1.写在前面
在spark streaming+kafka对流式数据处理过程中,往往是spark streaming消费kafka的数据写入hdfs中,再进行hive映射形成数仓,当然也可以利用sparkSQL直接写入hive形成数仓。对于写入hdfs中,如果是普通的rdd则API为saveAsTextFile(),如果是PairRDD则API为saveAsHadoopFile()。当然高版本的spark可能将这两个合二为一。这两种API在spark streaming中如果不自定义的话会导致新写入的hdfs文件覆盖历史写入的hdfs文件,下面来重现这个问题。
2.saveAsTextFile()写新写入的hdfs文件覆盖历史写入的hdfs文件测试代码
package com.surfilter.dp.timer.job;
import kafka.message.MessageAndMetadata;
import kafka.serializer.StringDecoder;
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 org.apache.spark.api.java.function.PairFlatMapFunction;
import org.apache.spark.api.java.function.VoidFunction;
import org.apache.spark.streaming.Seconds;
import org.apache.spark.streaming.api.java.JavaInputDStream;
import org.apache.spark.streaming.api.java.JavaStreamingContext;
import org.apache.spark.streaming.kafka.KafkaUtils;
import java.text.SimpleDateFormat;
import java.util.*;
public class TestStreaming extends BaseParams {
public static void main(String args[]) {
String totalParameterString = null;
if (null != args && args.length > 0) {
totalParameterString = args[0];
}
if (null != totalParameterString && !"".equals(totalParameterString)) {
ParameterParse parameterParse = new ParameterParse(totalParameterString);
SparkConf conf = new SparkConf().setAppName(parameterParse.getSpark_app_name());
setSparkConf(parameterParse, conf);
JavaSparkContext sparkContext = new JavaSparkContext(conf);
JavaStreamingContext streamingContext = new JavaStreamingContext(sparkContext, Seconds.apply(Long.parseLong(parameterParse.getSpark_streaming_duration())));
JavaInputDStream dStream = KafkaUtils.createDirectStream(streamingContext, String.class, String.class,
StringDecoder.class, StringDecoder.class, String.class,
generatorKafkaParams(parameterParse), generatorTopicOffsets(parameterParse, "test_20200509"),
new Function, String>() {
private static final long serialVersionUID = 1L;
@Override
public String call(MessageAndMetadata msgAndMd) throws Exception {
return msgAndMd.message();
}
});
dStream.foreachRDD(new VoidFunction>() {
@Override
public void call(JavaRDD rdd) throws Exception {
JavaRDD saveHdfsRdd = rdd.mapPartitions(new FlatMapFunction, String>() {
@Override
public Iterable call(Iterator iterator) throws Exception {
List returnList = new ArrayList<>();
while (iterator.hasNext()){
String message = iterator.next().toString();
returnList.add(message);
}
return returnList;
}
});
String dt = new SimpleDateFormat("yyyyMMdd").format(new Date());
String hour = new SimpleDateFormat("HH").format(new Date());
String savePath = "hdfs://gawh220:8020/user/hive/warehouse/rzx_standard.db/meijs_test/dt=" + dt + "/hour=" + hour + "/";
saveHdfsRdd.saveAsTextFile(savePath);
}
});
streamingContext.start();
streamingContext.awaitTermination();
streamingContext.close();
}
}
}
在yarn上执行spark streaming观察,用命令行的方式往test_20200509的topic手动生产一段测试数据,发现spark streaming立马检测到并执行完成
之后查看写入的hdfs文件
发现hdfs文件写入正常,也是有数据的。之后不再继续命令行生产数据,当sprak streaming新的一个批次记录为0的任务开始执行并执行完成
再观察写入的hdfs文件,发现文件依然有,但是文件的内容为空,这就证明了第一批有数据的被覆盖掉了
为什么被覆盖?
