入门三之dataStream API、flink连kafka"/>
Flink入门三之dataStream API、flink连kafka
Flink流处理API
运行环境
Environment
getExecutionEnvironment
创建一个执行环境,表示当前执行程序的上下文。 如果程序是独立调用的,则此方法返回本地执行环境;如果从命令行客户端调用程序以提交到集群,则此方法返回此集群的执行环境,也就是说,getExecutionEnvironment 会根据查询运行的方式决定返回什么样的运行环境,是最常用的一种创建执行环境的方式。
ExecutionEnvironment env = ExecutionEnvironment.getExecutionEnvironment();
createLocalEnvironment
本地执行环境
LocalStreamEnvironment env = StreamExecutionEnvironment.createLocalEnvironment(1);
createRemoteEnvironment
远程执行环境
返回集群执行环境,将 Jar 提交到远程服务器。需要在调用时指定 JobManager的 IP 和端口号,并指定要在集群中运行的 Jar 包。
StreamExecutionEnvironment env =
StreamExecutionEnvironment.createRemoteEnvironment("jobmanage-hostname", 6123, "YOURPATH//WordCount.jar");
source
Kafka 读数据
package com.guigu.sc.source;import com.guigu.sc.beans.SensorReading;
import org.apache.flink.api.common.serialization.SimpleStringSchema;
import org.apache.flink.streaming.api.datastream.DataStream;
import org.apache.flink.streaming.api.datastream.DataStreamSource;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
import org.apache.flink.streaming.connectors.kafka.FlinkKafkaConsumer011;import java.util.Arrays;
import java.util.Properties;public class SourceTest3_Kafka {public static void main(String[] args) throws Exception{StreamExecutionEnvironment env =StreamExecutionEnvironment.getExecutionEnvironment();// kafka 配置项Properties properties = new Properties();properties.setProperty("bootstrap.servers", "localhost:9092");properties.setProperty("group.id", "consumer-group");properties.setProperty("key.deserializer","org.apache.kafka.common.serialization.StringDeserializer");properties.setProperty("value.deserializer","org.apache.kafka.common.serialization.StringDeserializer");properties.setProperty("auto.offset.reset", "latest");// 1.从kafaka 读数据DataStream<String> dataStream = env.addSource(new FlinkKafkaConsumer011<String>("sensor", new SimpleStringSchema(), properties));env.execute();}
}
自定义source
模拟数据源做测试时比较有用
除了以上的 source 数据来源,我们还可以自定义 source。需要做的,只是传入一个 SourceFunction 就可以。具体调用如下
package com.guigu.sc.source;import com.guigu.sc.beans.SensorReading;
import org.apache.flink.streaming.api.datastream.DataStream;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
import org.apache.flink.streaming.api.functions.source.SourceFunction;import java.util.HashMap;
import java.util.Random;public class SourceTest4_UDF {public static void main(String[] args) throws Exception{StreamExecutionEnvironment env =StreamExecutionEnvironment.getExecutionEnvironment();// 自定义数据源DataStream<SensorReading> dataStream = env.addSource(new MySensor());dataStream.print();env.execute();}public static class MySensor implements SourceFunction<SensorReading>{private boolean running = true;public void run(SourceContext<SensorReading> ctx) throws Exception {Random random = new Random();HashMap<String, Double> sensorTempMap = new HashMap<String, Double>();for( int i = 0; i < 10; i++ ){sensorTempMap.put("sensor_" + (i + 1), 60 + random.nextGaussian() * 20); // 高斯分布均值 该值 正负60之间}while (running) {for( String sensorId: sensorTempMap.keySet() ){Double newTemp = sensorTempMap.get(sensorId) + random.nextGaussian();sensorTempMap.put(sensorId, newTemp);ctx.collect( new SensorReading(sensorId, System.currentTimeMillis(), newTemp));}Thread.sleep(2000L);}public void cancel() {this.running = false;}}
}
Transform
转换算子(转换计算)
map,flatmap,Filter,KeyBy ——基本转换算子
map
flatmap
filter
package com.guigu.apiTest.transform;import com.guigu.sc.beans.SensorReading;
import com.guigu.sc.source.SourceTest4_UDF;
import org.apache.flink.api.common.functions.FilterFunction;
import org.apache.flink.api.common.functions.FlatMapFunction;
import org.apache.flink.api.common.functions.MapFunction;
import org.apache.flink.streaming.api.datastream.DataStream;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
