Flink-03 DataStream

  1. Flink DataStream
    1. DataStream相关概念

5.1.1 ExecutionEnvironment执行环境

  1. 执行环境创建方式

和Flink交互需要一个入口,这个入口就是ExecutionEnvironment执行环境。在Stream API中,它的执行环境就使用StreamExecutionEnvironment来创建,里面包含了创建各种执行环境的静态方法。

Flink-03 DataStream

这里这些静态方法都可以创建执行环境,我们最常用的就是getExecutionEnvironment方法,它会根据实际的执行环境来判断是运行在Local还是集群上。

      // 创建流处理环境

        StreamExecutionEnvironment.getExecutionEnvironment();

        StreamExecutionEnvironment.getExecutionEnvironment(new Configuration());

        // 创建一个本地运行环境

        StreamExecutionEnvironment.createLocalEnvironment(2,new Configuration());

        // 创建一个本地带WebUI的执行环境,需要引入flink-runtime-web的依赖

        StreamExecutionEnvironment.createLocalEnvironmentWithWebUI(new Configuration());

        // 创建远程执行环境,给定远程地址将任务直接提交到集群上,需要提供jar包

        StreamExecutionEnvironment.createRemoteEnvironment("hadoop01",6123);

  1. 远程执行RemoteEnv

示例:

// 因为下面用到了一个类AddKeyMapFunction这是用户自己定义的一个类,这个类默认Flink的jar包环境里面是没有的,现在需要先将这个类打成一个jar包,

// 下面创建环境第三个参数填入这个jar包的路径,运行时就可以将jar包上传给Flink,任务加载到就可以开始运行了

        StreamExecutionEnvironment env = StreamExecutionEnvironment.createRemoteEnvironment(

                "hadoop01",

                8081,

                "com.example.flink.datastream/target/com.example.flink.datastream-0.0.1-SNAPSHOT.jar"

        );

        DataStreamSource<String> source = env.socketTextStream("hadoop01", 9999);

        source.map(new AddKeyMapFunction()).print();

        try {

            env.execute();

        } catch (Exception e) {

            e.printStackTrace();

        }

  1. 异步提交任务

DataStream的内部操作是懒执行的,要想触发执行动作需要执行execute动作。这里execute()方法是线程阻塞的,可以通过executeAsync()来异步提交任务,后续可以通过其返回值jobClient来监控任务状态。

try {

            // 如果是execution()程序就会卡在这里

            // 可以使用executeAsync()异步提交,这样程序就不会卡在这里,后续可以通过jobClient来监控任务状态

            JobClient jobClient = env.executeAsync();

            ExecutorService executor = Executors.newSingleThreadExecutor();

            executor.execute(() -> {

                while (true) {

                    try {

                        JobStatus jobStatus = jobClient.getJobStatus().get();

                        if (!JobStatus.RUNNING.equals(jobStatus)) break;

                        else {

                            TimeUnit.SECONDS.sleep(1);

                            System.out.println("运行状态:" + jobStatus.name());

                        }

                    } catch (InterruptedException e) {

                        e.printStackTrace();

                    } catch (ExecutionException e) {

                        e.printStackTrace();

                    }

                }

            });

        } catch (Exception e) {

            e.printStackTrace();

        }

5.1.2 什么是DataStream

DataStream API 得名于一个特殊的DataStream类,该类用于表示 Flink 程序中的数据集合。您可以将它们视为可以包含重复项的不可变数据集合。这些数据可以是有限的,也可以是无限的,用于处理它们的 API 是相同的。

在用法上DataStream与常规 Java 相似,Collection但在某些关键方面却大不相同。它们是不可变的,这意味着一旦它们被创建,你就不能添加或删除元素。您也不能简单地检查内部元素,而只能使用DataStreamAPI 操作(也称为转换)处理它们。

大概的意思就是DataStream就是一个大的数据集向Collection一样,但是这个数据集内的数据不能被改变只能用它提供出来的API进行转换。

    1. DataStreamAPI操作

5.2.1 Connector连接器

一些比较基本的 Source 和 Sink 已经内置在 Flink 里。 预定义 data sources 支持从文件、目录、socket,以及 collections 和 iterators 中读取数据。 预定义 data sinks 支持把数据写入文件、标准输出(stdout)、标准错误输出(stderr)和 socket。

