Flink基础
1、系统时间与事件时间
系统时间(处理时间)
在Sparksreaming的任务计算时,使用的是系统时间。
假设所用窗口为滚动窗口,大小为5分钟。那么每五分钟,都会对接收的数据进行提交任务.
但是,这里有个要注意的点,有个概念叫时间轴对齐 。若我们在12:12开始接收数据,按道理我们会在12:17进行提交任务。事实上我们会在12:20进行提交任务,因为会进行时间轴对齐,将一天按照五分钟进行划分,会对应到12:20。在此时提交任务,后面每个五分钟提交任务,都会对应到我们所划分的时间轴。
事件时间
flink支持带有事件时间的窗口(Window)操作
事件时间区别于系统时间,如下举例:
flink处理实时数据,对数据进行逐条处理。设定事件时间为5分钟,12:00开始接收数据,接收的第一条数据时间为12:01,接收的第二条数据为12:02。假设从此时起没有收到数据,那么将不会进行提交任务。**到了12:06,接收到了第三条数据。第三条数据的接收时间自12:00起,已经超过了五分钟,**那么此时便会进行任务提交。
2、wordcount简单案例的实现
java
import org.apache.flink.api.common.functions.FlatMapFunction;
import org.apache.flink.api.common.functions.MapFunction;
import org.apache.flink.api.java.functions.KeySelector;
import org.apache.flink.api.java.tuple.Tuple2;
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 Demo01StreamWordCount {
public static void main(String[] args) throws Exception {
// 1、构建Flink环境
StreamExecutionEnvironment env = StreamExecutionEnvironment
.getExecutionEnvironment();
// 2、通过Socket模拟无界流环境,方便FLink处理
// 虚拟机启动:nc -lk 8888
// 从Source构建第一个DataStream
// TODO C:\Windows\System32\drivers\etc\hosts文件中配置了master与IP地址的映射,所以这里可以使用master
DataStream<String> lineDS = env.socketTextStream("master", 8888);
// 统计每个单词的数量
// 第一步:将每行数据的每个单词切出来并进行扁平化处理
DataStream<String> wordsDS = lineDS.flatMap(new FlatMapFunction<String, String>() {
/**
*FlatMapFunction<String, String>: 表示输入、输出数据的类型
* @param line DS中的一条数据
* @param out 通过collect方法将数据发送到下游
* @throws Exception
*/
@Override
public void flatMap(String line, Collector<String> out) throws Exception {
for (String word : line.split(",")) {
// 将每个单词发送到下游
out.collect(word);
}
}
});
// 第二步:将每个单词变成 KV格式,V置为1;返回的数据是一个二元组Tuple2
DataStream<Tuple2<String, Integer>> wordKVDS = wordsDS.map(new MapFunction<String, Tuple2<String, Integer>>() {
@Override
public Tuple2<String, Integer> map(String word) throws Exception {
return Tuple2.of(word, 1);
}
});
/**
* 第三步:按每一个单词进行分组; 无法再使用其父类DataStream进行定义(无法向上转型)
* KeyedStream<T, K> 是 DataStream<T> 的一个特殊化版本,它添加了与键控操作相关的特定方法(如 reduce、aggregate、window 等)。
* 由于 KeyedStream 提供了额外的功能和方法,它不能简单地被视为 DataStream 的一个简单实例,
* 因为它实现了额外的接口(如 KeyedOperations<T, K>)并可能覆盖了某些方法的行为以支持键控操作。
*/
KeyedStream<Tuple2<String, Integer>, String> keyedDS = wordKVDS.keyBy(new KeySelector<Tuple2<String, Integer>, String>() {
@Override
public String getKey(Tuple2<String, Integer> tuple2) throws Exception {
// 对Key进行分组
return tuple2.f0;
}
});
// 第四步:对1进行聚合sum,下标是从0开始的
DataStream<Tuple2<String, Integer>> wordCntDS = keyedDS.sum(1);
// 3、打印结果:将DS中的内容Sink到控制台
wordCntDS.print();
// 执行任务
env.execute();
}
}
3、设置任务执行的并行度
本机为8核,可并行16的线程
手动改变任务的并行度,若不设置则会显示1-16,设置后只会显示1-2
env.setParallelism(2);
setBufferTimeout():设置输出缓冲区刷新的最大时间频率(毫秒)。
env.setBufferTimeout(200);
java
import org.apache.flink.api.common.functions.FlatMapFunction;
import org.apache.flink.api.common.functions.MapFunction;
import org.apache.flink.api.java.functions.KeySelector;
import org.apache.flink.api.java.tuple.Tuple2;
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 Demo01StreamWordCount {
public static void main(String[] args) throws Exception {
// 1、构建Flink环境
StreamExecutionEnvironment env = StreamExecutionEnvironment
.getExecutionEnvironment();
// 手动改变任务的并行度,默认并行度为最大,
env.setParallelism(2);
// setBufferTimeout():设置输出缓冲区刷新的最大时间频率(毫秒)。
env.setBufferTimeout(200);
// 2、通过Socket模拟无界流环境,方便FLink处理
// 虚拟机启动:nc -lk 8888
// 从Source构建第一个DataStream
DataStream<String> lineDS = env.socketTextStream("master", 8888);
System.out.println("lineDS并行度:" + lineDS.getParallelism());
// 统计每个单词的数量
// 第一步:将每行数据的每个单词切出来并进行扁平化处理
DataStream<String> wordsDS = lineDS.flatMap(new FlatMapFunction<String, String>() {
/**
*
* @param line DS中的一条数据
* @param out 通过collect方法将数据发送到下游
* @throws Exception
*/
@Override
public void flatMap(String line, Collector<String> out) throws Exception {
for (String word : line.split(",")) {
// 将每个单词发送到下游
out.collect(word);
}
}
});
System.out.println("wordsDS并行度:" + wordsDS.getParallelism());
// 第二步:将每个单词变成 KV格式,V置为1
DataStream<Tuple2<String, Integer>> wordKVDS = wordsDS.map(new MapFunction<String, Tuple2<String, Integer>>() {
@Override
public Tuple2<String, Integer> map(String word) throws Exception {
return Tuple2.of(word, 1);
}
});
System.out.println("wordKVDS并行度:" + wordKVDS.getParallelism());
// 第三步:按每一个单词进行分组
// keyBy之后数据流会进行分组,相同的key会进入同一个线程中被处理
// 传递数据的规则:hash取余(线程总数,默认CPU的总线程数)原理
KeyedStream<Tuple2<String, Integer>, String> keyedDS = wordKVDS.keyBy(new KeySelector<Tuple2<String, Integer>, String>() {
@Override
public String getKey(Tuple2<String, Integer> tuple2) throws Exception {
return tuple2.f0;
}
});
System.out.println("keyedDS并行度:" + keyedDS.getParallelism());
// 第四步:对1进行聚合sum
DataStream<Tuple2<String, Integer>> wordCntDS = keyedDS.sum(1);
System.out.println("wordCntDS并行度:" + wordCntDS.getParallelism());
// 3、打印结果:将DS中的内容Sink到控制台
keyedDS.print();
env.execute();
}
}
4、设置批/流处理方式,使用Lambda表达式,使用自定类实现接口中抽象的方法
java
package com.shujia.flink.core;
import org.apache.flink.api.common.RuntimeExecutionMode;
import org.apache.flink.api.common.functions.FlatMapFunction;
import org.apache.flink.api.common.typeinfo.Types;
import org.apache.flink.api.java.functions.KeySelector;
import org.apache.flink.api.java.tuple.Tuple2;
import org.apache.flink.streaming.api.datastream.DataStream;
import org.apache.flink.streaming.api.datastream.KeyedStream;
import org.apache.flink.streaming.api.datastream.SingleOutputStreamOperator;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
import org.apache.flink.util.Collector;
public class Demo02BatchWordCount {
public static void main(String[] args) throws Exception {
// 1、构建环境
StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
// 设置Flink程序的处理方式:默认是流处理
/**
* BATCH:批处理,只能处理有界流,底层是MR模型,可以进行预聚合
* STREAMING:流处理,可以处理无界流,也可以处理有界流,底层是持续流模型,数据一条一条处理
* AUTOMATIC:自动判断,当所有的Source都是有界流则使用BATCH模式,当Source中有一个是无界流则会使用STREAMING模式
*/
env.setRuntimeMode(RuntimeExecutionMode.BATCH);
// 2、获得第一个DS
// 通过readTextFile可以基于文件构建有界流
DataStream<String> wordsFileDS = env.readTextFile("flink/data/words.txt");
// 3、DS之间的转换
// 统计每个单词的数量
// 第一步:将每行数据的每个单词切出来并进行扁平化处理
// Flink处理逻辑传入的方式
// new XXXFunction 使用匿名内部类
// DataStream<String> wordsDS = wordsFileDS.flatMap(new FlatMapFunction<String, String>() {
// /**
// * @param line DS中的一条数据
// * @param out 通过collect方法将数据发送到下游
// * @throws Exception
// * Type parameters:
// * FlatMapFunction<T, O>
// * <T> -- Type of the input elements. <O> -- Type of the returned elements.