import org.apache.flink.util.Collector;public class TransformTest1_Base {public static void main(String[] args) throws Exception{StreamExecutionEnvironment env =StreamExecutionEnvironment.getExecutionEnvironment();// 自定义数据源env.setParallelism(1);DataStream<SensorReading> dataStream = env.addSource(new SourceTest4_UDF.MySensor());// 1.map, sensorReading 转字符串 输出长度DataStream<Integer> mapStream = dataStream.map(//匿名类new MapFunction<SensorReading, Integer>() {@Overridepublic Integer map(SensorReading sensorReading) throws Exception {return sensorReading.toString().length();}});//2.flatmap, 按逗号切分字段DataStream<String> flatMapStream = dataStream.flatMap(new FlatMapFunction<SensorReading, String>() {@Overridepublic void flatMap(SensorReading value, Collector<String> out) throws Exception {String[] fields = value.toString().split(",");for (String field : fields) {out.collect(field);}}});// 3. filter, 条件筛选,筛选sensor_1开头的id对应的数据DataStream<SensorReading> filterStream = dataStream.filter(new FilterFunction<SensorReading>() {@Overridepublic boolean filter(SensorReading sensorReading) throws Exception {// 返回true标识筛选出来return sensorReading.toString().contains("sensor_1");}});// 打印数据mapStream.print("map");flatMapStream.print("flatmap");filterStream.print("filter");env.execute();}
}
KeyBy
DataStream → KeyedStream:逻辑地将一个流拆分成不相交的分区,每个分区包含具有相同 key 的元素,在内部以 hash 的形式实现的。
滚动聚合算子
所有聚合操作,只有分组后才能聚合。以下可以针对 KeyedStream 的每一个支流做聚合。
- sum() 求和
- min() 只取最小
- max() 只取最大
- minBy() 返回具有最小值的整个元素
- maxBy() 返回具有最大值的整个元素
package com.guigu.apiTest.transform;import com.guigu.sc.beans.SensorReading;
import com.guigu.sc.source.SourceTest4_UDF;
import org.apache.flink.api.common.functions.FilterFunction;
import org.apache.flink.api.common.functions.FlatMapFunction;
import org.apache.flink.api.common.functions.MapFunction;
import org.apache.flink.api.java.tuple.Tuple;
import org.apache.flink.streaming.api.datastream.DataStream;
import org.apache.flink.streaming.api.datastream.KeyedStream;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
import org.apache.flink.util.Collector;public class TransformTest2_RollingAggregation {public static void main(String[] args) throws Exception{StreamExecutionEnvironment env =StreamExecutionEnvironment.getExecutionEnvironment();// 设置并行度为1env.setParallelism(1);// 自定义数据源DataStream<SensorReading> dataStream = env.addSource(new SourceTest4_UDF.MySensor());// 分组KeyedStream<SensorReading, Tuple> keyedStream = dataStream.keyBy("id"); // javaBean 一个field// keySelector//KeyedStream<SensorReading, String> keyedStream2 = dataStream.keyBy(SensorReading::getId); // 传方法引用// 滚动聚合,不停更新DataStream<SensorReading> resultStream = keyedStream.maxBy("temperature"); // 传key,传位置的话就必须要是元组类型resultStream.print();env.execute();}
}
Reduce
KeyedStream → DataStream:一个分组数据流的聚合操作,合并当前的元素和上次聚合的结果,产生一个新的值,返回的流中包含每一次聚合的结果,而不是
只返回最后一次聚合的最终结果。(一般化的聚合)
多流转换算子
Split 和 Select
Split
DataStream → SplitStream:根据某些特征把一个 DataStream 拆分成两个或者多个 DataStream。
Select
需求:传感器数据按照温度高低(以 30 度为界),拆分成两个流。
过时方法,可以用底层process 替代
Connect 和 CoMap
DataStream,DataStream → ConnectedStreams:连接两个保持他们类型的数据流,两个数据流被 Connect 之后,只是被放在了一个同一个流中,内部依然保持各自的数据和形式不发生任何变化,两个流相互独立。
CoMap,CoFlatMape
ConnectedStreams → DataStream:作用于 ConnectedStreams 上,功能与 map和 flatMap 一样,对 ConnectedStreams 中的每一个 Stream 分别进行 map 和 flatMap处理。
应用:
connect()经常被应用于使用一个控制流对另一个数据流进行控制的场景,控制流可以是阈值、规则、机器学习模型或其他参数。
package com.guigu.apiTest.transform;import com.guigu.sc.beans.SensorReading;
import com.guigu.sc.source.SourceTest4_UDF;
import org.apache.flink.api.common.functions.MapFunction;
import org.apache.flink.api.java.tuple.Tuple2;
import org.apache.flink.api.java.tuple.Tuple3;
import org.apache.flink.streaming.api.collector.selector.OutputSelector;