以下是1.12版本Flink社区已经开发完成的连接器source是输入sink是输出。

  1. 内部数据源读取

Flink-03 DataStream

以上都是可以从内部数据源读取数据的API

  1. 文本读取

        StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();

        DataStreamSource<String> source = env.readTextFile("data/streaming/AFINN-111.txt");

        source.print();

        env.execute();

  1. Socket读取

        StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();

        DataStreamSource<String> source = env.socketTextStream("hadoop01", 9999);

        source.print();

        env.execute();

  1. Kafka读取

 StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();

        String topic= "sensor";

        Properties consumerConfig = new Properties();

        consumerConfig.setProperty(ConsumerConfig.BOOTSTRAP_SERVERS_CONFIG,"hadoop01:9092");

        consumerConfig.setProperty(ConsumerConfig.KEY_DESERIALIZER_CLASS_CONFIG, StringDeserializer.class.getName());

        consumerConfig.setProperty(ConsumerConfig.VALUE_DESERIALIZER_CLASS_CONFIG, StringDeserializer.class.getName());

        consumerConfig.setProperty(ConsumerConfig.AUTO_OFFSET_RESET_CONFIG,"latest");

        consumerConfig.setProperty(ConsumerConfig.GROUP_ID_CONFIG,"flink_consumer");

        DataStreamSource<String> source = env.addSource(new FlinkKafkaConsumer<String>(topic,new SimpleStringSchema(),consumerConfig));

        source.print();

        env.execute();

  1. 自定义Source

 public static void main(String[] args) throws Exception {

        StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();

        String topic = "sensor";

        Properties consumerConfig = new Properties();

        consumerConfig.setProperty(ConsumerConfig.BOOTSTRAP_SERVERS_CONFIG, "hadoop01:9092");

        consumerConfig.setProperty(ConsumerConfig.KEY_DESERIALIZER_CLASS_CONFIG, StringDeserializer.class.getName());

        consumerConfig.setProperty(ConsumerConfig.VALUE_DESERIALIZER_CLASS_CONFIG, StringDeserializer.class.getName());

        consumerConfig.setProperty(ConsumerConfig.AUTO_OFFSET_RESET_CONFIG, "latest");

        consumerConfig.setProperty(ConsumerConfig.GROUP_ID_CONFIG, "flink_consumer");

        DataStreamSource<String> source = env.addSource(new MySource(topic, consumerConfig));

        source.print();

        env.execute();

    }

    /**

     * 自定义一个Source用来消费kafka数据

     */

    static class MySource implements SourceFunction<String> {

        private final String topic;

        private final Properties config;

        private volatile boolean run = true;

        public MySource(String topic, Properties config) {

            this.topic = topic;

            this.config = config;

            this.run = true;

        }

        @Override

        public void run(SourceContext<String> ctx) throws Exception {

            KafkaConsumer<String, String> consumer = new KafkaConsumer<String, String>(config);

            consumer.subscribe(Collections.singletonList(this.topic));

            while (run) {

                ConsumerRecords<String, String> consumerRecords = consumer.poll(Duration.ofSeconds(10));

                for (ConsumerRecord<String, String> consumerRecord : consumerRecords) {

                    ctx.collect(consumerRecord.value());

                }

            }

        }

        @Override

        public void cancel() {

            this.run = false;

        }

    }

        1. Sink
  1. Kafka Sink

        // 从Socket接收数据发送到Kafka

        StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();

        DataStreamSource<String> source = env.socketTextStream("hadoop01", 9999);

        String brokerList = "hadoop01:9092";

        String sendTopic = "sensor";

        source.addSink(new FlinkKafkaProducer<String>(brokerList, sendTopic, new SimpleStringSchema()));

        env.execute();

  1. JDBC Sink

        // 从Socket接收数据写入到mysql

        StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();

        DataStreamSource<String> source = env.socketTextStream("hadoop01", 9999);