// */
// @Override
// public void flatMap(String line, Collector<String> out) throws Exception {
// for (String word : line.split(",")) {
// // 将每个单词发送到下游
// out.collect(word);
// }
// }
// });
/**
* 使用Lambda表达式
* 使用时得清楚FlatMapFunction中所要实现的抽象方法flatMap的两个参数的含义
* ()->{}
* 通过 -> 分隔,左边是函数的参数,右边是函数实现的具体逻辑
* 并且需要给出 flatMap函数的输出类型,Types.STRING
* line: 输入数据类型, out: 输出数据类型
*/
DataStream<String> wordsDS = wordsFileDS.flatMap((line, out) -> {
for (String word : line.split(",")) {
out.collect(word);
}
}, Types.STRING);
//TODO 使用自定类实现接口中抽象的方法,一般不使用这种方法
wordsFileDS.flatMap(new MyFunction()).print();
// 第二步:将每个单词变成 KV格式,V置为1
// DataStream<Tuple2<String, Integer>> wordKVDS = wordsDS.map(new MapFunction<String, Tuple2<String, Integer>>() {
// @Override
// public Tuple2<String, Integer> map(String word) throws Exception {
// return Tuple2.of(word, 1);
// }
// });
// TODO 此处需要给出 map函数的输出类型,Types.TUPLE(Types.STRING, Types.INT),是一个二元组
DataStream<Tuple2<String, Integer>> wordKVDS = wordsDS.map(word -> Tuple2.of(word, 1), Types.TUPLE(Types.STRING, Types.INT));
/**
* 第三步:按每一个单词进行分组
* keyBy之后数据流会进行分组,相同的key会进入同一个线程中被处理
* 传递数据的规则:hash取余(线程总数,默认CPU的总线程数,本机为16)原理
*/
// KeyedStream<Tuple2<String, Integer>, String> keyedDS = wordKVDS.keyBy(new KeySelector<Tuple2<String, Integer>, String>() {
// @Override
// public String getKey(Tuple2<String, Integer> tuple2) throws Exception {
// return tuple2.f0;
// }
// });
// TODO 此处的Types.STRING 并不是直接表示某个方法的输出类型,而是用来指定 keyBy 方法中键(key)的类型。这里可以省略!
KeyedStream<Tuple2<String, Integer>, String> keyedDS = wordKVDS.keyBy(kv -> kv.f0, Types.STRING);
// 第四步:对1进行聚合sum,无需指定返回值类型
DataStream<Tuple2<String, Integer>> wordCntDS = keyedDS.sum(1);
// 4、最终结果的处理(保存/输出打印)
wordCntDS.print();
env.execute();
}
}
class MyFunction implements FlatMapFunction<String,String>{
@Override
public void flatMap(String line, Collector<String> out) throws Exception {
for (String word : line.split(",")) {
// 将每个单词发送到下游
out.collect(word);
}
}
}
5、source
Flink 在流处理和批处理上的 source 大概有 4 类:
基于本地集合的 source、
基于文件的 source、
基于网络套接字的 source、
自定义的 source。自定义的 source 常见的有 Apache kafka、Amazon Kinesis Streams、RabbitMQ、Twitter Streaming API、Apache NiFi 等,当然你也可以定义自己的 source。
1、从本地集合source中读取数据
java
package com.shujia.flink.source;
import org.apache.flink.streaming.api.datastream.DataStream;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
import java.util.ArrayList;
public class Demo01ListSource {
public static void main(String[] args) throws Exception {
StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
// 本地集合Source
ArrayList<String> arrList = new ArrayList<>();
arrList.add("flink");
arrList.add("flink");
arrList.add("flink");
arrList.add("flink");
//TODO 有界流,fromCollection
DataStream<String> listDS = env.fromCollection(arrList);
listDS.print();
env.execute();
}
}
2、新版本从本地文件中读取数据,有界流和无界流两种方式
java
package com.shujia.flink.source;
import org.apache.flink.api.common.eventtime.WatermarkStrategy;
import org.apache.flink.connector.file.src.FileSource;
import org.apache.flink.connector.file.src.reader.TextLineInputFormat;
import org.apache.flink.core.fs.Path;
import org.apache.flink.streaming.api.datastream.DataStream;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
import java.io.File;
import java.time.Duration;
public class Demo02FileSource {
public static void main(String[] args) throws Exception {
StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
//TODO 历史版本读文件的方式,有界流
DataStream<String> oldFileDS = env.readTextFile("flink/data/words.txt");
// oldFileDS.print();
//TODO 读取案例一: 新版本加载文件的方式:FileSource,默认是有界流
FileSource<String> fileSource = FileSource
.forRecordStreamFormat(
new TextLineInputFormat()
, new Path("flink/data/words.txt")
)
.build();
//TODO 从Source加载数据构建DS,使用自带source类,使用 fromSource
DataStream<String> fileSourceDS = env.fromSource(fileSource, WatermarkStrategy.noWatermarks(), "fileSource");
fileSourceDS.print();
//TODO 读取案例二: 将读取文件变成无界流
FileSource<String> fileSource2 = FileSource
.forRecordStreamFormat(
new TextLineInputFormat()
, new Path("flink/data/words")
)
//TODO 使成为无界流读取一个文件夹中的数据,类似Flume中的spool dir,可以监控一个目录下文件的变化
// Duration.ofSeconds(5) 以5秒为间隔持续监控
.monitorContinuously(Duration.ofSeconds(5))
.build();
DataStream<String> fileSourceDS2 = env.fromSource(fileSource2,WatermarkStrategy.noWatermarks(),"fileSource2");
fileSourceDS2.print();
env.execute();
}
}
3、自定义source类,区分有界流与无界流
- 只有在Source启动时会执行一次
run方法如果会结束,则Source会得到一个有界流
run方法如果不会结束,则Source会得到一个无界流
java
package com.shujia.flink.source;
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;
public class Demo03MySource {
public static void main(String[] args) throws Exception {
StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
// TODO 使用自定义source类,通过addSource对其进行添加
DataStream<String> mySourceDS = env.addSource(new MySource());
mySourceDS.print();
env.execute();
}
}
class MySource implements SourceFunction<String>{
/**
* 只有在Source启动时会执行一次
* run方法如果会结束,则Source会得到一个有界流
* run方法如果不会结束,则Source会得到一个无界流
* 下面的例子Source会得到一个无界流
*/
@Override
public void run(SourceContext<String> ctx) throws Exception {
System.out.println("run方法启动了");
// ctx 可以通过collect方法向下游发送数据
long cnt = 0L;
while(true){
ctx.collect(cnt+"");
cnt ++;
// 休眠一会
Thread.sleep(1000);
}
}
// Source结束时会执行
@Override
public void cancel() {
System.out.println("Source结束了");
}
}
4、自定义source类,读取MySQL中的数据,并进行处理
java
package com.shujia.flink.source;
import org.apache.flink.api.common.typeinfo.Types;
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.source.SourceFunction;
import java.sql.Connection;
import java.sql.DriverManager;