import org.apache.flink.streaming.api.datastream.ConnectedStreams;
import org.apache.flink.streaming.api.datastream.DataStream;
import org.apache.flink.streaming.api.datastream.SplitStream;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
import org.apache.flink.streaming.api.functions.co.CoMapFunction;import java.util.Collections;public class TransformTest5_connect_comap {public static void main(String[] args) throws Exception{StreamExecutionEnvironment env =StreamExecutionEnvironment.getExecutionEnvironment();// 设置并行度为1env.setParallelism(1);// 自定义数据源DataStream<SensorReading> inputStream = env.addSource(new SourceTest4_UDF.MySensor());// split 分流,方法过时SplitStream<SensorReading> splitStream = inputStream.split(new OutputSelector<SensorReading>() {@Overridepublic Iterable<String> select(SensorReading sensorReading) {return (sensorReading.getTemperature() > 30) ? Collections.singletonList("high") :Collections.singletonList("low");}});// selectDataStream<SensorReading> highStream = splitStream.select("high");DataStream<SensorReading> lowStream = splitStream.select("low");// 2. 合流, 将 high 流转换为二元组类型 (id, temperature),与low流链接合并之后输出状态信息DataStream<Tuple2<String, Double>> warningStream = highStream.map(new MapFunction<SensorReading, Tuple2<String, Double>>() {@Overridepublic Tuple2<String, Double> map(SensorReading sensorReading) throws Exception {return new Tuple2<>(sensorReading.getId(), sensorReading.getTemperature());}});// 合流, high流输出报警,low输出健康状态ConnectedStreams<Tuple2<String, Double>, SensorReading> connectedStreams =warningStream.connect(lowStream);// 返回Object流,因为connect两个流类型可以不一样DataStream<Object> resultStream = connectedStreams.map(new CoMapFunction<Tuple2<String, Double>, SensorReading, Object>() {//处理第一个流@Overridepublic Object map1(Tuple2<String, Double> stringDoubleTuple2) throws Exception {return new Tuple3<>(stringDoubleTuple2.f0, stringDoubleTuple2.f1, "warning");}// 处理第二个流@Overridepublic Object map2(SensorReading sensorReading) throws Exception {return new Tuple2<>(sensorReading.getId(), "healthy");}});resultStream.print();env.execute();}
}
Flink不保证map1 和 map2 的执行顺序,两个方法的调用顺序依赖于两个数据流中数据流入的先后顺序。
Union
Union必须是相同数据类型才能进行合流。DataStream → DataStream:对两个或者两个以上的 DataStream 进行 union 操
作,产生一个包含所有 DataStream 元素的新 DataStream
// union联合多条流
DataStream<SensorReading> unionStream = highStream.union(lowStream);
connect Union区别
-
Union 之前两个流的类型必须是一样,Connect 可以不一样,在之后的 coMap中再去调整成为一样的。
-
Connect 只能操作两个流,Union 可以操作多个。
UDF——更细粒度的控制流
Flink 暴露了所有 udf 函数的接口(实现方式为接口或者抽象类)。例如MapFunction, FilterFunction, ProcessFunction 等等。
自定义算子接口
例如实现以下FilterFunction接口(过滤出含flink字段的字符串)
DataStream<String> flinkTweets = tweets.filter(new FlinkFilter()); //tweets 也是一个流public static class FlinkFilter implements FilterFunction<String> {@Overridepublic boolean filter(String value) throws Exception {return value.contains("flink");}
}//匿名类
DataStream<String> flinkTweets = tweets.filter(new FilterFunction<String>() {@Overridepublic boolean filter(String value) throws Exception {return value.contains("flink");}
});
我们 filter 的字符串"flink"还可以当作参数传进去。
DataStream<String> tweets = env.readTextFile("INPUT_FILE ");
DataStream<String> flinkTweets = tweets.filter(new KeyWordFilter("flink"));public static class KeyWordFilter implements FilterFunction<String> {private String keyWord;KeyWordFilter(String keyWord) { this.keyWord = keyWord; }@Overridepublic boolean filter(String value) throws Exception {return value.contains(this.keyWord);}
}
匿名函数形式
DataStream<String> tweets = env.readTextFile("INPUT_FILE");
DataStream<String> flinkTweets = tweets.filter( tweet -> tweet.contains("flink") );
富函数 Rich Functions
“富函数”是 DataStream API 提供的一个函数类的接口,所有 Flink 函数类都有其 Rich 版本。它与常规函数的不同在于,可以获取运行环境的上下文,并拥有一些生命周期方法,所以可以实现更复杂的功能。
- RichMapFunction
- RichFlatMapFunction.
