spark读写hbase性能对比
一、spark写入hbase
hbase client以put方式封装数据，并支持逐条或批量插入。spark中内置saveAsHadoopDataset和saveAsNewAPIHadoopDataset两种方式写入hbase。为此，将同样的数据插入其中对比性能。
依赖如下：
<!-- https://mvnrepository.com/artifact/org.apache.spark/spark-core -->
<dependency>
<groupId>org.apache.spark</groupId>
<artifactId>spark-core_2.11</artifactId>
<version>2.3.1</version>
</dependency>
<!-- https://mvnrepository.com/artifact/org.apache.hbase/hbase-client -->
<dependency>
<groupId>org.apache.hbase</groupId>
<artifactId>hbase-client</artifactId>
<version>1.4.6</version>
</dependency>
<!-- https://mvnrepository.com/artifact/org.apache.hbase/hbase-common -->
<dependency>
<groupId>org.apache.hbase</groupId>
<artifactId>hbase-common</artifactId>
<version>1.4.6</version>
</dependency>
<!-- https://mvnrepository.com/artifact/org.apache.hbase/hbase-server -->
<dependency>
<groupId>org.apache.hbase</groupId>
<artifactId>hbase-server</artifactId>
<version>1.4.6</version>
</dependency>
<!-- https://mvnrepository.com/artifact/org.apache.hbase/hbase-protocol -->
<dependency>
<groupId>org.apache.hbase</groupId>
<artifactId>hbase-protocol</artifactId>
<version>1.4.6</version>
</dependency>
<!-- https://mvnrepository.com/artifact/commons-cli/commons-cli -->
<dependency>
<groupId>commons-cli</groupId>
<artifactId>commons-cli</artifactId>
<version>1.4</version>
</dependency>
1. put逐条插入
1.1 hbase客户端建表
create 'keyword1',{NAME=>'info',BLOCKSIZE=>'16384',BLOCKCACHE=>'false'},{NUMREGIONS=>10,SPLITALGO=>'HexStringSplit'}
1.2 code
val start_time1 = new Date().getTime
keyword.foreachPartition(records =>{
HBaseUtils1x.init()
records.foreach(f => {
val keyword = f.getString(0)
val app_id = f.getString(1)
val catalog_name = f.getString(2)
val keyword_catalog_pv = f.getString(3)
val keyword_catalog_pv_rate = f.getString(4)
val rowKey = MD5Hash.getMD5AsHex(Bytes.toBytes(keyword+app_id)).substring(0,8)
val cols = Array(keyword,app_id,catalog_name,keyword_catalog_pv,keyword_catalog_pv_rate)
HBaseUtils1x.insertData(tableName1, HBaseUtils1x.getPutAction(rowKey, cf, columns, cols))
})
HBaseUtils1x.closeConnection()
})
var end_time1 =new Date().getTime
println("HBase逐条插入运行时间为：" + (end_time1 - start_time1))
2.put批量插入
2.1 建表
create 'keyword2',{NAME=>'info',BLOCKSIZE=>'16384',BLOCKCACHE=>'false'},{NUMREGIONS=>10,SPLITALGO=>'HexStringSplit'}
2.2 代码
val start_time2 = new Date().getTime
keyword.foreachPartition(records =>{
HBaseUtils1x.init()
val puts = ArrayBuffer[Put]()
records.foreach(f => {
val keyword = f.getString(0)
val app_id = f.getString(1)
val catalog_name = f.getString(2)
val keyword_catalog_pv = f.getString(3)
val keyword_catalog_pv_rate = f.getString(4)
val rowKey = MD5Hash.getMD5AsHex(Bytes.toBytes(keyword+app_id)).substring(0,8)
val cols = Array(keyword,app_id,catalog_name,keyword_catalog_pv,keyword_catalog_pv_rate)
try{
puts.append(HBaseUtils1x.getPutAction(rowKey,
cf, columns, cols))
}catch{
case e:Throwable => println(f)
}
})
import collection.JavaConverters._
HBaseUtils1x.addDataBatchEx(tableName2, puts.asJava)
HBaseUtils1x.closeConnection()
})
val end_time2 = new Date().getTime
println("HBase批量插入运行时间为：" + (end_time2 - start_time2))
3. saveAsHadoopDataset写入
使用旧的Hadoop API将RDD输出到任何Hadoop支持的存储系统，为该存储系统使用Hadoop JobConf对象。JobConf设置一个OutputFormat和任何需要输出的路径，就像为Hadoop MapReduce作业配置那样。
3.1 建表
create 'keyword3',{NAME=>'info',BLOCKSIZE=>'16384',BLOCKCACHE=>'false'},{NUMREGIONS=>10,SPLITALGO=>'HexStringSplit'}
3.2 代码
val start_time3 = new Date().getTime
keyword.rdd.map(f =>{
val keyword = f.getString(0)