import java.sql.ResultSet;
import java.sql.Statement;
public class Demo04MyMySQLSource {
public static void main(String[] args) throws Exception {
StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
DataStream<Students> studentDS = env.addSource(new MyMySQLSource());
// 统计班级人数
DataStream<Tuple2<String, Integer>> clazzCntDS = studentDS
.map(stu -> Tuple2.of(stu.clazz, 1), Types.TUPLE(Types.STRING, Types.INT))
.keyBy(t2 -> t2.f0)
.sum(1);
clazzCntDS.print();
// 统计性别人数
DataStream<Tuple2<String, Integer>> genderCntDS = studentDS
.map(stu -> Tuple2.of(stu.gender, 1), Types.TUPLE(Types.STRING, Types.INT))
.keyBy(t2 -> t2.f0)
.sum(1);
genderCntDS.print();
env.execute();
}
}
// TODO 自定义source类从MySQL中读取数据
class MyMySQLSource implements SourceFunction<Students> {
@Override
public void run(SourceContext<Students> ctx) throws Exception {
//TODO run方法只会执行一次创建下列对象的操作
// 建立连接
Connection conn = DriverManager.getConnection("jdbc:mysql://master:3306/bigdata_30", "root", "123456");
// 创建Statement
Statement st = conn.createStatement();
// 执行查询
ResultSet rs = st.executeQuery("select * from students2");
// 遍历rs提取每一条数据
while (rs.next()) {
long id = rs.getLong("id");
String name = rs.getString("name");
int age = rs.getInt("age");
String gender = rs.getString("gender");
String clazz = rs.getString("clazz");
Students stu = new Students(id, name, age, gender, clazz);
ctx.collect(stu);
/**
* 16> (文科四班,1)
* 15> (女,1)
* 15> (女,2)
* 2> (男,1)
* 7> (文科六班,1)
* 15> (女,3)
* 2> (男,2)
* 17> (理科六班,1)
* 17> (理科六班,2)
* 13> (理科五班,1)
* 20> (理科二班,1)
* 13> (理科四班,1)
*/
}
rs.close();
st.close();
conn.close();
}
@Override
public void cancel() {
}
}
// TODO 创建一个类,用于存储从MySQL中取出的数据
class Students {
Long id;
String name;
Integer age;
String gender;
String clazz;
public Students(Long id, String name, Integer age, String gender, String clazz) {
this.id = id;
this.name = name;
this.age = age;
this.gender = gender;
this.clazz = clazz;
}
}
6、sink
Flink 将转换计算后的数据发送的地点 。
Flink 常见的 Sink 大概有如下几类:写入文件、
打印出来、
写入 socket 、
自定义的 sink 。自定义的 sink 常见的有 Apache kafka、RabbitMQ、MySQL、ElasticSearch、Apache Cassandra、Hadoop FileSystem 等,同理你也可以定义自己的 sink。
1、构建FileSink,监控一个端口中的数据并将其写入到本地文件夹中
java
package com.shujia.flink.sink;
import org.apache.flink.api.common.serialization.SimpleStringEncoder;
import org.apache.flink.configuration.MemorySize;
import org.apache.flink.connector.file.sink.FileSink;
import org.apache.flink.core.fs.Path;
import org.apache.flink.streaming.api.datastream.DataStreamSource;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
import org.apache.flink.streaming.api.functions.sink.filesystem.rollingpolicies.DefaultRollingPolicy;
import java.time.Duration;
public class Demo01FileSink {
public static void main(String[] args) throws Exception {
StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
DataStreamSource<String> lineDS = env.socketTextStream("master", 8888);
// 构建FileSink
FileSink<String> fileSink = FileSink
.<String>forRowFormat(new Path("flink/data/fileSink"), new SimpleStringEncoder<String>("UTF-8"))
.withRollingPolicy(
DefaultRollingPolicy.builder()
// 这个设置定义了滚动的时间间隔。
.withRolloverInterval(Duration.ofSeconds(10))
// 这个设置定义了一个不活动间隔。
.withInactivityInterval(Duration.ofSeconds(10))
// 这个设置定义了单个日志文件可以增长到的最大大小。在这个例子中,每个日志文件在被滚动之前可以增长到最多1MB。
.withMaxPartSize(MemorySize.ofMebiBytes(1))
.build())
.build();
lineDS.sinkTo(fileSink);
env.execute();
}
}
2、自定义sink类
java
package com.shujia.flink.sink;
import org.apache.flink.streaming.api.datastream.DataStreamSource;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
import org.apache.flink.streaming.api.functions.sink.SinkFunction;
import java.util.ArrayList;
public class Demo02MySink {
public static void main(String[] args) throws Exception {
StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
ArrayList<String> arrList = new ArrayList<>();
arrList.add("flink");
arrList.add("flink");
arrList.add("flink");
arrList.add("flink");
DataStreamSource<String> ds = env.fromCollection(arrList);
ds.addSink(new MySinkFunction());
env.execute();
/**
* 进入了invoke方法
* flink
* 进入了invoke方法
* flink
* 进入了invoke方法
* flink
* 进入了invoke方法
* flink
*/
}
}
class MySinkFunction implements SinkFunction<String>{
@Override
public void invoke(String value, Context context) throws Exception {
System.out.println("进入了invoke方法");
// invoke 每一条数据会执行一次
// 最终数据需要sink到哪里,就对value进行处理即可
System.out.println(value);
}
}
7、Transformation:数据转换的常用操作
1、Map
java
package com.shujia.flink.tf;
import org.apache.flink.api.common.functions.MapFunction;
import org.apache.flink.api.common.typeinfo.Types;
import org.apache.flink.api.java.tuple.Tuple2;
import org.apache.flink.streaming.api.datastream.DataStream;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
public class Demo01Map {
public static void main(String[] args) throws Exception {
// 传入一条数据返回一条数据
StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
DataStream<String> ds = env.socketTextStream("master", 8888);
// 1、使用匿名内部类
DataStream<Tuple2<String, Integer>> mapDS = ds.map(new MapFunction<String, Tuple2<String, Integer>>() {
@Override
public Tuple2<String, Integer> map(String word) throws Exception {
return Tuple2.of(word, 1);
}
});
// mapDS.print();
// 2、使用lambda表达式
DataStream<Tuple2<String, Integer>> mapDS2 =
ds.map(word -> Tuple2.of(word, 1), Types.TUPLE(Types.STRING, Types.INT));
mapDS2.print();
env.execute();
}
}
2、FlatMap
java
package com.shujia.flink.tf;
import org.apache.flink.api.common.functions.FlatMapFunction;
import org.apache.flink.api.common.typeinfo.Types;
import org.apache.flink.api.java.tuple.Tuple;