spark streaming是按照特定的配置时间去一批批的拉取kafka的数据,在写入的时候也是按照分区的状态写入hdfs中的,比如下图
可以看出三个分区写成三个文件,每一批写入都是按照这种方式自动生成文件名并写入文件中,所以会造成最新一批覆盖之前的一批
3.利用saveAsHadoopFile()自定义输出文件格式避免覆盖问题
package com.surfilter.dp.timer.job;
import com.surfilter.dp.timer.parse.BaseParams;
import com.surfilter.dp.timer.parse.ParameterParse;
import kafka.message.MessageAndMetadata;
import kafka.serializer.StringDecoder;
import org.apache.hadoop.mapred.lib.MultipleTextOutputFormat;
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.Function;
import org.apache.spark.api.java.function.PairFlatMapFunction;
import org.apache.spark.api.java.function.VoidFunction;
import org.apache.spark.sql.hive.HiveContext;
import org.apache.spark.streaming.Seconds;
import org.apache.spark.streaming.api.java.JavaInputDStream;
import org.apache.spark.streaming.api.java.JavaStreamingContext;
import org.apache.spark.streaming.kafka.KafkaUtils;
import scala.Tuple2;
import java.text.SimpleDateFormat;
import java.util.ArrayList;
import java.util.Date;
import java.util.Iterator;
import java.util.List;
public class TestStreaming extends BaseParams {
public static void main(String args[]) {
String totalParameterString = null;
if (null != args && args.length > 0) {
totalParameterString = args[0];
}
if (null != totalParameterString && !"".equals(totalParameterString)) {
ParameterParse parameterParse = new ParameterParse(totalParameterString);
SparkConf conf = new SparkConf().setAppName(parameterParse.getSpark_app_name());
setSparkConf(parameterParse, conf);
JavaSparkContext sparkContext = new JavaSparkContext(conf);
JavaStreamingContext streamingContext = new JavaStreamingContext(sparkContext, Seconds.apply(Long.parseLong(parameterParse.getSpark_streaming_duration())));
JavaInputDStream dStream = KafkaUtils.createDirectStream(streamingContext, String.class, String.class,
StringDecoder.class, StringDecoder.class, String.class,
generatorKafkaParams(parameterParse), generatorTopicOffsets(parameterParse, "test_20200509"),
new Function, String>() {
private static final long serialVersionUID = 1L;
@Override
public String call(MessageAndMetadata msgAndMd) throws Exception {
return msgAndMd.message();
}
});
dStream.foreachRDD(new VoidFunction>() {
@Override
public void call(JavaRDD rdd) {
JavaPairRDD pairRDD = rdd.mapPartitionsToPair(new PairFlatMapFunction, String, String>() {
@Override
public Iterable> call(Iterator iterator) {
List> returnTuple = new ArrayList<>();
while (iterator.hasNext()) {
String message = iterator.next().toString();
returnTuple.add(new Tuple2<>(message, ""));
}
return returnTuple;
}
});
String dt = new SimpleDateFormat("yyyyMMdd").format(new Date());
String hour = new SimpleDateFormat("HH").format(new Date());
String savePath = "hdfs://gawh220:8020/user/hive/warehouse/rzx_standard.db/meijs_test/dt=" + dt + "/hour=" + hour + "/";
pairRDD.saveAsHadoopFile(savePath, String.class, String.class, RDDMultipleTextOutputFormat.class);
}
});
streamingContext.start();
streamingContext.awaitTermination();
streamingContext.close();
}
}
}
class RDDMultipleTextOutputFormat extends MultipleTextOutputFormat {
private static String system_time = System.currentTimeMillis() + "";
@Override
protected String generateFileNameForKeyValue(Object key, Object value, String name) {
name = system_time + "-" + name;
return super.generateFileNameForKeyValue(key, value, name);
}
}
用命令行的方式往test_20200509的topic手动生产一段测试数据,发现spark streaming立马检测到并执行完成
之后查看写入的hdfs文件
发现hdfs文件写入正常,也是有数据的。之后不再继续命令行生产数据,当sprak streaming新的一个批次记录为0的任务开始执行并执行完成
再观察写入的hdfs文件,发现并没有产生新的hdfs文件
再命令行的方式往test_20200509的topic手动生产一段测试数据,发现spark streaming立马检测到并执行完成
之后查看写入的hdfs文件,发现新写入的hdfs文件是追加到之前的文件的方式并且有数据的,如果之前的文件大小超过hdfs设定的大小,则会追加新的文件方式
说明:这种方式不但可以避免覆盖问题,而且可以避免大量小文件