Flink中的批和流
批处理的特点是有界、持久、大量,非常适合需要访问全部记录才能完成的计算工作,一般用于离线统计。
流处理的特点是无界、实时, 无需针对整个数据集执行操作,而是对通过系统 传输的每个数据项执行操作,一般用于实时统计。
一个无界流可以分解为多个有界流
性能 Flink > Spark > Hadoop
Flink的四种安装模式:
-
Local
-
Standalone
-
standaloneHA
-
Yarn
flink在使用input、output执行测试文件WordCount.jar 的时候,报出找不到文件的错误(但是文件路径存在),原因是:
因为我们的flink是task节点在执行任务的,task在三台机器上都有分布,我们当前文件只在一台服务器中,所以当其他task运行的时候,就会报出找不到文件的错误,将此文件分发到每台服务器中就不会出现这个错误。(我们以后在使用flink的时候,数据都是存放在hdfs上(一式三份),就不存在找不到文件的错误)
Flink-WordCount案例:
-
第一版代码
这一版代码比较简单,看代码就可以看懂
package com.bigdata;
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.DataStreamSource;
import org.apache.flink.streaming.api.datastream.SingleOutputStreamOperator;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
import org.apache.flink.util.Collector;
public class _01WorkCount {
public static void main(String[] args) throws Exception {
StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
DataStreamSource<String> dataStream01 = env.fromElements("spark flink kafka", "spark sqoop flink", "kafka hadoop flink");
// 首先先对字符串进行切割,形成一个新的数组
SingleOutputStreamOperator<String> flatMapStream = dataStream01.flatMap(new FlatMapFunction<String, String>() {
@Override
public void flatMap(String line, Collector<String> collector) throws Exception {
String[] arr = line.split(" ");
for (String word : arr) {
collector.collect(word);
}
}
});
// 将切割好的字符串形成 (word,1)的二元组的形式
SingleOutputStreamOperator<Tuple2<String, Integer>> map = flatMapStream.map(new MapFunction<String, Tuple2<String, Integer>>() {
@Override
public Tuple2<String, Integer> map(String word) throws Exception {
return Tuple2.of(word, 1);
}
});
// 聚合
DataStream<Tuple2<String, Integer>> sumResult = map.keyBy(new KeySelector<Tuple2<String, Integer>, String>() {
@Override
public String getKey(Tuple2<String, Integer> tuple2) throws Exception {
return tuple2.f0;
}
// 此处的1 指的是元组的第二个元素,进行相加的意思
}).sum(1);
sumResult.print();
env.execute();
}
}
第二版代码:简化了第一版的书写形式
第一版中 SingleOutputStreamOperator、DataStreamSource的父类其实都是DataStream,所以可以连着写下来
package com.bigdata;
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.DataStreamSource;
import org.apache.flink.streaming.api.datastream.SingleOutputStreamOperator;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
import org.apache.flink.util.Collector;
public class _02WorkCount {
/**
*
* 简化版案例
* @param args
* @throws Exception
*/
public static void main(String[] args) throws Exception {
StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
env.fromElements("spark flink kafka", "spark sqoop flink", "kafka hadoop flink")
.flatMap(new FlatMapFunction<String, String>() {
@Override
public void flatMap(String line, Collector<String> collector) throws Exception {
String[] arr = line.split(" ");
for (String word : arr) {
collector.collect(word);
}
}
}).map(new MapFunction<String, Tuple2<String, Integer>>() {
@Override
public Tuple2<String, Integer> map(String word) throws Exception {
return Tuple2.of(word, 1);
}
}).keyBy(new KeySelector<Tuple2<String, Integer>, String>() {
@Override
public String getKey(Tuple2<String, Integer> tuple2) throws Exception {
return tuple2.f0;
}
// 此处的1 指的是元组的第二个元素,进行相加的意思
}).sum(1).print();
env.execute();
}
}
第三版,使用lambda表达式,更加简化的书写
不过在使用lambda的时候,需要在后面指定一个方法的返回值,要不然会报错
package com.bigdata;
import org.apache.flink.api.common.functions.FlatMapFunction;
import org.apache.flink.api.common.functions.MapFunction;
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.DataStreamSource;
import org.apache.flink.streaming.api.datastream.SingleOutputStreamOperator;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
import org.apache.flink.util.Collector;
public class _03WorkCount_lambda {
/**
* lambda 表达式简化版
* @param args
* @throws Exception
*/
public static void main(String[] args) throws Exception {
// 使用lambda简化的时候,需要指定返回值类型,不指定的话会报错
StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
env.fromElements("spark flink kafka", "spark sqoop flink", "kafka hadoop flink")
.flatMap((String line, Collector<String> collector) -> {
String[] arr = line.split(" ");
for (String word : arr) {
collector.collect(word);
