Apache HBase是一个分布式的、面向列的开源数据库,它可以让我们随机的、实时的访问大数据。但是怎样有效的将数据导入到HBase呢?HBase有多种导入数据的方法,最直接的方法就是在MapReduce作业中使用TableOutputFormat作为输出,或者使用标准的客户端API,但是这些都不非常有效的方法。
Bulkload利用MapReduce作业输出HBase内部数据格式的表数据,然后将生成的StoreFiles直接导入到集群中。与使用HBase API相比,使用Bulkload导入数据占用更少的CPU和网络资源。
Bulkload过程主要包括三部分:
1.从数据源(通常是文本文件或其他的数据库)提取数据并上传到HDFS
这一步不在HBase的考虑范围内,不管数据源是什么,只要在进行下一步之前将数据上传到HDFS即可。
2.利用一个MapReduce作业准备数据
这一步需要一个MapReduce作业,并且大多数情况下还需要我们自己编写Map函数,而Reduce函数不需要我们考虑,由HBase提供。该作业需要使用rowkey(行键)作为输出Key,KeyValue、Put或者Delete作为输出Value。MapReduce作业需要使用HFileOutputFormat2来生成HBase数据文件。为了有效的导入数据,需要配置HFileOutputFormat2使得每一个输出文件都在一个合适的区域中。为了达到这个目的,MapReduce作业会使用Hadoop的TotalOrderPartitioner类根据表的key值将输出分割开来。HFileOutputFormat2的方法configureIncrementalLoad()会自动的完成上面的工作。
3.告诉RegionServers数据的位置并导入数据
这一步是最简单的,通常需要使用LoadIncrementalHFiles(更为人所熟知是completebulkload工具),将文件在HDFS上的位置传递给它,它就会利用RegionServer将数据导入到相应的区域。
下图简单明确的说明了整个过程
Note:在进行BulkLoad之前,要在HBase中创建与程序中同名且结构相同的空表
Java实现如下:
BulkLoadDriver.java
import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.conf.Configured;
import org.apache.hadoop.fs.FileSystem;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.hbase.HBaseConfiguration;
import org.apache.hadoop.hbase.TableName;
import org.apache.hadoop.hbase.client.Connection;
import org.apache.hadoop.hbase.client.ConnectionFactory;
import org.apache.hadoop.hbase.client.Put;
import org.apache.hadoop.hbase.io.ImmutableBytesWritable;
import org.apache.hadoop.hbase.mapreduce.HFileOutputFormat2;
import org.apache.hadoop.mapreduce.Job;
import org.apache.hadoop.mapreduce.lib.input.FileInputFormat;
import org.apache.hadoop.mapreduce.lib.input.TextInputFormat;
import org.apache.hadoop.mapreduce.lib.output.FileOutputFormat;
import org.apache.hadoop.util.Tool;
import org.apache.hadoop.util.ToolRunner;
/**
* Created by shaobo on 15-6-9.
