OutputStremingDemo {
    public static void main(String[] args) throws Exception {
        StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();

        DataStreamSource<Product> source = env.addSource(new ProductStremingSource());

        // 使用Split分流
        SplitStream<Product> dataSelect = source.split(new OutputSelector<Product>() {
            @Override
            public Iterable<String> select(Product product) {
                List<String> seasonTypes = new ArrayList<>();
                String seasonType = product.getSeasonType();
                switch (seasonType){
                    case "spring":
                        seasonTypes.add(seasonType);
                        break;
                    case "summer":
                        seasonTypes.add(seasonType);
                        break;
                    case "autumn":
                        seasonTypes.add(seasonType);
                        break;
                    case "winter":
                        seasonTypes.add(seasonType);
                        break;
                    default:
                        break;
                }
                return seasonTypes;
            }
        });
        DataStream<Product> spring = dataSelect.select("machine");
        DataStream<Product> summer = dataSelect.select("docker");
        DataStream<Product> autumn = dataSelect.select("application");
        DataStream<Product> winter = dataSelect.select("middleware");
        source.print();
        winter.printToErr();

        env.execute("output");
    }
}

使用Side Output分流

推薦使用這種方式

首先需要定義一個OutputTag用於標識不同流

可以使用下面的幾種函式處理流傳送到分流中:

ProcessFunction
KeyedProcessFunction
CoProcessFunction
KeyedCoProcessFunction
ProcessWindowFunction
ProcessAllWindowFunction

之後再用getSideOutput(OutputTag)選擇流。

public class OutputStremingDemo {
    public static void main(String[] args) throws Exception {
        StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();

        DataStreamSource<Product> source = env.addSource(new ProductStremingSource());

        // 使用Side Output分流
        final OutputTag<Product> spring = new OutputTag<Product>("spring");
        final OutputTag<Product> summer = new OutputTag<Product>("summer");
        final OutputTag<Product> autumn = new OutputTag<Product>("autumn");
        final OutputTag<Product> winter = new OutputTag<Product>("winter");
        SingleOutputStreamOperator<Product> sideOutputData = source.process(new ProcessFunction<Product, Product>() {
            @Override
            public void processElement(Product product, Context ctx, Collector<Product> out) throws Exception {
                String seasonType = product.getSeasonType();
                switch (seasonType){
                    case "spring":
                        ctx.output(spring,product);
                        break;
                    case "summer":
                        ctx.output(summer,product);
                        break;
                    case "autumn":
                        ctx.output(autumn,product);
                        break;
                    case "winter":
                        ctx.output(winter,product);
                        break;
                    default:
                        out.collect(product);
                }
            }
        });

        DataStream<Product> springStream = sideOutputData.getSideOutput(spring);
        DataStream<Product> summerStream = sideOutputData.getSideOutput(summer);
        DataStream<Product> autumnStream = sideOutputData.getSideOutput(autumn);
        DataStream<Product> winterStream = sideOutputData.getSideOutput(winter);

        // 輸出標籤為:winter 的資料流
        winterStream.print();

        env.execute("output");
    }
}

結果輸出: