Spark应用的数据源：
1）Driver驱动中的一个集合(parallelizePairs parallelize)
2）从本地（file:///d:/test）或者网络(file:///hdfs:localhost:7777)存上获取
textFile textWholeFiles
3）流式数据源：Socket (socketTextStream)
一、Spark封装的格式：
1、普通文件
2、JSON
3、CSV
如果CSV的所有数据字段均没有包含换行符，可以使用 textFile() 读取并解析数据，如果在字段中嵌有换行符，就需要用wholeTextFiles()完整读入每个文件，然后解析各段.
由于在 CSV 中我们不会在每条记录中输出字段名，因此为了使输出保持一致，需要 创建一种映射关系。一种简单做法是写一个函数，用于将各字段转为指定顺序的数组。
4、sequence file 二进制形式 键值对
5、object file JDK 序列化（看起来是对sequenceFile进行了简单封装，他允许存储只包含值的RDD，和sequenceFile不一样的是，对象文件是java序列化写出的，读取的对象不能改变（输出会依赖对象））
普通文件file
import java.io.Serializable;
import java.io.StringReader;
import java.util.ArrayList;
import java.util.Iterator;
import org.apache.spark.SparkConf;
import org.apache.spark.api.java.JavaPairRDD;
import org.apache.spark.api.java.JavaRDD;
import org.apache.spark.api.java.JavaSparkContext;
import org.apache.spark.api.java.function.FlatMapFunction;
import org.apache.spark.api.java.function.Function;
import scala.Tuple2;
import au.com.bytecode.opencsv.CSVReader;
import com.fasterxml.jackson.databind.ObjectMapper;
public class SparkIO_File {
public static void main(String[] args) {
SparkConf conf = new SparkConf().setMaster("local[2]").setAppName("SparkIO").set("spark.testing.memory", "2147480000");
JavaSparkContext sc = new JavaSparkContext(conf);
sc.setLogLevel("WARN");
fileTest(sc);
sc.stop();
sc.close();
}
static void fileTest(JavaSparkContext sc){
//每行都是rdd
// JavaRDD<String> rdd = sc.textFile("file:///E:/codes2016/workspace/Spark1/src/spark1106_StreamSpark/UpdateStateByKeyDemo.java");
//wholeTextFiles返回一个键值对类型，键为文件全路径，值为文件内容，分区数是2
JavaPairRDD<String, String> rdd = sc.wholeTextFiles("file:///E:/codes2016/workspace/Spark1/src/spark1106_StreamSpark");
System.out.println("分区数:"+rdd.getNumPartitions()); //分区数为2
rdd.foreach(x->{
System.out.println("当前元素：" + x);
});
System.out.println(rdd.count());
rdd.saveAsTextFile("file:///d:/jsontext/filewholetext");
}
}
json文件
import java.io.Serializable;
import java.io.StringReader;
import java.util.ArrayList;
import java.util.Iterator;
import org.apache.spark.SparkConf;
import org.apache.spark.api.java.JavaPairRDD;
import org.apache.spark.api.java.JavaRDD;
import org.apache.spark.api.java.JavaSparkContext;
import org.apache.spark.api.java.function.FlatMapFunction;
import org.apache.spark.api.java.function.Function;
import scala.Tuple2;
import au.com.bytecode.opencsv.CSVReader;
import com.fasterxml.jackson.databind.ObjectMapper;
public class SparkIO_JSON {
public static void main(String[] args) {
SparkConf conf = new SparkConf().setMaster("local[2]").setAppName("SparkIO").set("spark.testing.memory", "2147480000");
JavaSparkContext sc = new JavaSparkContext(conf);
sc.setLogLevel("WARN");
writeJsonTest(sc);
sc.stop();
sc.close();
}
//读JSON
static void readJsonTest(JavaSparkContext sc){
//如果json文件中断了行就读不出来了，没截断的部分任然会显示
JavaRDD<String> input = sc.textFile("file:///d:/jsontext/jsonsong.json");
//使用wholetextfile就不会有断行的错误，因为读的是整个文件
// JavaRDD<String> input = sc.wholeTextFiles("file:///d:/jsontext/jsonsong.json");
// JavaRDD<Mp3Info> result = input.mapPartitions(new ParseJson());
JavaRDD<Mp3Info> result = input.map(x->{
ObjectMapper mapper=new ObjectMapper();
return mapper.readValue(x, Mp3Info.class);
});
result.foreach(x->System.out.println(x));
}
//写JSON
static void writeJsonTest(JavaSparkContext sc){
JavaRDD<String> input = sc.textFile("file:///d:/jsontext/jsonsong.json");
JavaRDD<Mp3Info> result = input.mapPartitions(new ParseJson()).
filter(
x->x.getAlbum().equals("怀旧专辑")
);
// JavaRDD<String> formatted = result.mapPartitions(new WriteJson());
JavaRDD<String> formatted = result.map(x->{
ObjectMapper mapper=new ObjectMapper();
return mapper.writeValueAsString(x);
});
result.foreach(x->System.out.println(x));
formatted.saveAsTextFile("file:///d:/jsontext/jsonsongout");
}
}
class ParseJson implements FlatMapFunction<Iterator<String>, Mp3Info>, Serializable {
public Iterator<Mp3Info> call(Iterator<String> lines) throws Exception {
ArrayList<Mp3Info> people = new ArrayList<Mp3Info>();
ObjectMapper mapper = new ObjectMapper();
while (lines.hasNext()) {
String line = lines.next();
try {
people.add(mapper.readValue(line, Mp3Info.class));
} catch (Exception e) {
e.printStackTrace();
}
}
return people.iterator();
}
}
class WriteJson implements FlatMapFunction<Iterator<Mp3Info>, String> {
public Iterator<String> call(Iterator<Mp3Info> song) throws Exception {
ArrayList<String> text = new ArrayList<String>();
ObjectMapper mapper = new ObjectMapper();
while (song.hasNext()) {
Mp3Info person = song.next();
text.add(mapper.writeValueAsString(person));
}
return text.iterator();
}
}
class Mp3Info implements Serializable{
/*
{"name":"上海滩","singer":"叶丽仪","album":"香港电视剧主题歌","path":"mp3/shanghaitan.mp3"}
{"name":"一生何求","singer":"陈百强","album":"香港电视剧主题歌","path":"mp3/shanghaitan.mp3"}
{"name":"红日","singer":"李克勤","album":"怀旧专辑","path":"mp3/shanghaitan.mp3"}
{"name":"爱如潮水","singer":"张信哲","album":"怀旧专辑","path":"mp3/airucaoshun.mp3"}
{"name":"红茶馆","singer":"陈惠嫻","album":"怀旧专辑","path":"mp3/redteabar.mp3"}
*/
private String name;
private String album;
private String path;
private String singer;
public String getSinger() {
return singer;
}
public void setSinger(String singer) {
this.singer = singer;
}
public String getName() {
return name;
}
public void setName(String name) {
this.name = name;
}
public String getAlbum() {
return album;
}
public void setAlbum(String album) {
this.album = album;
}
public String getPath() {
return path;
}
public void setPath(String path) {
this.path = path;
}
@Override
public String toString() {
return "Mp3Info [name=" + name + ", album="
+ album + ", path=" + path + ", singer=" + singer + "]";
}
}
/*
{"name":"上海滩","singer":"叶丽仪","album":"香港电视剧主题歌","path":"mp3/shanghaitan.mp3"}
{"name":"一生何求","singer":"陈百强","album":"香港电视剧主题歌","path":"mp3/shanghaitan.mp3"}
{"name":"红日","singer":"李克勤","album":"怀旧专辑","path":"mp3/shanghaitan.mp3"}
{"name":"爱如潮水","singer":"张信哲","album":"怀旧专辑","path":"mp3/airucaoshun.mp3"}
{"name":"红茶馆","singer":"陈惠嫻","album":"怀旧专辑","path":"mp3/redteabar.mp3"}
*/csv文件
import java.io.StringReader;
import java.io.StringWriter;
import java.util.Arrays;
import java.util.Iterator;
import org.apache.spark.SparkConf;
import org.apache.spark.api.java.JavaPairRDD;
import org.apache.spark.api.java.JavaRDD;
import org.apache.spark.api.java.JavaSparkContext;
import org.apache.spark.api.java.function.FlatMapFunction;
import org.apache.spark.api.java.function.Function;
import scala.Tuple2;
import au.com.bytecode.opencsv.CSVReader;
import au.com.bytecode.opencsv.CSVWriter;
public class SparkIO_CSV {
public static void main(String[] args) {
SparkConf conf = new SparkConf().setMaster("local[2]").setAppName("SparkIO").set("spark.testing.memory", "2147480000");
JavaSparkContext sc = new JavaSparkContext(conf);
sc.setLogLevel("WARN");
readCsv2(sc);
sc.stop();
sc.close();
}
static void readCsv1(JavaSparkContext sc) {
JavaRDD<String> csvFile1 = sc.textFile("file:///d:/jsontext/csvsong.csv");
// csvFile1.foreach(x->System.out.println(x));
JavaRDD<String[]> csvData = csvFile1.map(new ParseLine());
csvData.foreach(x->{
for(String s : x){
System.out.println(s);
}
}
);
}
static void writeCsv1(JavaSparkContext sc) {
JavaRDD<String> csvFile1 = sc.textFile("file:///d:/jsontext/csvsong.csv");
JavaRDD<String[]> parsedData = csvFile1.map(new ParseLine());
parsedData = parsedData.filter(x->x[2].equals("怀旧专辑")); //过滤 如果在这里存文件的话，存的是数组类型的对象
parsedData.foreach(
x->{
long id = Thread.currentThread().getId();
System.out.println("在线程 "+ id +" 中" + "打印当前数据元素：");
for(String s : x){
System.out.print(s+ " ");
}
System.out.println();
}
);
parsedData.map(x->{
StringWriter stringWriter = new StringWriter();
CSVWriter csvWriter = new CSVWriter(stringWriter);
csvWriter.writeNext(x); //把数组转换成为CSV的格式
csvWriter.close();
return stringWriter.toString();
}).saveAsTextFile("file:///d:/jsontext/csvout");
}
public static class ParseLine implements Function<String, String[]> {
public String[] call(String line) throws Exception {
CSVReader reader = new CSVReader(new StringReader(line));
String[] lineData = reader.readNext();
reader.close(); //关闭流资源
// String[] lineData =line.split(","); //这样还有
return lineData;
}
}
static void readCsv2(JavaSparkContext sc){
//如果文件中有断行，wholetextfile可以跳行
JavaPairRDD<String, String> csvData = sc.wholeTextFiles("d:/jsontext/csvsong.csv");
JavaRDD<String[]> keyedRDD = csvData.flatMap(new ParseLineWhole());
keyedRDD.foreach(x->
{
for(String s : x){
System.out.println(s);
}
}
);
}
public static class ParseLineWhole implements FlatMapFunction<Tuple2<String, String>, String[]> {
public Iterator<String[]> call(Tuple2<String, String> file) throws Exception {
CSVReader reader = new CSVReader(new StringReader(file._2()));
Iterator<String[]> data = reader.readAll().iterator();
reader.close();
return data;
}
}
}
/*
"上海滩","叶丽仪","香港电视剧主题歌","mp3/shanghaitan.mp3"
"一生何求","陈百强","香港电视剧主题歌","mp3/shanghaitan.mp3"
"红日","李克勤","怀旧专辑","mp3/shanghaitan.mp3"
"爱如潮水","张信哲","怀旧专辑","mp3/airucaoshun.mp3"