val app_id = f.getString(1)
val catalog_name = f.getString(2)
val keyword_catalog_pv = f.getString(3)
val keyword_catalog_pv_rate = f.getString(4)
val rowKey = MD5Hash.getMD5AsHex(Bytes.toBytes(keyword+app_id)).substring(0,8)
val cols = Array(keyword,app_id,catalog_name,keyword_catalog_pv,keyword_catalog_pv_rate)
(new ImmutableBytesWritable, HBaseUtils1x.getPutAction(rowKey, cf, columns, cols))
}).saveAsHadoopDataset(HBaseUtils1x.getJobConf(tableName3))
val end_time3 = new Date().getTime
println("saveAsHadoopDataset方式写入运行时间为：" + (end_time3 - start_time3))
4. saveAsNewAPIHadoopDataset写入
使用新的Hadoop API将RDD输出到任何Hadoop支持存储系统，为该存储系统使用Hadoop Configuration对象.Conf设置一个OutputFormat和任何需要的输出路径，就像为Hadoop MapReduce作业配置那样。
4.1 建表
create 'keyword4',{NAME=>'info',BLOCKSIZE=>'16384',BLOCKCACHE=>'false'},{NUMREGIONS=>10,SPLITALGO=>'HexStringSplit'}
4.2 code
val start_time4 = new Date().getTime
keyword.rdd.map(f =>{
val keyword = f.getString(0)
val app_id = f.getString(1)
val catalog_name = f.getString(2)
val keyword_catalog_pv = f.getString(3)
val keyword_catalog_pv_rate = f.getString(4)
val rowKey = MD5Hash.getMD5AsHex(Bytes.toBytes(keyword+app_id)).substring(0,8)
val cols = Array(keyword,app_id,catalog_name,keyword_catalog_pv,keyword_catalog_pv_rate)
(new ImmutableBytesWritable, HBaseUtils1x.getPutAction(rowKey, cf, columns, cols))
}).saveAsNewAPIHadoopDataset(HBaseUtils1x.getNewJobConf(tableName4,spark.sparkContext))
val end_time4 = new Date().getTime
println("saveAsNewAPIHadoopDataset方式写入运行时间为：" + (end_time4 - start_time4))
5. 性能对比
可以看出，saveAsHadoopDataset和saveAsNewAPIHadoopDataset方式要优于put逐条插入和批量插入。
二、spark读取hbase
newAPIHadoopRDD API可以将hbase表转化为RDD，具体使用如下：
val start_time1 = new Date().getTime
val hbaseRdd = spark.sparkContext.newAPIHadoopRDD(HBaseUtils1x.getNewConf(tableName1), classOf[TableInputFormat], classOf[ImmutableBytesWritable], classOf[Result])
println(hbaseRdd.count())
hbaseRdd.foreach{
case(_,result) => {
// 获取行键
val rowKey = Bytes.toString(result.getRow)
val keyword = Bytes.toString(result.getValue(cf.getBytes(), "keyword".getBytes()))
val keyword_catalog_pv_rate = Bytes.toDouble(result.getValue(cf.getBytes(), "keyword_catalog_pv_rate".getBytes()))
println(rowKey + "," + keyword + "," + keyword_catalog_pv_rate)
}
}
三、完整代码
package com.sparkStudy.utils
import java.util.Date
import org.apache.hadoop.hbase.client.{Put, Result}
import org.apache.hadoop.hbase.io.ImmutableBytesWritable
import org.apache.hadoop.hbase.mapreduce.TableInputFormat
import org.apache.hadoop.hbase.util.{Bytes, MD5Hash}
import org.apache.spark.sql.SparkSession
import scala.collection.mutable.ArrayBuffer
/**
* @Author: JZ.lee
* @Description: TODO
* @Date: 18-8-28 下午4:28
* @Modified By:
*/
object SparkRWHBase {
def main(args: Array[String]): Unit = {
val spark = SparkSession.builder()
.appName("SparkRWHBase")
.master("local[2]")
.config("spark.some.config.option", "some-value")
.getOrCreate()
val keyword = spark.read
.format("org.apache.spark.sql.execution.datasources.csv.CSVFileFormat")
.option("header",false)
.option("delimiter",",")
.load("file:/opt/data/keyword_catalog_day.csv")
val tableName1 = "keyword1"
val tableName2 = "keyword2"
val tableName3 = "keyword3"
val tableName4 = "keyword4"
val cf = "info"
val columns = Array("keyword", "app_id", "catalog_name", "keyword_catalog_pv", "keyword_catalog_pv_rate")
val start_time1 = new Date().getTime
keyword.foreachPartition(records =>{
HBaseUtils1x.init()
records.foreach(f => {
val keyword = f.getString(0)
val app_id = f.getString(1)