        String writeSql = "INSERT INTO tbl2(value) VALUES(?)";

        source.addSink(JdbcSink.sink(writeSql,

                (JdbcStatementBuilder<String>) (preparedStatement, s) -> {

                    preparedStatement.setObject(1, s);

                },

                // JDBC是以批的方式写入的,这里改下批次大小好看到效果

                new JdbcExecutionOptions.Builder().withBatchSize(1)

                        .build()

                , new JdbcConnectionOptions.JdbcConnectionOptionsBuilder()

                        .withDriverName("com.mysql.jdbc.Driver")

                        .withUsername("root")

                        .withPassword("")

                        .withUrl("jdbc:mysql:///test")

                        .build()

        ));

        env.execute();

      1. Operators
        1. Transform数据流转换
  1. Map & FatMap

        StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();

        env.setParallelism(1);

        DataStreamSource<String> source = env.socketTextStream("hadoop01", 9999);

        // 在接收到的消息前加个前缀打印出来

        SingleOutputStreamOperator<String> result = source.map(new MapFunction<String, String>() {

            private static final String PREFIX = "input message is : ";

            @Override

            public String map(String value) throws Exception {

                return PREFIX.concat(value);

            }

        });

        result.print("map result:");

        SingleOutputStreamOperator<String> flatMapResult = source.flatMap(new FlatMapFunction<String, String>() {

            @Override

            public void flatMap(String value, Collector<String> out) throws Exception {

                for (String word : value.split(" ")) {

                    out.collect(word);

                }

            }

        });

        flatMapResult.print("flatmap result:");

        env.execute();

  1. Fiter

      StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();

        env.setParallelism(1);

        DataStreamSource<String> source = env.socketTextStream("hadoop01", 9999);

        // 在接收到的消息前加个前缀打印出来

        SingleOutputStreamOperator<String> result = source.filter(new FilterFunction<String>() {

            @Override

            public boolean filter(String value) throws Exception {

                // 对内容进行过滤,如果消息的长度超过10就被滤掉

                return StringUtils.length(value) <= 10;

            }

        });

        result.print("长度不超过10的消息:");

        env.execute();

  1. Union

      StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();

        env.setParallelism(1);

        DataStreamSource<String> hadoop01Stream = env.socketTextStream("hadoop01", 9999);

        DataStreamSource<String> hadoop02Stream = env.socketTextStream("hadoop02", 9999);

        // 将两个相同泛型的DataStream Union到一起联合处理

        DataStream<String> unionResult = hadoop01Stream.union(hadoop02Stream);

        unionResult.print("union stream:");

        env.execute();

  1. Connect

        StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();

        env.setParallelism(1);

        DataStreamSource<String> hadoop01Stream = env.socketTextStream("hadoop01", 9999);

        DataStream<Integer> hadoop02Stream = env.socketTextStream("hadoop02", 9999).map(Integer::parseInt);

        // 将两个流connect到一起,不管泛型一不一致

        ConnectedStreams<String, Integer> connect = hadoop01Stream.connect(hadoop02Stream);

        // 后续对两个流的操作就是两个方法单独处理都是Co开头的Function最终整合成一个流

        SingleOutputStreamOperator<String> result = connect.map(new CoMapFunction<String, Integer, String>() {

            @Override

            public String map1(String value) throws Exception {

                return value;

            }

            @Override

            public String map2(Integer value) throws Exception {

                return value.toString();

            }

        });

        result.print("connect stream : ");

        env.execute();

  1. KeyBy

keyBy算子是用来分组的算子,返回结果就是KeyedStream键控流,后面的Reduce算子Aggravate算子都是需要分组之后才能操作。

        // 对source进行分组,WordCount统计单词个数

        KeyedStream<Tuple2<String, Long>, String> wordKeyedStream = wordStream.keyBy(new KeySelector<Tuple2<String, Long>, String>() {

            @Override

            public String getKey(Tuple2<String, Long> tuple2) throws Exception {

                return tuple2.f0;

            }

        });