"红茶馆","陈惠嫻","怀旧专辑","mp3/redteabar.mp3"
*/seq二进制文件
import java.util.ArrayList;
import java.util.Arrays;
import java.util.List;
import org.apache.hadoop.io.IntWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapreduce.lib.output.SequenceFileOutputFormat;
import org.apache.spark.SparkConf;
import org.apache.spark.api.java.JavaPairRDD;
import org.apache.spark.api.java.JavaSparkContext;
import org.apache.spark.api.java.function.PairFunction;
import org.apache.spark.serializer.KryoSerializer;
import scala.Tuple2;
public class SparkIO_SeqFile {
public static void main(String[] args) {
//多线程，开了两个线程
SparkConf conf = new SparkConf().setMaster("local[2]").setAppName("SparkIO")
.set("spark.testing.memory", "2147480000")
.set("spark.serializer", "org.apache.spark.serializer.KryoSerializer");
JavaSparkContext sc = new JavaSparkContext(conf);
sc.setLogLevel("WARN");
//sequenceFile存取的是键值对，是序列化文本文件（将对象转换为二进制形式）
writeSeqFile(sc);
readSeqFile(sc);
sc.stop();
sc.close();
}
private static class ConvertToNativeTypes implements PairFunction<Tuple2<Text, IntWritable>, String, Integer> {
public Tuple2<String, Integer> call(Tuple2<Text, IntWritable> record) {
return new Tuple2<String, Integer>(record._1.toString(), record._2.get());
}
}
private static void writeSeqFile(JavaSparkContext sc) {
List<Tuple2<String, Integer>> data = new ArrayList<Tuple2<String, Integer>>();
data.add(new Tuple2<String, Integer>("ABC", 1));
data.add(new Tuple2<String, Integer>("DEF", 3));
data.add(new Tuple2<String, Integer>("GHI", 2));
data.add(new Tuple2<String, Integer>("JKL", 4));
data.add(new Tuple2<String, Integer>("ABC", 1));
// JavaPairRDD<String, Integer> rdd1 = sc.parallelizePairs(Arrays.asList(("d",1)),1);
//设置分区数，有多少个分区数就有多少个输出文件
JavaPairRDD<String, Integer> rdd = sc.parallelizePairs(data, 1);
String dir = "file:///D:jsontext/sequenceFile";
//sequenceFile将键值对使用maptoPair装换为文本类型的键值对
JavaPairRDD<Text, IntWritable> result = rdd.mapToPair(new ConvertToWritableTypes());
//四个参数，文件名，输出键值对的类型，输出格式 saveAsNewAPIHadoopFile是新接口
result.saveAsNewAPIHadoopFile(dir, Text.class, IntWritable.class, SequenceFileOutputFormat.class);
}
static class ConvertToWritableTypes implements PairFunction<Tuple2<String, Integer>, Text, IntWritable> {
public Tuple2<Text, IntWritable> call(Tuple2<String, Integer> record) {
return new Tuple2<Text, IntWritable>(new Text(record._1), new IntWritable(record._2));
}
}
private static void readSeqFile(JavaSparkContext sc) {
//读取sequenceFile文件，输出到PairRDD，三个参数，文件名，输入键值对类型
JavaPairRDD<Text, IntWritable> input = sc.sequenceFile(
"file:///D:/jsontext/sequenceFile",
Text.class,
IntWritable.class);
// input.foreach(System.out::println);
//调用mapToPair将文件的键值对装换为string的键值对类型，输出
JavaPairRDD<String, Integer> result = input.mapToPair(new ConvertToNativeTypes());