import org.apache.flink.api.java.tuple.Tuple2;
import org.apache.flink.streaming.api.datastream.DataStream;
import org.apache.flink.streaming.api.datastream.SingleOutputStreamOperator;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
import org.apache.flink.util.Collector;
public class Demo02FlatMap {
public static void main(String[] args) throws Exception {
// 传入一条数据返回多条数据,类似UDTF函数
StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
DataStream<String> ds = env.socketTextStream("master", 8888);
// 1、使用匿名内部类
SingleOutputStreamOperator<Tuple2<String, Integer>> flatMapDS01 = ds.flatMap(new FlatMapFunction<String, Tuple2<String, Integer>>() {
@Override
public void flatMap(String line, Collector<Tuple2<String, Integer>> out) throws Exception {
for (String word : line.split(",")) {
out.collect(Tuple2.of(word, 1));
}
}
});
flatMapDS01.print();
// 2、使用lambda表达式
SingleOutputStreamOperator<Tuple> flatMapDS02 = ds.flatMap((line, out) -> {
for (String word : line.split(",")) {
out.collect(Tuple2.of(word, 1));
}
}, Types.TUPLE(Types.STRING, Types.INT));
flatMapDS02.print();
env.execute();
}
}
3、Filter
java
package com.shujia.flink.tf;
import org.apache.flink.api.common.functions.FilterFunction;
import org.apache.flink.streaming.api.datastream.DataStream;
import org.apache.flink.streaming.api.datastream.SingleOutputStreamOperator;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
public class Demo03Filter {
public static void main(String[] args) throws Exception {
// 过滤数据,注意返回值必须是布尔类型,返回true则保留数据,返回false则过滤数据
StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
DataStream<String> ds = env.socketTextStream("master", 8888);
/**
* Integer.valueOf:该方法将字符串参数转换为 Integer 对象。返回的是 Integer 类型,即 java.lang.Integer 的一个实例。
* Integer.parseInt:该方法将字符串参数解析为基本数据类型 int 的值。返回的是 int 类型的值,而不是对象。
* 无需指定返回值类型
*/
// 只输出大于10的数字
SingleOutputStreamOperator<String> filterDS = ds.filter(new FilterFunction<String>() {
@Override
public boolean filter(String value) throws Exception {
return Integer.parseInt(value) > 10;
}
});
filterDS.print();
ds.filter(value -> Integer.parseInt(value) > 10).print();
env.execute();
}
}
4、KeyBy
// 两种不同的简写方式
ds.keyBy(value -> value.toLowerCase(), Types.STRING).print();
ds.keyBy(String::toLowerCase, Types.STRING).print();
java
package com.shujia.flink.tf;
import org.apache.flink.api.common.typeinfo.Types;
import org.apache.flink.api.java.functions.KeySelector;
import org.apache.flink.streaming.api.datastream.DataStream;
import org.apache.flink.streaming.api.datastream.KeyedStream;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
public class Demo04KeyBy {
public static void main(String[] args) throws Exception {
// 用于就数据流分组,让相同的Key进入到同一个任务中进行处理,后续可以跟聚合操作
StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
DataStream<String> ds = env.socketTextStream("master", 8888);
KeyedStream<String, String> keyByDS = ds.keyBy(new KeySelector<String, String>() {
@Override
public String getKey(String value) throws Exception {
return value;
}
});
keyByDS.print();
// 两种不同的简写方式
ds.keyBy(value -> value.toLowerCase(), Types.STRING).print();
ds.keyBy(String::toLowerCase, Types.STRING).print();
env.execute();
}
}
5、Reduce
java
package com.shujia.flink.tf;
import org.apache.flink.api.common.functions.ReduceFunction;
import org.apache.flink.api.common.typeinfo.Types;
import org.apache.flink.api.java.tuple.Tuple2;
import org.apache.flink.api.java.tuple.Tuple3;
import org.apache.flink.streaming.api.datastream.DataStream;
import org.apache.flink.streaming.api.datastream.KeyedStream;
import org.apache.flink.streaming.api.datastream.SingleOutputStreamOperator;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
public class Demo05Reduce {
public static void main(String[] args) throws Exception {
// 用于对KeyBy之后的数据流进行聚合计算
StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
DataStream<String> ds = env.socketTextStream("master", 8888);
// 统计班级的平均年龄
/*
* 文科一班,20
* 文科一班,22
* 文科一班,21
* 文科一班,20
* 文科一班,22
*
* 理科一班,20
* 理科一班,21
* 理科一班,20
* 理科一班,21
* 理科一班,20
*
*/
SingleOutputStreamOperator<Tuple3<String, Integer, Integer>> kvDS = ds.map(line -> {
String[] split = line.split(",");
String clazz = split[0];
int age = Integer.parseInt(split[1]);
return Tuple3.of(clazz, age, 1);
}, Types.TUPLE(Types.STRING, Types.INT, Types.INT));
KeyedStream<Tuple3<String, Integer, Integer>, String> keyByDS = kvDS.keyBy(kv -> kv.f0, Types.STRING);
keyByDS.reduce(new ReduceFunction<Tuple3<String, Integer, Integer>>() {
@Override
public Tuple3<String, Integer, Integer> reduce(Tuple3<String, Integer, Integer> value1, Tuple3<String, Integer, Integer> value2) throws Exception {
return Tuple3.of(value1.f0, value1.f1 + value2.f1, value1.f2 + value2.f2);
}
}).map(t3 -> Tuple2.of(t3.f0, (double) t3.f1 / t3.f2),Types.TUPLE(Types.STRING,Types.DOUBLE))
.print();
keyByDS.reduce((v1,v2)->Tuple3.of(v1.f0, v1.f1 + v2.f1, v1.f2 + v2.f2))
.map(t3 -> Tuple2.of(t3.f0, (double) t3.f1 / t3.f2),Types.TUPLE(Types.STRING,Types.DOUBLE))
.print();
env.execute();
}
}
6、Window
java
package com.shujia.flink.tf;
import org.apache.flink.api.common.typeinfo.Types;
import org.apache.flink.api.java.tuple.Tuple2;
import org.apache.flink.streaming.api.datastream.DataStream;
import org.apache.flink.streaming.api.datastream.SingleOutputStreamOperator;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
import org.apache.flink.streaming.api.windowing.assigners.SlidingProcessingTimeWindows;
import org.apache.flink.streaming.api.windowing.assigners.TumblingProcessingTimeWindows;