  1. Reduce

Reduce累计,接收的是KeyedStream,按照什么Key来汇总,group by后的聚合操作。

        SingleOutputStreamOperator<Tuple2<String, Long>> reduceStream = wordKeyedStream.reduce(new ReduceFunction<Tuple2<String, Long>>() {

            @Override

            public Tuple2<String, Long> reduce(Tuple2<String, Long> value1, Tuple2<String, Long> value2) throws Exception {

                return new Tuple2<>(value1.f0, value1.f1 + value2.f1);

            }

        });

  1. Aggregations

Aggregations操作就是分组之后的聚合操作,简单的聚合例如sum,count,min,max之类的函数。

        KeyedStream<Word, String> wordSumKeyedStream = wordStream.keyBy(new KeySelector<Word, String>() {

            @Override

            public String getKey(Word value) throws Exception {

                return "长度比较";

            }

        });

        // 再看哪些单词长度长

        SingleOutputStreamOperator<Word> minByResult = wordSumKeyedStream.minBy("len");

        SingleOutputStreamOperator<Word> maxByResult = wordSumKeyedStream.maxBy("len");

        minByResult.print("minByResult:");

        maxByResult.print("maxByResult:");

        env.execute();

        1. 分区

前面运行架构中有数据从上游到下游的发送方式,可以是一对一可以是重新分发。

重新分发又有几种分发方式,用户定义分发方式、随机分发、平衡随机分发、组内随机、下游每个分区一份。

  • RebalancePartitioner

Rebalance方式将数据发送至下游

@Override

public int selectChannel(SerializationDelegate<StreamRecord<T>> record) {

nextChannelToSendTo = (nextChannelToSendTo + 1) % numberOfChannels;

return nextChannelToSendTo;

}

通过这种方式来得出是发往哪个分区的,并不是随机而是一种轮询的方式发送。

API操作:

source.rebalance().print("rebalance").setParallelism(6);

  • RescalePartitioner

这个是个限制版的rebalence,和它很像,但是rescale是会对下游或上游进行分组,rebalance不分组就是直接轮询假如下游有4个分区也不管上游几个分区,向下游发送时就是固定的一个顺序1,2,3,4,1,2,3,4,1,2,3,4....但是Rescale就不会这么粗暴的轮询,而是上游和下游进行一个对应分组,假如上游有2个分区,下游有4个分区那么,上游的0分区就会在1,2之间轮询,1分区就会在3,4之间轮询。如果上游是4个下游是2个那么就是1,2向下游的0发送,3,4向下游的1发送。

@Override

public int selectChannel(SerializationDelegate<StreamRecord<T>> record) {

if (++nextChannelToSendTo >= numberOfChannels) {

nextChannelToSendTo = 0;

}

return nextChannelToSendTo;

}

API操作

.rescale().print("rescale").setParallelism(4);

  • GlobalPartitioner

简单粗暴,直接给到0分区,不管怎么来数据都是给到下游的ID=0分区。

@Override

public int selectChannel(SerializationDelegate<StreamRecord<T>> record) {

return 0;

}

API操作:

source.global().print("global").setParallelism(4);

  • KeyGroupStreamPartitioner

通过Key的Hash值判断发送给下游的哪个分区

@Override

public int selectChannel(SerializationDelegate<StreamRecord<T>> record) {

    K key;

    try {

        key = keySelector.getKey(record.getInstance().getValue());

    } catch (Exception e) {

        throw new RuntimeException("Could not extract key from " + record.getInstance().getValue(), e);

    }

    return KeyGroupRangeAssignment.assignKeyToParallelOperator(key, maxParallelism, numberOfChannels);

}

// KeyGroupRangeAssignment中的方法

public static int assignKeyToParallelOperator(Object key, int maxParallelism, int parallelism) {

    return computeOperatorIndexForKeyGroup(maxParallelism, parallelism, assignToKeyGroup(key, maxParallelism));

}

// KeyGroupRangeAssignment中的方法

public static int assignToKeyGroup(Object key, int maxParallelism) {

    return computeKeyGroupForKeyHash(key.hashCode(), maxParallelism);