val catalog_name = f.getString(2)
val keyword_catalog_pv = f.getString(3)
val keyword_catalog_pv_rate = f.getString(4)
val rowKey = MD5Hash.getMD5AsHex(Bytes.toBytes(keyword+app_id)).substring(0,8)
val cols = Array(keyword,app_id,catalog_name,keyword_catalog_pv,keyword_catalog_pv_rate)
HBaseUtils1x.insertData(tableName1, HBaseUtils1x.getPutAction(rowKey, cf, columns, cols))
})
HBaseUtils1x.closeConnection()
})
var end_time1 =new Date().getTime
println("HBase逐条插入运行时间为：" + (end_time1 - start_time1))
val start_time2 = new Date().getTime
keyword.foreachPartition(records =>{
HBaseUtils1x.init()
val puts = ArrayBuffer[Put]()
records.foreach(f => {
val keyword = f.getString(0)
val app_id = f.getString(1)
val catalog_name = f.getString(2)
val keyword_catalog_pv = f.getString(3)
val keyword_catalog_pv_rate = f.getString(4)
val rowKey = MD5Hash.getMD5AsHex(Bytes.toBytes(keyword+app_id)).substring(0,8)
val cols = Array(keyword,app_id,catalog_name,keyword_catalog_pv,keyword_catalog_pv_rate)
try{
puts.append(HBaseUtils1x.getPutAction(rowKey,
cf, columns, cols))
}catch{
case e:Throwable => println(f)
}
})
import collection.JavaConverters._
HBaseUtils1x.addDataBatchEx(tableName2, puts.asJava)
HBaseUtils1x.closeConnection()
})
val end_time2 = new Date().getTime
println("HBase批量插入运行时间为：" + (end_time2 - start_time2))
val start_time3 = new Date().getTime
keyword.rdd.map(f =>{
val keyword = f.getString(0)
val app_id = f.getString(1)
val catalog_name = f.getString(2)
val keyword_catalog_pv = f.getString(3)
val keyword_catalog_pv_rate = f.getString(4)
val rowKey = MD5Hash.getMD5AsHex(Bytes.toBytes(keyword+app_id)).substring(0,8)
val cols = Array(keyword,app_id,catalog_name,keyword_catalog_pv,keyword_catalog_pv_rate)
(new ImmutableBytesWritable, HBaseUtils1x.getPutAction(rowKey, cf, columns, cols))
}).saveAsHadoopDataset(HBaseUtils1x.getJobConf(tableName3))
val end_time3 = new Date().getTime
println("saveAsHadoopDataset方式写入运行时间为：" + (end_time3 - start_time3))
//
val start_time4 = new Date().getTime
keyword.rdd.map(f =>{
val keyword = f.getString(0)
val app_id = f.getString(1)
val catalog_name = f.getString(2)
val keyword_catalog_pv = f.getString(3)
val keyword_catalog_pv_rate = f.getString(4)
val rowKey = MD5Hash.getMD5AsHex(Bytes.toBytes(keyword+app_id)).substring(0,8)
val cols = Array(keyword,app_id,catalog_name,keyword_catalog_pv,keyword_catalog_pv_rate)
(new ImmutableBytesWritable, HBaseUtils1x.getPutAction(rowKey, cf, columns, cols))
}).saveAsNewAPIHadoopDataset(HBaseUtils1x.getNewJobConf(tableName4,spark.sparkContext))
val end_time4 = new Date().getTime
println("saveAsNewAPIHadoopDataset方式写入运行时间为：" + (end_time4 - start_time4))
val hbaseRdd = spark.sparkContext.newAPIHadoopRDD(HBaseUtils1x.getNewConf(tableName1), classOf[TableInputFormat], classOf[ImmutableBytesWritable], classOf[Result])
println(hbaseRdd.count())
hbaseRdd.foreach{
case(_,result) => {
// 获取行键
val rowKey = Bytes.toString(result.getRow)
val keyword = Bytes.toString(result.getValue(cf.getBytes(), "keyword".getBytes()))
val keyword_catalog_pv_rate = Bytes.toDouble(result.getValue(cf.getBytes(), "keyword_catalog_pv_rate".getBytes()))
println(rowKey + "," + keyword + "," + keyword_catalog_pv_rate)
}
}
}
}
package com.sparkStudy.utils
import org.apache.hadoop.conf.Configuration
import org.apache.hadoop.hbase.client.BufferedMutator.ExceptionListener
import org.apache.hadoop.hbase.client._
import org.apache.hadoop.hbase.io.ImmutableBytesWritable
import org.apache.hadoop.hbase.protobuf.ProtobufUtil