import org.apache.flink.streaming.api.windowing.time.Time;
public class Demo06Window {
public static void main(String[] args) throws Exception {
// Flink窗口操作:时间、计数、会话
StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
DataStream<String> ds = env.socketTextStream("master", 8888);
SingleOutputStreamOperator<Tuple2<String, Integer>> kvDS = ds.map(word -> Tuple2.of(word, 1), Types.TUPLE(Types.STRING, Types.INT));
// 每隔5s统计每个单词的数量 ---> 滚动窗口实现(与spark中的定义相同)
SingleOutputStreamOperator<Tuple2<String, Integer>> outputDS01 = kvDS
// 按照Tuple2中的第一个元素进行分组
.keyBy(kv -> kv.f0, Types.STRING)
// 设置滚动时间
.window(TumblingProcessingTimeWindows.of(Time.seconds(5)))
// 对Tuple2中的第二个元素(索引为1的元素,即Integer类型)进行求和
.sum(1);
// outputDS01.print();
// 每隔5s统计最近10s内的每个单词的数量 ---> 滑动窗口实现(与spark中的定义相同)
kvDS
.keyBy(kv -> kv.f0, Types.STRING)
// 设置窗口大小和滑动大小
.window(SlidingProcessingTimeWindows.of(Time.seconds(10), Time.seconds(5)))
.sum(1)
.print();
env.execute();
}
}
7、Union
java
package com.shujia.flink.tf;
import org.apache.flink.streaming.api.datastream.DataStream;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
public class Demo07Union {
public static void main(String[] args) throws Exception {
StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
DataStream<String> ds01 = env.socketTextStream("master", 8888);
DataStream<String> ds02 = env.socketTextStream("master", 9999);
DataStream<String> unionDS = ds01.union(ds02);
// union 就是将两个相同结构的DS合并成一个DS(上下合并)
unionDS.print();
env.execute();
}
}
8、Process
通过processElement实现Map算子操作、flatMap算子操作(实现扁平化)、filter算子操作
java
package com.shujia.flink.tf;
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.ProcessFunction;
import org.apache.flink.util.Collector;
public class Demo08Process {
public static void main(String[] args) throws Exception {
StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
DataStream<String> ds01 = env.socketTextStream("master", 8888);
ds01.process(new ProcessFunction<String, Object>() {
/*
* 每进来一条数据就会执行一次
* value :一条数据
* ctx:可以获取任务执行时的信息
* out:用于输出数据
* ProcessFunction<String, Object>.Context ctx:flink的上下文对象
*/
@Override
public void processElement(String value, ProcessFunction<String, Object>.Context ctx, Collector<Object> out) throws Exception {
// 通过processElement实现Map算子操作
out.collect(Tuple2.of(value, 1));
// 通过processElement实现flatMap算子操作(实现扁平化)
for (String word : value.split(",")) {
out.collect(word);
}
// 通过processElement实现filter算子操作
if("java".equals(value)){
out.collect("java ok");
}
}
}).print();
env.execute();
}
}
通过processElement实现KeyBy算子操作
java
package com.shujia.flink.tf;
import org.apache.flink.api.common.typeinfo.Types;
import org.apache.flink.api.java.tuple.Tuple2;
import org.apache.flink.configuration.Configuration;
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.streaming.api.functions.KeyedProcessFunction;
import org.apache.flink.streaming.api.functions.ProcessFunction;
import org.apache.flink.util.Collector;
import java.util.HashMap;
public class Demo09KeyByProcess {
public static void main(String[] args) throws Exception {
StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
DataStream<String> ds01 = env.socketTextStream("master", 8888);
KeyedStream<Tuple2<String, Integer>, String> keyedDS = ds01.process(new ProcessFunction<String, Tuple2<String, Integer>>() {
@Override
public void processElement(String value, ProcessFunction<String, Tuple2<String, Integer>>.Context ctx, Collector<Tuple2<String, Integer>> out) throws Exception {
for (String word : value.split(",")) {
out.collect(Tuple2.of(word, 1));
}
}
}).keyBy(t2 -> t2.f0, Types.STRING);
// 基于分组之后的数据流同样可以调用process方法
/**
* KeyedProcessFunction<K, I, O>
* Type parameters:
* <K> -- Type of the key. <I> -- Type of the input elements. <O> -- Type of the output elements.
*/
keyedDS
.process(new KeyedProcessFunction<String, Tuple2<String, Integer>, String>() {
HashMap<String, Integer> wordCntMap;
// 当KeyedProcessFunction构建时只会执行一次,这样就避免了重复创建HashMap对象
@Override
public void open(Configuration parameters) throws Exception {
wordCntMap = new HashMap<String, Integer>();
}
// 每一条数据会执行一次
@Override
public void processElement(Tuple2<String, Integer> value, KeyedProcessFunction<String, Tuple2<String, Integer>, String>.Context ctx, Collector<String> out) throws Exception {
// 通过process实现word count
// 判断word是不是第一次进入,通过HashMap查找word是否有count值
String word = value.f0;
int cnt = 1;
if (wordCntMap.containsKey(word)) {
//get 在集合中通过value来获取对应的值
int newCnt = wordCntMap.get(word) + 1;
wordCntMap.put(word, newCnt);
cnt = newCnt;
} else {
wordCntMap.put(word, 1);
}
out.collect(word + ":" + cnt);
}
}).print();
env.execute();
}
}
8、Flink并行度
如何设置并行度?
1、考虑吞吐量
有聚合操作的任务:1w条/s 一个并行度
无聚合操作的任务:10w条/s 一个并行度
2、考虑集群本身的资源
注:
Task的数量由并行度以及有无Shuffle一起决定(可在shuffle之前观察是否有可合并的Task,可以来减少Task数量)
Task Slot数量 是由任务中最大的并行度决定
TaskManager的数量由配置文件中每个TaskManager设置的Slot数量及任务所需的Slot数量一起决定
FLink 并行度设置的几种方式:1、通过env设置,不推荐,如果需要调整并行度得修改代码重新打包提交任务
2、每个算子可以单独设置并行度,视实际情况决定,一般不常用
3、还可以在提交任务的时候指定并行度,最常用 比较推荐的方式
命令行:flink run 可以通过 -p 参数设置全局并行度
4、配置文件flink-conf.yaml中设置
web UI:填写parallelism输入框即可设置,优先级:算子本身的设置 > env做的全局设置 > 提交任务时指定的 > 配置文件flink-conf.yaml
java
package com.shujia.flink.core;
import org.apache.flink.api.common.typeinfo.Types;
import org.apache.flink.api.java.tuple.Tuple2;
import org.apache.flink.streaming.api.datastream.DataStreamSource;
import org.apache.flink.streaming.api.datastream.SingleOutputStreamOperator;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
public class Demo03Parallelism {
public static void main(String[] args) throws Exception {
/**
* 如何设置并行度?