}

// KeyGroupRangeAssignment中的方法

public static int computeKeyGroupForKeyHash(int keyHash, int maxParallelism) {

    return MathUtils.murmurHash(keyHash) % maxParallelism;

}

API操作

.KeyBy(new KeySelector<String, String>() {

            @Override

            public String getKey(String value) throws Exception {

                return value;

            }

        });

  • ForwardPartitioner

仅将元素转发到本地运行的下游操作的分区器,将记录输出到下游本地的operator实例。ForwardPartitioner分区器要求上下游算子并行度一样,上下游Operator同属一个SubTasks。

注意上下游并行度一定得是对等的,否则会运行就会报错: Forward partitioning does not allow change of parallelism. Upstream operation: Map-15 parallelism: 2, downstream operation: Sink: Print to Std. Out-17 parallelism: 4 You must use another partitioning strategy, such as broadcast, rebalance, shuffle or global.意思就是使用Forward分区就得一致,不然你用其他的分区策略。

@Override

public int selectChannel(SerializationDelegate<StreamRecord<T>> record) {

return 0;

}

API操作:

.forward().print(“forward”);

  • ShufflePartitioner

也不是轮询就是从下游通道中随机选一个分区。

@Override

public int selectChannel(SerializationDelegate<StreamRecord<T>> record) {

return random.nextInt(numberOfChannels);

}

API操作

source.shuffle().print("shuffle").setParallelism(4);

  • CustomPartitionerWrapper

用户自定义分区,用户指定这条数据要发送的分区编号。

public CustomPartitionerWrapper(Partitioner<K> partitioner, KeySelector<T, K> keySelector) {

this.partitioner = partitioner;

this.keySelector = keySelector;

}

API操作:

        source.partitionCustom(new Partitioner<String>() {

            @Override

            public int partition(String key, int numPartitions) {

                int mid = numPartitions / 2;

                return key.length() > 5 ? RandomUtils.nextInt(0, mid) : RandomUtils.nextInt(mid, numPartitions);

            }

        }, new KeySelector<String, String>() {

            @Override

            public String getKey(String value) throws Exception {

                return value;

            }

        }).map(Object::toString).print("custom").setParallelism(4);

  • BroadcastPartitioner

下游分区每个分区都分发。

@Override

public int selectChannel(SerializationDelegate<StreamRecord<T>> record) {

throw new UnsupportedOperationException("Broadcast partitioner does not support select channels.");

}

@Override

public boolean isBroadcast() {

return true;

}

API操作

source.broadcast().print("broadcast").setParallelism(4);

        1. 共享组

禁用算子链操作

算子链就是Operator chain,运行架构中介绍了操作链,可以将相同并行度的算子放到同一个Slot中执行,这样能避免数据在多个task之间流转来提高性能。

操作链可以通过env.disableChaining()全局禁用,也可以在某个算子操作后对后面的算子禁用。

Source -> map -> sink

  1. 并行度始终为1,默认开启效果

Flink-03 DataStream

  1. 并行度始终为1,全局禁用效果

Flink-03 DataStream

  1. 并行度始终为1,Source禁用算子链效果

Flink-03 DataStream

  1. Map并行度增大,任何阶段不禁用算子链效果

Flink-03 DataStream

由此可见并行度并不是随着上个算子传递下来的,而是没有设置就是默认,*别依次是 算子后设置并行度 >  env设置并行度 > 任务提交参数设置的并行度 > flink-conf.yaml配置文件中的并行度。

设置共享组操作

设置操作的槽位共享组。Flink 会将具有相同槽共享组的操作放在同一个槽中,而将没有槽共享组的操作保留在其他槽中。这可用于隔离插槽。如果所有输入操作都在同一个槽共享组中,则槽共享组从输入操作继承。默认插槽共享组的名称为“default”,可以通过调用 slotSharingGroup(“default”) 将操作显式放入该组。

上一篇:1.18.2.8与DataStream和DataSet API结合,Scala隐式转换,通过DataSet或DataStream创建视图,将DataStream或DataSet转换成表 等


下一篇:Flink 1.13,面向流批一体的运行时与 DataStream API 优化