import org.apache.hadoop.hbase.util.{Base64, Bytes}
import org.apache.hadoop.hbase.{HBaseConfiguration, HColumnDescriptor, HTableDescriptor, TableName}
import org.apache.hadoop.mapred.JobConf
import org.apache.hadoop.mapreduce.Job
import org.apache.spark.SparkContext
import org.slf4j.LoggerFactory
/**
* @Author: JZ.Lee
* @Description:HBase1x增删改查
* @Date: Created at 上午11:02 18-8-14
* @Modified By:
*/
object HBaseUtils1x {
private val LOGGER = LoggerFactory.getLogger(this.getClass)
private var connection:Connection = null
private var conf:Configuration = null
def init() = {
conf = HBaseConfiguration.create()
conf.set("hbase.zookeeper.quorum", "lee")
connection = ConnectionFactory.createConnection(conf)
}
def getJobConf(tableName:String) = {
val conf = HBaseConfiguration.create()
val jobConf = new JobConf(conf)
jobConf.set("hbase.zookeeper.quorum", "lee")
jobConf.set("hbase.zookeeper.property.clientPort", "2181")
jobConf.set(org.apache.hadoop.hbase.mapred.TableOutputFormat.OUTPUT_TABLE,tableName)
jobConf.setOutputFormat(classOf[org.apache.hadoop.hbase.mapred.TableOutputFormat])
jobConf
}
def getNewConf(tableName:String) = {
conf = HBaseConfiguration.create()
conf.set("hbase.zookeeper.quorum", "lee")
conf.set("hbase.zookeeper.property.clientPort", "2181")
conf.set(org.apache.hadoop.hbase.mapreduce.TableInputFormat.INPUT_TABLE,tableName)
val scan = new Scan()
conf.set(org.apache.hadoop.hbase.mapreduce.TableInputFormat.SCAN,Base64.encodeBytes(ProtobufUtil.toScan(scan).toByteArray))
conf
}
def getNewJobConf(tableName:String) = {
val conf = HBaseConfiguration.create()
conf.set("hbase.zookeeper.quorum", Constants.ZOOKEEPER_SERVER_NODE)
conf.set("hbase.zookeeper.property.clientPort", "2181")
conf.set("hbase.defaults.for.version.skip", "true")
conf.set(org.apache.hadoop.hbase.mapreduce.TableOutputFormat.OUTPUT_TABLE, tableName)
conf.setClass("mapreduce.job.outputformat.class", classOf[org.apache.hadoop.hbase.mapreduce.TableOutputFormat[String]],
classOf[org.apache.hadoop.mapreduce.OutputFormat[String, Mutation]])
new JobConf(conf)
}
def closeConnection(): Unit = {
connection.close()
}
def getGetAction(rowKey: String):Get = {
val getAction = new Get(Bytes.toBytes(rowKey));
getAction.setCacheBlocks(false);
getAction
}
def getPutAction(rowKey: String, familyName:String, column: Array[String], value: Array[String]):Put = {
val put: Put = new Put(Bytes.toBytes(rowKey));
for (i <- 0 until(column.length)) {
put.add(Bytes.toBytes(familyName), Bytes.toBytes(column(i)), Bytes.toBytes(value(i)));
}
put
}
def insertData(tableName:String, put: Put) = {
val name = TableName.valueOf(tableName)
val table = connection.getTable(name)
table.put(put)
}
def addDataBatchEx(tableName:String, puts:java.util.List[Put]): Unit = {
val name = TableName.valueOf(tableName)
val table = connection.getTable(name)
val listener = new ExceptionListener {
override def onException
(e: RetriesExhaustedWithDetailsException, bufferedMutator: BufferedMutator): Unit = {
for(i <-0 until e.getNumExceptions){
LOGGER.info("写入put失败:" + e.getRow(i))
}
}
}
val params = new BufferedMutatorParams(name)
.listener(listener)
.writeBufferSize(4*1024*1024)
try{
val mutator = connection.getBufferedMutator(params)
mutator.mutate(puts)
mutator.close()
}catch {
case e:Throwable => e.printStackTrace()
}
}
}
https://blog.csdn.net/baymax_007/article/details/82191188
spark读写hbase性能对比的更多相关文章
- Spark读写HBase
Spark读写HBase示例 1.HBase shell查看表结构 hbase(main)::> desc 'SDAS_Person' Table SDAS_Person is ENABLED ...