* 1、考虑吞吐量
* 有聚合操作的任务:1w条/s 一个并行度
* 无聚合操作的任务:10w条/s 一个并行度
* 2、考虑集群本身的资源
*
* Task的数量由并行度以及有无Shuffle一起决定(可在shuffle之前观察是否有可合并的Task,可以来减少Task数量)
* Task Slot数量 是由任务中最大的并行度决定
* TaskManager的数量由配置文件中每个TaskManager设置的Slot数量及任务所需的Slot数量一起决定
*
*/
// FLink 并行度设置的几种方式
StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
// 1、通过env设置,不推荐,如果需要调整并行度得修改代码重新打包提交任务
env.setParallelism(3);
// socketTextStream的并行度为1,无法调整
DataStreamSource<String> ds = env.socketTextStream("master", 8888);
// 2、每个算子可以单独设置并行度,视实际情况决定,一般不常用
SingleOutputStreamOperator<Tuple2<String, Integer>> kvDS = ds
.map(word -> Tuple2.of(word, 1), Types.TUPLE(Types.STRING, Types.INT))
.setParallelism(4);
SingleOutputStreamOperator<Tuple2<String, Integer>> wordCntDS2P =
kvDS.keyBy(kv -> kv.f0)
.sum(1)
.setParallelism(2);
// 如果算子不设置并行度则以全局为准
wordCntDS2P.print();
/**
* 3、还可以在提交任务的时候指定并行度,最常用 比较推荐的方式
* 命令行:flink run 可以通过 -p 参数设置全局并行度
*
* web UI:填写parallelism输入框即可设置,优先级:算子本身的设置 > env做的全局设置 > 提交任务时指定的 > 配置文件flink-conf.yaml
*/
env.execute();
}
}
上述代码执行如下:
9、事件时间
事件时间 :指的是数据产生的时间或是数据发生的时间。它是数据本身所携带的时间信息,代表了事件真实发生的时间。在Flink中,事件时间通过数据元素自身带有的时间戳来表示,这个时间戳具有业务含义,并与系统时间独立。
1、案例一:基于事件事件的滚动窗口的实现
窗口的触发条件:
1、水位线大于等于窗口的结束时间
2、窗口内有数据
水位线:某个线程中所接收到的数据中最大的时间戳
水位线设置1: 单调递增时间戳策略,不考虑数据乱序问题。所传入数据的最大事件时间作为水位线
.<Tuple2<String, Long>>forMonotonousTimestamps()
水位线设置2 设置水位线前移,容忍5s的数据乱序到达,本质上将水位线前移5s,缺点:导致任务延时变大
.<Tuple2<String, Long>>forBoundedOutOfOrderness(Duration.ofSeconds(5))
java
package com.shujia.flink.core;
import org.apache.flink.api.common.eventtime.SerializableTimestampAssigner;
import org.apache.flink.api.common.eventtime.TimestampAssignerSupplier;
import org.apache.flink.api.common.eventtime.WatermarkStrategy;
import org.apache.flink.api.common.typeinfo.Types;
import org.apache.flink.api.java.tuple.Tuple2;
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 org.apache.flink.streaming.api.windowing.assigners.TumblingEventTimeWindows;
import org.apache.flink.streaming.api.windowing.time.Time;
import java.time.Duration;
public class Demo04EventTime {
public static void main(String[] args) throws Exception {
// 事件时间:数据本身自带的时间
StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
// 设置全局并行度
env.setParallelism(1);
/*
数据格式:单词,时间戳(很大的整数,Long类型)
a,1722233813000
a,1722233814000
a,1722233815000
a,1722233816000
a,1722233817000
a,1722233818000
a,1722233819000
a,1722233820000
a,1722233822000
a,1722233827000
*/
DataStreamSource<String> wordTsDS = env.socketTextStream("master", 8888);
SingleOutputStreamOperator<Tuple2<String, Long>> mapDS = wordTsDS
.map(line -> Tuple2.of(line.split(",")[0], Long.parseLong(line.split(",")[1])), Types.TUPLE(Types.STRING, Types.LONG));
// 指定数据的时间戳,告诉Flink,将其作为事件时间进行处理
SingleOutputStreamOperator<Tuple2<String, Long>> assDS = mapDS
.assignTimestampsAndWatermarks(
WatermarkStrategy
// 水位线:某个线程中所接收到的数据中最大的时间戳
// //水位线设置1: 单调递增时间戳策略,不考虑数据乱序问题。所传入数据的最大事件时间作为水位线
// .<Tuple2<String, Long>>forMonotonousTimestamps()
//TODO :水位线设置2 设置水位线前移,容忍5s的数据乱序到达,本质上将水位线前移5s,缺点:导致任务延时变大
.<Tuple2<String, Long>>forBoundedOutOfOrderness(Duration.ofSeconds(5))
// 指定事件时间,可以提取数据的某一部分作为事件时间
.withTimestampAssigner(new SerializableTimestampAssigner<Tuple2<String, Long>>() {
@Override
public long extractTimestamp(Tuple2<String, Long> t2, long recordTimestamp) {
return t2.f1;
}
})
);
// 不管是事件时间还是处理时间都需要搭配窗口操作一起使用
assDS.map(kv -> Tuple2.of(kv.f0, 1), Types.TUPLE(Types.STRING, Types.INT))
.keyBy(t2 -> t2.f0)
/**
* 窗口的触发条件
* 1、水位线大于等于窗口的结束时间
* 2、窗口内有数据
*TumblingEventTimeWindows:滚动窗口
*/
.window(TumblingEventTimeWindows.of(Time.seconds(5)))
.sum(1)
.print();
env.execute();
}
}
2、案例二:自定义水平线策略
多并行度,map之后指定水位线生成策略
注:必须两个线程中的水位线都超过了窗口的大小,才能触发窗口的执行
当窗口满足执行条件:
1、所有线程的水位线都超过了窗口的结束时间 (依次每两个不同编号的线程为一组,该组均超过)
2、窗口有数据 触发一次process方法
java
package tfTest;
import com.shujia.flink.event.MyEvent;
import org.apache.flink.api.common.eventtime.*;
import org.apache.flink.api.common.functions.MapFunction;
import org.apache.flink.api.common.typeinfo.Types;
import org.apache.flink.api.java.tuple.Tuple2;
import org.apache.flink.streaming.api.datastream.DataStreamSource;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
import org.apache.flink.streaming.api.functions.windowing.ProcessWindowFunction;
import org.apache.flink.streaming.api.windowing.assigners.TumblingEventTimeWindows;
import org.apache.flink.streaming.api.windowing.time.Time;
import org.apache.flink.streaming.api.windowing.windows.TimeWindow;
import org.apache.flink.util.Collector;
public class Demo05WaterMarkStrategy {
public static void main(String[] args) throws Exception {
// 自定义水位线策略
// 参考链接:https://blog.csdn.net/zznanyou/article/details/121666563
StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
env.setParallelism(2);
DataStreamSource<String> eventDS = env.socketTextStream("master", 8888);
// 将每条数据变成MyEvent类型
eventDS.map(new MapFunction<String, MyEvent>() {
@Override
public MyEvent map(String value) throws Exception {
String[] split = value.split(",");
return new MyEvent(split[0],Long.parseLong(split[1]));
}
})
// TODO 设置事件时间和自定义水平线策略
.assignTimestampsAndWatermarks(new WatermarkStrategy<MyEvent>() {
@Override
public TimestampAssigner<MyEvent> createTimestampAssigner(TimestampAssignerSupplier.Context context) {
