为什么Apache Flink从数据流中删除事件？

Question

在以下单元测试用例中，将生成由numberOfElements指定的某些事件，并将其作为数据流提供。 该单位案例在生产线上随机失败。

assertEquals（numberOfElements，CollectSink.values.size（））;

为什么Apache Flink跳过事件的任何解释。

import org.apache.flink.api.java.tuple.Tuple2;
import org.apache.flink.streaming.api.datastream.DataStream;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
import org.apache.flink.streaming.api.functions.sink.SinkFunction;
import org.junit.Before;
import org.junit.Test;

import java.util.ArrayList;
import java.util.List;

import static java.lang.Thread.sleep;
import static org.junit.Assert.assertEquals;

public class FlinkTest {

StreamExecutionEnvironment env;

@Before
public void setup() {
    env = StreamExecutionEnvironment.createLocalEnvironment();
}

@Test
public void testStream1() throws Exception {
    testStream();
}

@Test
public void testStream2() throws Exception {
    testStream();
}

@Test
public void testStream3() throws Exception {
    testStream();
}

@Test
public void testStream4() throws Exception {
    testStream();
}


@Test
public void testStream() throws Exception {

    final int numberOfElements = 50;

    DataStream<Tuple2<String, Integer>> tupleStream = env.fromCollection(getCollectionOfBucketImps(numberOfElements));
    CollectSink.values.clear();
    tupleStream.addSink(new CollectSink());
    env.execute();
    sleep(2000);

    assertEquals(numberOfElements, getCollectionOfBucketImps(numberOfElements).size());
    assertEquals(numberOfElements, CollectSink.values.size());
}


public static List<Tuple2<String, Integer>> getCollectionOfBucketImps(int numberOfElements) throws InterruptedException {
    List<Tuple2<String, Integer>> records = new ArrayList<>();
    for (int i = 0; i < numberOfElements; i++) {
        records.add(new Tuple2<>(Integer.toString(i % 10), i));
    }
    return records;
}

// create a testing sink
private static class CollectSink implements SinkFunction<Tuple2<String, Integer>> {

    public static final List<Tuple2<String, Integer>> values = new ArrayList<>();

    @Override
    public synchronized void invoke(Tuple2<String, Integer> value, Context context) throws Exception {
        values.add(value);
    }
 }
}

例如，testStreamX情况之一随机失败。

上下文：代码以8作为并行度setu运行，因为运行它的cpu具有8个内核

Answer 1

我不知道您的工作是否与众不同（我想这是Flink可以分配的最高限额）。 看起来您可以在接收器的增加值上具有“竞争”条件。

解

我已经运行了您的示例代码，将环境并行度设置为1，并且一切正常。 有关测试的文档示例使用此解决方案链接到文档 。

@Before
public void setup() {
    env = StreamExecutionEnvironment.createLocalEnvironment();
    env.setParallelism(1);
}

更好

您只能在接收器运算符上将并行度设置为1，并保持其余管道的并行度。 在以下示例中，我为tha映射运算符添加了一个额外的映射函数，强制并行度为8。

public void testStream() throws Exception {

    final int numberOfElements = 50;

    DataStream<Tuple2<String, Integer>> tupleStream = env.fromCollection(getCollectionOfBucketImps(numberOfElements));
    CollectSink.values.clear();
    tupleStream
            .map(new MapFunction<Tuple2<String,Integer>, Tuple2<String,Integer>>() {
                @Override
                public Tuple2<String,Integer> map(Tuple2<String, Integer> stringIntegerTuple2) throws Exception {

                    stringIntegerTuple2.f0 += "- concat something";

                    return stringIntegerTuple2;
                }
            }).setParallelism(8)
            .addSink(new CollectSink()).setParallelism(1);
    env.execute();
    sleep(2000);

    assertEquals(numberOfElements, getCollectionOfBucketImps(numberOfElements).size());
    assertEquals(numberOfElements, CollectSink.values.size());
}

Answer 2

当环境的并行计算大于1时，将有多个CollectSink实例，这有可能导致竞争状况。

这些是避免竞争情况的解决方案：

1. 在类对象上同步

private static class CollectSink implements SinkFunction<Tuple2<String, Integer>> {

    public static final List<Tuple2<String, Integer>> values = new ArrayList<>();

    @Override
    public void invoke(Tuple2<String, Integer> value, Context context) throws Exception {
        synchronized(CollectSink.class) {
            values.add(value);
        }
    }
 }

2. Collections.synchronizedList()

import java.util.Collections;
private static class CollectSink implements SinkFunction<Tuple2<String, Integer>> {

    public static final List<Tuple2<String, Integer>> values = Collections.synchronizedList(new ArrayList<>());

    @Override
    public void invoke(Tuple2<String, Integer> value, Context context) throws Exception {
        values.add(value);
    }
 }