- RichFilterFunction
Rich Function 有一个生命周期的概念。典型的生命周期方法有: (通过重写做一些操作)
-
open()方法是 rich function 的初始化方法,当一个算子例如 map 或者 filter
被调用之前 open()会被调用。
-
close()方法是生命周期中的最后一个调用的方法,做一些清理工作。
-
getRuntimeContext()方法提供了函数的 RuntimeContext 的一些信息,例如函
数执行的并行度,任务的名字,以及 state 状态
public static class MyMapFunction extends RichMapFunction<SensorReading,
Tuple2<Integer, String>> {@Overridepublic Tuple2<Integer, String> map(SensorReading value) throws Exception {return new Tuple2<>(getRuntimeContext().getIndexOfThisSubtask(), value.getId());}@Overridepublic void open(Configuration parameters) throws Exception {System.out.println("my map open");// 以下可以做一些初始化工作,例如建立一个和 HDFS 的连接}@Overridepublic void close() throws Exception {System.out.println("my map close");// 以下做一些清理工作,例如断开和 HDFS 的连接}
}
数据重区分操作
-
keyBy
-
forward 直传,只在当前分区计算
-
broadcast 广播,向下游子任务都广播一份
-
shuffe 洗牌,随机发牌给下游并行子任务上
-
rebalance 均匀分布下游的操作实例里 (round-robin)
-
rescale() 分组均衡,组内轮询
-
global() 所有的output丢给下游第一个实例(汇总数据, 慎用
-
partitionCustom 用户自定义重分区方式
DataStream 关系图
Sink
Flink 没有类似于 spark 中 foreach 方法,让用户进行迭代的操作。虽有对外的
输出操作都要利用 Sink 完成。最后通过类似如下方式完成整个任务最终输出操作。
stream.addSink(new MySink(xxxx))
kafka
输出到kafka
package com.guigu.wc;import com.guigu.wc.beans.SensorReading;
import org.apache.flink.api.common.functions.MapFunction;
import org.apache.flink.api.common.serialization.SimpleStringSchema;
import org.apache.flink.streaming.api.datastream.DataStream;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
import org.apache.flink.streaming.connectors.kafka.FlinkKafkaProducer011;public class SinkTest1_Kafka {public static void main(String[] args) throws Exception{StreamExecutionEnvironment env =StreamExecutionEnvironment.getExecutionEnvironment();// 自定义数据源env.setParallelism(4);DataStream<SensorReading> inputStream = env.addSource(new SourceTest4_UDF.MySensor());DataStream<String> mapStream = inputStream.map(new MapFunction<SensorReading, String>() {@Overridepublic String map(SensorReading sensorReading) throws Exception {return sensorReading.toString();}});// 与kafka连接mapStream.addSink(new FlinkKafkaProducer011<String>("localhost:9092","test", new SimpleStringSchema()));env.execute();}
}
kafka数据管道,kafka进 kafka出。
package com.guigu.wc;import com.guigu.wc.beans.SensorReading;
import org.apache.flink.api.common.functions.MapFunction;
import org.apache.flink.api.common.serialization.SimpleStringSchema;
import org.apache.flink.streaming.api.datastream.DataStream;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
import org.apache.flink.streaming.connectors.kafka.FlinkKafkaProducer011;public class SinkTest1_Kafka {public static void main(String[] args) throws Exception{StreamExecutionEnvironment env =StreamExecutionEnvironment.getExecutionEnvironment();// kafka 配置项Properties properties = new Properties();properties.setProperty("bootstrap.servers", "localhost:9092");properties.setProperty("group.id", "consumer-group");properties.setProperty("key.deserializer","org.apache.kafka.common.serialization.StringDeserializer");properties.setProperty("value.deserializer","org.apache.kafka.common.serialization.StringDeserializer");properties.setProperty("auto.offset.reset", "latest");// 1.从kafaka 读数据DataStream<String> dataStream = env.addSource(new FlinkKafkaConsumer011<String>("sensor", new SimpleStringSchema(), properties));// 与kafka连接dataStream.addSink(new FlinkKafkaProducer011<String>("localhost:9092","test", new SimpleStringSchema()));env.execute();}
}
更多推荐
Flink入门三之dataStream API、flink连kafka
发布评论