return new SerializableTimestampAssigner<MyEvent>() {
@Override
public long extractTimestamp(MyEvent element, long recordTimestamp) {
return element.getTs();
}
};
}
@Override
public WatermarkGenerator<MyEvent> createWatermarkGenerator(WatermarkGeneratorSupplier.Context context) {
return new MyMapWatermarkGenerator();
}
}).keyBy(my-> my.getWord())
.window(TumblingEventTimeWindows.of(Time.seconds(5)))
// 当窗口满足执行条件:1、所有线程的水位线都超过了窗口的结束时间 2、窗口有数据 触发一次process方法
.process(new ProcessWindowFunction<MyEvent, String, String, TimeWindow>() {
@Override
public void process(String s, ProcessWindowFunction<MyEvent, String, String, TimeWindow>.Context context, Iterable<MyEvent> elements, Collector<String> out) throws Exception {
System.out.println("窗口触发执行了。");
System.out.println("当前水位线为:" + context.currentWatermark() + ",当前窗口的开始时间:" + context.window().getStart() + ",当前窗口的结束时间:" + context.window().getEnd());
// 基于elements做统计 通过out可以将结果发送到下游
}
}).print();
env.execute();
}
}
// 用于map之后指定水位线生成策略
class MyMapWatermarkGenerator implements WatermarkGenerator<MyEvent> {
private final long maxOutOfOrderness = 0;
private long currentMaxTimeStamp;
//TODO 每来一条数据会处理一次,若maxOutOfOrderness为0,则为单调递增时间戳策略;若不为0,则是水位线前移策略
@Override
public void onEvent(MyEvent event, long eventTimestamp, WatermarkOutput output) {
currentMaxTimeStamp = Math.max(currentMaxTimeStamp, eventTimestamp);
System.out.println("当前线程编号为:" + Thread.currentThread().getId() + ",当前水位线为:" + (currentMaxTimeStamp - maxOutOfOrderness));
}
// 周期性的执行:env.getConfig().getAutoWatermarkInterval(); 默认是200ms
@Override
public void onPeriodicEmit(WatermarkOutput output) {
// 发送
output.emitWatermark(new Watermark(currentMaxTimeStamp - maxOutOfOrderness));
}
}
执行结果:
多并行度,source之后设置水位线策略
效果通线程并行度为1的情况
java
package com.shujia.flink.core;
import com.shujia.flink.event.MyEvent;
import org.apache.flink.api.common.eventtime.*;
import org.apache.flink.api.common.functions.MapFunction;
import org.apache.flink.api.common.typeinfo.Types;
import org.apache.flink.api.java.tuple.Tuple2;
import org.apache.flink.streaming.api.datastream.DataStreamSource;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
import org.apache.flink.streaming.api.functions.windowing.ProcessWindowFunction;
import org.apache.flink.streaming.api.windowing.assigners.TumblingEventTimeWindows;
import org.apache.flink.streaming.api.windowing.time.Time;
import org.apache.flink.streaming.api.windowing.windows.TimeWindow;
import org.apache.flink.util.Collector;
public class Demo05WaterMarkStrategy {
public static void main(String[] args) throws Exception {
// 自定义水位线策略
// 参考链接:https://blog.csdn.net/zznanyou/article/details/121666563
StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
env.setParallelism(2);
DataStreamSource<String> eventDS = env.socketTextStream("master", 8888);
// 在Source之后就指定水位线策略
eventDS.assignTimestampsAndWatermarks(new WatermarkStrategy<String>() {
// 指定时间戳的提取策略
@Override
public TimestampAssigner<String> createTimestampAssigner(TimestampAssignerSupplier.Context context) {
return new SerializableTimestampAssigner<String>() {
@Override
public long extractTimestamp(String element, long recordTimestamp) {
return Long.parseLong(element.split(",")[1]);
}
};
// 简写方式
// return (ele,ts)->Long.parseLong(ele.split(",")[1]);
}
// 指定水位线的策略
@Override
public WatermarkGenerator<String> createWatermarkGenerator(WatermarkGeneratorSupplier.Context context) {
return new MyWatermarkGenerator();
}
})
// 将数据变成KV格式,即:单词,1
.map(line -> Tuple2.of(line.split(",")[0], 1), Types.TUPLE(Types.STRING, Types.INT))
.keyBy(t2 -> t2.f0)
.window(TumblingEventTimeWindows.of(Time.seconds(5)))
// 当窗口满足执行条件:1、水位线超过了窗口的结束时间 2、窗口有数据 触发一次process方法
.process(new ProcessWindowFunction<Tuple2<String, Integer>, Tuple2<String, Integer>, String, TimeWindow>() {
@Override
public void process(String s, ProcessWindowFunction<Tuple2<String, Integer>, Tuple2<String, Integer>, String, TimeWindow>.Context context, Iterable<Tuple2<String, Integer>> elements, Collector<Tuple2<String, Integer>> out) throws Exception {
System.out.println("窗口触发执行了。");
System.out.println("当前水位线为:" + context.currentWatermark() + ",当前窗口的开始时间:" + context.window().getStart() + ",当前窗口的结束时间:" + context.window().getEnd());
// 基于elements做统计 通过out可以将结果发送到下游
}
}).print();
env.execute();
}
}
// 用于Source之后直接指定水位线生成策略
class MyWatermarkGenerator implements WatermarkGenerator<String> {
private final long maxOutOfOrderness = 0;
private long currentMaxTimeStamp;
// 每来一条数据会处理一次
@Override
public void onEvent(String event, long eventTimestamp, WatermarkOutput output) {
currentMaxTimeStamp = Math.max(currentMaxTimeStamp, eventTimestamp);
System.out.println("当前线程编号为:" + Thread.currentThread().getId() + ",当前水位线为:" + (currentMaxTimeStamp - maxOutOfOrderness));
}
// 周期性的执行:env.getConfig().getAutoWatermarkInterval(); 默认是200ms
@Override
public void onPeriodicEmit(WatermarkOutput output) {
output.emitWatermark(new Watermark(currentMaxTimeStamp - maxOutOfOrderness));
}
}
10、窗口
1、时间窗口:滚动与滑动窗口
时间窗口:滚动、滑动
时间类型:处理时间、事件时间
java
package com.shujia.flink.window;
import com.shujia.flink.event.MyEvent;
import org.apache.flink.api.common.eventtime.WatermarkStrategy;
import org.apache.flink.api.common.functions.MapFunction;