- Spark读写Hbase的二种方式对比
作者:Syn良子 出处:http://www.cnblogs.com/cssdongl 转载请注明出处 一.传统方式 这种方式就是常用的TableInputFormat和TableOutputForm ...
- Spark读写HBase时出现的问题--RpcRetryingCaller: Call exception
问题描述 Exception in thread "main" org.apache.hadoop.hbase.client.RetriesExhaustedException: ...
- 顺序、随机IO和Java多种读写文件性能对比
概述 对于磁盘的读写分为两种模式,顺序IO和随机IO. 随机IO存在一个寻址的过程,所以效率比较低.而顺序IO,相当于有一个物理索引,在读取的时候不需要寻找地址,效率很高. 基本流程 总体结构 我们编 ...
- Spark读写Hbase中的数据
def main(args: Array[String]) { val sparkConf = new SparkConf().setMaster("local").setAppN ...
- Spark-读写HBase，SparkStreaming操作，Spark的HBase相关操作
Spark-读写HBase,SparkStreaming操作,Spark的HBase相关操作 1.sparkstreaming实时写入Hbase(saveAsNewAPIHadoopDataset方法 ...
- HBase在单Column和多Column情况下批量Put的性能对比分析
作者: 大圆那些事 | 文章可以转载,请以超链接形式标明文章原始出处和作者信息 网址: http://www.cnblogs.com/panfeng412/archive/2013/11/28/hba ...
- Spark实战之读写HBase
1 配置 1.1 开发环境: HBase:hbase-1.0.0-cdh5.4.5.tar.gz Hadoop:hadoop-2.6.0-cdh5.4.5.tar.gz ZooKeeper:zooke ...
- Hadoop vs Spark性能对比
http://www.cnblogs.com/jerrylead/archive/2012/08/13/2636149.html Hadoop vs Spark性能对比 基于Spark-0.4和Had ...
随机推荐
- C++、Java语法差异对照表
C++.Java语法差异对照表 C++ and Java Syntax Differences Cheat Sheet First, two big things--the main function ...
- leetcode — distinct-subsequences
import java.util.Arrays; /** * * Source : https://oj.leetcode.com/problems/distinct-subsequences/ * ...
- linux centos6.5安装KVM
1.安装kvm,vnc软件 http://www.server110.com/kvm/201404/10467.html yum install kvm,缺啥补啥 yum install kvm km ...
- EF Core新增迁移时无法加载程序集“System.ValueTuple”的错误
EF Core使用迁移命令时,如: Add-Migration Init 有时会出现如下错误: System.IO.FileLoadException: Could not load file or ...
- Perl进程：僵尸进程和孤儿进程
概念 僵尸进程:当子进程退出时,父进程还没有(使用wait或waitpid)接收其退出状态时,子进程就成了僵尸进程 孤儿进程:当子进程还在运行时,父进程先退出了,子进程就会成为孤儿进程被pid=1的i ...
- Puppet部署Nginx返代示例
一.创建目录并编辑Nginx安装模块 mkdir -pv /etc/puppet/modules/nginx/{manifests,files,templates,spec,tests,lib} ]# ...
- (摘)老司机也必须掌握的MySQL优化指南
当 MySQL 单表记录数过大时,增删改查性能都会急剧下降,本文会提供一些优化参考,大家可以参考以下步骤来优化. 单表优化 除非单表数据未来会一直不断上涨,否则不要一开始就考虑拆分,拆分会带来逻辑.部 ...
- mysql 的优化
=>记录慢速查询 在一个 SQL 服务器中,数据表都是保存在磁盘上的.索引为服务器提供了一种在表中查找特定数据行的方法,而不用搜索整个表.当必须要搜索整个表时,就称为表扫描.通常 来说,您可能只 ...
- year 和 weak year 的区别
java 中使用 SimpleDateFormat 时,会遇到 year 和 week year 这两个概念,特此记录. google 答案: A week year is in sync with ...
- 做一个开源的小程序登录模块组件（token）
先了解下SSO 对于单点登陆浅显一点的说就是两种,一种web端的基于Cookie.另一种是跨端的基于Token,一般想要做的都优先做Token吧,个人建议,因为后期扩展也方便哦. 小程序也是呢,做成t ...