result.foreach(x->System.out.println(x));
}
}
object文件
import java.io.Serializable;
import java.util.ArrayList;
import java.util.List;
import org.apache.hadoop.io.IntWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapreduce.lib.output.SequenceFileOutputFormat;
import org.apache.spark.SparkConf;
import org.apache.spark.api.java.JavaPairRDD;
import org.apache.spark.api.java.JavaRDD;
import org.apache.spark.api.java.JavaSparkContext;
import org.apache.spark.api.java.function.PairFunction;
import scala.Tuple2;
public class SparkIO_ObjFile {
public static void main(String[] args) {
SparkConf conf = new SparkConf().setMaster("local[2]").setAppName("SparkIO").set("spark.testing.memory", "2147480000");
JavaSparkContext sc = new JavaSparkContext(conf);
sc.setLogLevel("WARN");
writeObjFile(sc);
//文件所读取的对象是person对象，输出的形式为person对象，所以如果没有了person对象，foreach输出将会报错
readObjFile(sc);
sc.stop();
sc.close();
}
private static void readObjFile(JavaSparkContext sc) {
//object二进制文件读取为rdd
JavaRDD<Object> input = sc.objectFile("file:///D:/jsontext/objFile");
//输出object文件时自动读取引用的person对象，如果person对象不存在，将会报错，终止操作
input.foreach(x->System.out.println(x));
}
private static void writeObjFile(JavaSparkContext sc) {
List<Person> data = new ArrayList<Person>();
data.add(new Person("ABC", 1));
data.add(new Person("DEF", 3));
data.add(new Person("GHI", 2));
data.add(new Person("JKL", 4));
data.add(new Person("ABC", 1));
//设置分区数，多少个分区数有多少个个输出文件
JavaRDD<Person> rdd = sc.parallelize(data, 2);
//将文件保存为textFile类型，输出为文本文件，可见的文本为tostring方法
String dir = "file:///D:/jsontext/textFile";
rdd.saveAsTextFile(dir);
//输出为objectFile类型，为二进制文件，此文件保存的是对象的类型和值，类型为文本类型，值为二进制类型，使用saveAsObject方法存到文件
//objectFile存储只包含值的rdd
String dir1 = "file:///D:/jsontext/objFile";
rdd.saveAsObjectFile(dir1);
}
static class Person implements Serializable{
public Person(String name, int id) {
super();
this.name = name;
this.id = id;
}
@Override
public String toString() {
return "Person [name=" + name + ", id=" + id + "]";
}
String name;
int id;
}
}
二、Hadoop支持格式
1、例如：KeyValueTextInputFormat 是最简单的 Hadoop 输入格式之一，可以用于从文本文件中读取 键值对数据。每一行都会被独立处理，键和值之间用制表符隔开。
newAPIHadoopFile/saveAsNewAPIHadoopFile
2、非文件系统数据（HBase/MongoDB）
使用newAPIHadoopDataset/saveAsNewAPIHadoopDataset
3、Protocol buffer（简称 PB，https://github.com/google/protobuf）
三、文件压缩
四、文件系统
1、本地文件系统
file:///D:/sequenceFile
file:///home/sequenceFile
Spark 支持从本地文件系统中读取文件，不过它要求文件在集群中所有节点的相同路径下 都可以找到。
一些像 NFS、AFS 以及 MapR 的 NFS layer 这样的网络文件系统会把文件以本地文件系统 的形式暴露给用户。如果你的数据已经在这些系统中，那么你只需要指定输入为一个 file:// 路径；只要这个文件系统挂载在每个节点的同一个路径下，Spark 就会自动处理（如例 5-29 所示）。如果文件还没有放在集群中的所有节点上，你可以在驱动器程序中从本地读取该文件而无 需使用整个集群，然后再调用 parallelize 将内容分发给工作节点。不过这种方式可能会 比较慢，所以推荐的方法是将文件先放到像 HDFS、NFS、S3 等共享文件系统上。
2、 网络文件系统
file:///hdfs:localhost:7088/ sequenceFile
五、数据库
1、JDBC
2、Cassandra
3、HBase
4、Elasticsearch