*/
public class BulkLoadDriver extends Configured implements Tool {
private static final String DATA_SEPERATOR = "\\s+";
private static final String TABLE_NAME = "temperature";//表名
private static final String COLUMN_FAMILY_1="date";//列组1
private static final String COLUMN_FAMILY_2="tempPerHour";//列组2
public static void main(String[] args) {
try {
int response = ToolRunner.run(HBaseConfiguration.create(), new BulkLoadDriver(), args);
if(response == 0) {
System.out.println("Job is successfully completed...");
} else {
System.out.println("Job failed...");
}
} catch(Exception exception) {
exception.printStackTrace();
}
}
public int run(String[] args) throws Exception {
String outputPath = args[1];
/**
* 设置作业参数
*/
Configuration configuration = getConf();
configuration.set("data.seperator", DATA_SEPERATOR);
configuration.set("hbase.table.name", TABLE_NAME);
configuration.set("COLUMN_FAMILY_1", COLUMN_FAMILY_1);
configuration.set("COLUMN_FAMILY_2", COLUMN_FAMILY_2);
Job job = Job.getInstance(configuration, "Bulk Loading HBase Table::" + TABLE_NAME);
job.setJarByClass(BulkLoadDriver.class);
job.setInputFormatClass(TextInputFormat.class);
job.setMapOutputKeyClass(ImmutableBytesWritable.class);//指定输出键类
job.setMapOutputValueClass(Put.class);//指定输出值类
job.setMapperClass(BulkLoadMapper.class);//指定Map函数
FileInputFormat.addInputPaths(job, args[0]);//输入路径
FileSystem fs = FileSystem.get(configuration);
Path output = new Path(outputPath);
if (fs.exists(output)) {
fs.delete(output, true);//如果输出路径存在,就将其删除
}
FileOutputFormat.setOutputPath(job, output);//输出路径
Connection connection = ConnectionFactory.createConnection(configuration);
TableName tableName = TableName.valueOf(TABLE_NAME);
HFileOutputFormat2.configureIncrementalLoad(job, connection.getTable(tableName), connection.getRegionLocator(tableName));
job.waitForCompletion(true);
if (job.isSuccessful()){
HFileLoader.doBulkLoad(outputPath, TABLE_NAME);//导入数据
return 0;
} else {
return 1;
}
}
}
BulkLoadMapper.java
import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.hbase.client.Put;
import org.apache.hadoop.hbase.io.ImmutableBytesWritable;
import org.apache.hadoop.hbase.util.Bytes;
import org.apache.hadoop.io.LongWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapreduce.Mapper;
/**
* Created by shaobo on 15-6-9.
*/
public class BulkLoadMapper extends Mapper<LongWritable, Text, ImmutableBytesWritable, Put> {
private String hbaseTable;
private String dataSeperator;
private String columnFamily1;
private String columnFamily2;
public void setup(Context context) {
Configuration configuration = context.getConfiguration();//获取作业参数
hbaseTable = configuration.get("hbase.table.name");
dataSeperator = configuration.get("data.seperator");
columnFamily1 = configuration.get("COLUMN_FAMILY_1");
columnFamily2 = configuration.get("COLUMN_FAMILY_2");
}
public void map(LongWritable key, Text value, Context context){
try {
String[] values = value.toString().split(dataSeperator);
ImmutableBytesWritable rowKey = new ImmutableBytesWritable(values[0].getBytes());
Put put = new Put(Bytes.toBytes(values[0]));
put.addColumn(Bytes.toBytes(columnFamily1), Bytes.toBytes("month"), Bytes.toBytes(values[1]));
put.addColumn(Bytes.toBytes(columnFamily1), Bytes.toBytes("day"), Bytes.toBytes(values[2]));
for (int i = 3; i < values.length; ++i){
put.addColumn(Bytes.toBytes(columnFamily2), Bytes.toBytes("hour : " + i), Bytes.toBytes(values[i]));
}
context.write(rowKey, put);
} catch(Exception exception) {
exception.printStackTrace();
}
}
}
HFileLoader.java
import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.hbase.HBaseConfiguration;
import org.apache.hadoop.hbase.client.HTable;
import org.apache.hadoop.hbase.mapreduce.LoadIncrementalHFiles;
/**
* Created by shaobo on 15-6-9.