import org.apache.flink.api.common.typeinfo.Types;
import org.apache.flink.api.java.tuple.Tuple2;
import org.apache.flink.streaming.api.datastream.DataStream;
import org.apache.flink.streaming.api.datastream.SingleOutputStreamOperator;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
import org.apache.flink.streaming.api.windowing.assigners.SlidingEventTimeWindows;
import org.apache.flink.streaming.api.windowing.assigners.SlidingProcessingTimeWindows;
import org.apache.flink.streaming.api.windowing.time.Time;
import java.time.Duration;
public class Demo01TimeWindow {
public static void main(String[] args) throws Exception {
/*
* 时间窗口:滚动、滑动
* 时间类型:处理时间、事件时间
*/
StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
env.setParallelism(1);
DataStream<MyEvent> myDS = env.socketTextStream("master", 8888)
.map(new MapFunction<String, MyEvent>() {
@Override
public MyEvent map(String value) throws Exception {
String[] split = value.split(",");
return new MyEvent(split[0], Long.parseLong(split[1]));
}
});
// 基于处理时间的滚动、滑动窗口
SingleOutputStreamOperator<Tuple2<String, Integer>> processDS = myDS
.map(e -> Tuple2.of(e.getWord(), 1), Types.TUPLE(Types.STRING, Types.INT))
.keyBy(t2 -> t2.f0)
// 滚动窗口 每隔5s统计一次
// .window(TumblingProcessingTimeWindows.of(Time.seconds(5)))
// 滑动窗口 每隔5s统计最近10s内的数据
.window(SlidingProcessingTimeWindows.of(Time.seconds(10), Time.seconds(5)))
.sum(1);
// 基于事件时间的滚动、滑动窗口
SingleOutputStreamOperator<MyEvent> assDS = myDS.assignTimestampsAndWatermarks(
// 设置水位线策略、指定事件时间
WatermarkStrategy
// Duration.ofSeconds(5):水位线前移5s
.<MyEvent>forBoundedOutOfOrderness(Duration.ofSeconds(5))
.withTimestampAssigner((event, ts) -> event.getTs())
);
SingleOutputStreamOperator<Tuple2<String, Integer>> eventDS = assDS
.map(e -> Tuple2.of(e.getWord(), 1), Types.TUPLE(Types.STRING, Types.INT))
.keyBy(t2 -> t2.f0)
// 滚动窗口,由于水位线前移了5s,整体有5s的延时
// .window(TumblingEventTimeWindows.of(Time.seconds(5)))
// 滑动窗口
.window(SlidingEventTimeWindows.of(Time.seconds(10),Time.seconds(5)))
.sum(1);
// processDS.print();
eventDS.print();
env.execute();
}
}
2、会话窗口
基于处理时间的会话窗口,当一段时间没有数据,那么就认定此次会话结束并触发窗口的执行
基于事件时间的会话窗口,连续接收的两条数据的事件时间之差要大于5s(窗口大小),才能触发窗口的执行
java
package com.shujia.flink.window;
import com.shujia.flink.event.MyEvent;
import org.apache.flink.api.common.eventtime.WatermarkStrategy;
import org.apache.flink.api.common.functions.MapFunction;
import org.apache.flink.api.common.typeinfo.Types;
import org.apache.flink.api.java.tuple.Tuple2;
import org.apache.flink.streaming.api.datastream.DataStream;
import org.apache.flink.streaming.api.datastream.SingleOutputStreamOperator;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
import org.apache.flink.streaming.api.windowing.assigners.EventTimeSessionWindows;
import org.apache.flink.streaming.api.windowing.assigners.ProcessingTimeSessionWindows;
import org.apache.flink.streaming.api.windowing.assigners.SlidingProcessingTimeWindows;
import org.apache.flink.streaming.api.windowing.time.Time;
public class Demo02Session {
public static void main(String[] args) throws Exception {
StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
env.setParallelism(1);
DataStream<MyEvent> myDS = env.socketTextStream("master", 8888)
.map(new MapFunction<String, MyEvent>() {
@Override
public MyEvent map(String value) throws Exception {
String[] split = value.split(",");
return new MyEvent(split[0], Long.parseLong(split[1]));
}
});
// 基于处理时间的会话窗口,当一段时间没有数据,那么就认定此次会话结束并触发窗口的执行
SingleOutputStreamOperator<Tuple2<String, Integer>> processSessionDS = myDS.map(e -> Tuple2.of(e.getWord(), 1), Types.TUPLE(Types.STRING, Types.INT))
.keyBy(t2 -> t2.f0)
// 10秒内没有数据,则认定此次会话结束并触发窗口的执行
.window(ProcessingTimeSessionWindows.withGap(Time.seconds(10)))
.sum(1);
//TODO 基于事件时间的会话窗口,连续接收的两条数据的事件时间之差要大于5s(窗口大小),才能触发窗口的执行
// 指定水位线策略并提供数据中的时间戳解析规则
SingleOutputStreamOperator<MyEvent> assDS = myDS.assignTimestampsAndWatermarks(
WatermarkStrategy
.<MyEvent>forMonotonousTimestamps()
.withTimestampAssigner((e, ts) -> e.getTs())
);
SingleOutputStreamOperator<Tuple2<String, Integer>> eventSessionDS = assDS.map(e -> Tuple2.of(e.getWord(), 1), Types.TUPLE(Types.STRING, Types.INT))
.keyBy(t2 -> t2.f0)
.window(EventTimeSessionWindows.withGap(Time.seconds(5)))
.sum(1);
// processSessionDS.print();
eventSessionDS.print();
env.execute();
}
}
3、计数窗口:滚动、滑动
滚动下:每同一个key的5条数据会统计一次
滑动下:每隔同一个key的5条数据,统计最近的同一个key的10条数据
java
package com.shujia.flink.window;
import org.apache.flink.api.common.typeinfo.Types;
import org.apache.flink.api.java.tuple.Tuple2;
import org.apache.flink.streaming.api.datastream.DataStreamSource;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
public class Demo03CountWindow {
public static void main(String[] args) throws Exception {
// 计数窗口:滚动、滑动
StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
DataStreamSource<String> ds = env.socketTextStream("master", 8888);
ds.map(word-> Tuple2.of(word,1), Types.TUPLE(Types.STRING,Types.INT))
.keyBy(t2->t2.f0)
// .countWindow(5) // 每同一个key的5条数据会统计一次
.countWindow(10,5) // 每隔同一个key的5条数据,统计最近的同一个key的10条数据
.sum(1)
.print();
env.execute();
/**
* 每隔同一个key的5条数据,统计最近的同一个key的10条数据
* 输入:
* a
* a
* a
* a
* a
* b
* b
* b
* a
* a
* a
* a
* a
* 输出:
* 13> (a,5)
* 13> (a,10)
*/
}
}