}
}).returns(Types.STRING).map((String word) -> Tuple2.of(word, 1)
).returns(Types.TUPLE(Types.STRING, Types.INT)).keyBy((Tuple2<String, Integer> tuple2) -> tuple2.f0).sum(1).print();
// 此处的1 指的是元组的第二个元素,进行相加的意思
env.execute();
}
}
复习lambda表达式:
-
lambda可以用来简化匿名内部类的书写
-
lambda只能简化函数式接口(有且仅有一个方法的接口)的匿名内部类的书写
省略规则:
-
只拿小括号里面的 加上 -> 指向大括号
-
只有一个参数的时候,参数类型可以省略
-
如果方法体中的代码只有一行,大括号和return等都可以省略(但是需要同时省略)
没省略之前的 (第一版)
省略后(第三版)
第四版,自定义输入与输出的路径地址
可以打包到集群中使用,使用的时候在jar包的后面跟上input路径以及output路径即可
package com.bigdata;
import org.apache.flink.api.common.RuntimeExecutionMode;
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.core.fs.FileSystem;
import org.apache.flink.streaming.api.datastream.DataStream;
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.util.Collector;
public class _04WorkCount_zidingyipass {
public static void main(String[] args) throws Exception {
StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
// 这个是 自动 ,根据流的性质,决定是批处理还是流处理
//env.setRuntimeMode(RuntimeExecutionMode.AUTOMATIC);
// 批处理流, 一口气把数据算出来
// env.setRuntimeMode(RuntimeExecutionMode.BATCH);
// 流处理,默认是这个 可以通过打印批和流的处理结果,体会流和批的含义
env.setRuntimeMode(RuntimeExecutionMode.STREAMING);
// 将任务的并行度设置为2
// env.setParallelism(2);
// 通过args传参
DataStreamSource<String> dataStream01 = null;
if (args.length == 0){
dataStream01 = env.fromElements("spark flink kafka", "spark sqoop flink", "kafka hadoop flink");
}else {
String input = args[0];
dataStream01 = env.readTextFile(input);
}
// 首先先对字符串进行切割,形成一个新的数组
SingleOutputStreamOperator<Tuple2<String, Integer>> sumResult = dataStream01
.flatMap((String line, Collector<String> collector) -> {
String[] arr = line.split(" ");
for (String word : arr) {
collector.collect(word);
}
}).map((String word) -> Tuple2.of(word, 1)
).keyBy((Tuple2<String, Integer> tuple2) -> tuple2.f0
// 此处的1 指的是元组的第二个元素,进行相加的意思
).sum(1);
if (args.length == 0){
sumResult.print();
}else {
String output = args[1];
sumResult.writeAsText(output, FileSystem.WriteMode.OVERWRITE).setParallelism(1);
}
env.execute();
}
}
打包后执行结果如下:
第五版,采用和flink中相同的书写方式 即带(--input 以及 --output)
也可以打包到集群中使用,使用的时候在jar包的后面跟上 --input +路径以及 -output + 路径即可
package com.bigdata;
import org.apache.flink.api.common.RuntimeExecutionMode;
import org.apache.flink.api.common.typeinfo.Types;
import org.apache.flink.api.java.tuple.Tuple2;
import org.apache.flink.api.java.utils.ParameterTool;
import org.apache.flink.core.fs.FileSystem;
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.util.Collector;
public class _05WorkCount_zidingyipass_input {
public static void main(String[] args) throws Exception {
StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
// 这个是 自动 ,根据流的性质,决定是批处理还是流处理
//env.setRuntimeMode(RuntimeExecutionMode.AUTOMATIC);
// 批处理流, 一口气把数据算出来
// env.setRuntimeMode(RuntimeExecutionMode.BATCH);
// 流处理,默认是这个 可以通过打印批和流的处理结果,体会流和批的含义
env.setRuntimeMode(RuntimeExecutionMode.STREAMING);
// 将任务的并行度设置为2
// env.setParallelism(2);
ParameterTool parameterTool = ParameterTool.fromArgs(args);
String input = "";
String output = "";
if (parameterTool.has("output") && parameterTool.has("input")) {
input = parameterTool.get("input");
output = parameterTool.get("output");
} else {
output = "hdfs://bigdata01:9820/home/wordcount/output";
}
// 通过args传参
DataStreamSource<String> dataStream01 = null;
if (args.length == 0){
dataStream01 = env.fromElements("spark flink kafka", "spark sqoop flink", "kafka hadoop flink");
}else {
dataStream01 = env.readTextFile(input);
}
// 首先先对字符串进行切割,形成一个新的数组
SingleOutputStreamOperator<Tuple2<String, Integer>> sumResult = dataStream01
.flatMap((String line, Collector<String> collector) -> {
String[] arr = line.split(" ");
for (String word : arr) {
collector.collect(word);
}
}).returns(Types.STRING).map((String word) -> Tuple2.of(word, 1)
).returns(Types.TUPLE(Types.STRING, Types.INT)).keyBy((Tuple2<String, Integer> tuple2) -> tuple2.f0
// 此处的1 指的是元组的第二个元素,进行相加的意思
).sum(1);
if (args.length == 0){
sumResult.print();
}else {
sumResult.writeAsText(output, FileSystem.WriteMode.OVERWRITE).setParallelism(1);
}
env.execute();
}
}