*/
public class HFileLoader {
public static void doBulkLoad(String pathToHFile, String tableName){
try {
Configuration configuration = new Configuration();
HBaseConfiguration.addHbaseResources(configuration);
LoadIncrementalHFiles loadFfiles = new LoadIncrementalHFiles(configuration);
HTable hTable = new HTable(configuration, tableName);//指定表名
loadFfiles.doBulkLoad(new Path(pathToHFile), hTable);//导入数据
System.out.println("Bulk Load Completed..");
} catch(Exception exception) {
exception.printStackTrace();
}
}
}
将程序编译打包,提交到Hadoop运行
HADOOP_CLASSPATH=$(hbase mapredcp):/path/to/hbase/conf hadoop jar BulkLoad.jar inputpath outputpath
上述命令用法可参考44. HBase, MapReduce, and the CLASSPATH
作业运行情况:
15/06/14 14:31:07 INFO mapreduce.HFileOutputFormat2: Looking up current regions for table temperature(表名)
15/06/14 14:31:07 INFO mapreduce.HFileOutputFormat2: Configuring 1 reduce partitions to match current region count
15/06/14 14:31:07 INFO mapreduce.HFileOutputFormat2: Writing partition information to /home/shaobo/hadoop/tmp/partitions_5d464f1e-d412-4dbe-bb98-367f8431bdc9
15/06/14 14:31:07 INFO zlib.ZlibFactory: Successfully loaded & initialized native-zlib library
15/06/14 14:31:07 INFO compress.CodecPool: Got brand-new compressor [.deflate]
15/06/14 14:31:08 INFO mapreduce.HFileOutputFormat2: Incremental table temperature(表名) output configured.
15/06/14 14:31:08 INFO client.RMProxy: Connecting to ResourceManager at localhost/127.0.0.1:8032
15/06/14 14:31:15 INFO input.FileInputFormat: Total input paths to process : 2
15/06/14 14:31:15 INFO mapreduce.JobSubmitter: number of splits:2
15/06/14 14:31:16 INFO mapreduce.JobSubmitter: Submitting tokens for job: job_1434262360688_0002
15/06/14 14:31:17 INFO impl.YarnClientImpl: Submitted application application_1434262360688_0002
15/06/14 14:31:17 INFO mapreduce.Job: The url to track the job: http://shaobo-ThinkPad-E420:8088/proxy/application_1434262360688_0002/
15/06/14 14:31:17 INFO mapreduce.Job: Running job: job_1434262360688_0002
15/06/14 14:31:28 INFO mapreduce.Job: Job job_1434262360688_0002 running in uber mode : false
15/06/14 14:31:28 INFO mapreduce.Job: map 0% reduce 0%
15/06/14 14:32:24 INFO mapreduce.Job: map 49% reduce 0%
15/06/14 14:32:37 INFO mapreduce.Job: map 67% reduce 0%
15/06/14 14:32:43 INFO mapreduce.Job: map 100% reduce 0%
15/06/14 14:33:39 INFO mapreduce.Job: map 100% reduce 67%
15/06/14 14:33:42 INFO mapreduce.Job: map 100% reduce 70%
15/06/14 14:33:45 INFO mapreduce.Job: map 100% reduce 88%
15/06/14 14:33:48 INFO mapreduce.Job: map 100% reduce 100%
15/06/14 14:33:52 INFO mapreduce.Job: Job job_1434262360688_0002 completed successfully
...
...
...
15/06/14 14:34:02 WARN mapreduce.LoadIncrementalHFiles: Skipping non-directory hdfs://localhost:9000/user/output/_SUCCESS
15/06/14 14:34:03 INFO hfile.CacheConfig: CacheConfig:disabled
15/06/14 14:34:03 INFO hfile.CacheConfig: CacheConfig:disabled
15/06/14 14:34:07 INFO mapreduce.LoadIncrementalHFiles: Trying to load hfile=hdfs://localhost:9000/user/output/date/c64cd2524fba48738bab26630d550b61 first=AQW00061705 last=USW00094910
15/06/14 14:34:07 INFO mapreduce.LoadIncrementalHFiles: Trying to load hfile=hdfs://localhost:9000/user/output/tempPerHour/43af29456913444795a820544691eb3d first=AQW00061705 last=USW00094910
Bulk Load Completed..
Job is successfully completed...
BulLoad过程的第三步也可以在用MapReduce作业生成HBase数据文件后在命令行中进行,不一定要与MapReduce过程写在一起。
$ hadoop jar hbase-server-VERSION.jar completebulkload [-c /path/to/hbase/config/hbase-site.xml] outputpath tablename
参考资料:
http://hbase.apache.org/book.html#arch.bulk.load
http://blog.cloudera.com/blog/2013/09/how-to-use-hbase-bulk-loading-and-why/