最近经常被问到这个问题，所以简单写一下总结。
Hive数据导入到HBase基本有2个方案：
1、HBase中建表，然后Hive中建一个外部表，这样当Hive中写入数据后，HBase中也会同时更新
2、MapReduce读取Hive数据，然后写入（API或者Bulkload）到HBase
1、Hive 外部表
创建hbase表
(1) 建立一个表格classes具有1个列族user
create 'classes','user'
(2) 查看表的构造
1hbase(main):005:0> describe 'classes' 2DESCRIPTION ENABLED 3 'classes', {NAME => 'user', DATA_BLOCK_ENCODING => 'NONE', BLOOMFILTER => 'ROW', REPLICATION_SCOPE => '0', true 4 VERSIONS => '1', COMPRESSION => 'NONE', MIN_VERSIONS => '0', TTL => '2147483647', KEEP_DELETED_CELLS => ' 5 false', BLOCKSIZE => '65536', IN_MEMORY => 'false', BLOCKCACHE => 'true'}
(3) 加入2行数据
1put 'classes','001','user:name','jack' 2put 'classes','001','user:age','20' 3put 'classes','002','user:name','liza' 4put 'classes','002','user:age','18'
(4) 查看classes中的数据
1hbase(main):016:0> scan 'classes' 2ROW COLUMN+CELL 3 001 column=user:age, timestamp=1404980824151, value=20 4 001 column=user:name, timestamp=1404980772073, value=jack 5 002 column=user:age, timestamp=1404980963764, value=18 6 002 column=user:name, timestamp=1404980953897, value=liza
(5) 创建外部hive表,查询验证
1create external table classes(id int, name string, age int) 2STORED BY 'org.apache.hadoop.hive.hbase.HBaseStorageHandler' 3WITH SERDEPROPERTIES ("hbase.columns.mapping" = ":key,user:name,user:age") 4TBLPROPERTIES("hbase.table.name" = "classes"); 5 6select * from classes; 7OK 81 jack 20 92 liza 18
（6）再添加数据到HBase
1put 'classes','003','user:age','1820183291839132' 2hbase(main):025:0> scan 'classes' 3ROW COLUMN+CELL 4 001 column=user:age, timestamp=1404980824151, value=20 5 001 column=user:name, timestamp=1404980772073, value=jack 6 002 column=user:age, timestamp=1404980963764, value=18 7 002 column=user:name, timestamp=1404980953897, value=liza 8 003 column=user:age, timestamp=1404981476497, value=1820183291839132
（7）Hive查询，看看新数据
1select * from classes; 2OK 31 jack 20 42 liza 18 53 NULL NULL --这里是null了，因为003没有name，所以补位Null，而age为Null是因为超过最大值
（8）如下作为验证
1put 'classes','004','user:name','test' 2put 'classes','004','user:age','1820183291839112312' -- 已经超int了 3hbase(main):030:0> scan 'classes' 4ROW COLUMN+CELL 5 001 column=user:age, timestamp=1404980824151, value=20 6 001 column=user:name, timestamp=1404980772073, value=jack 7 002 column=user:age, timestamp=1404980963764, value=18 8 002 column=user:name, timestamp=1404980953897, value=liza 9 003 column=user:age, timestamp=1404981476497, value=1820183291839132 10 004 column=user:age, timestamp=1404981558125, value=1820183291839112312 11 004 column=user:name, timestamp=1404981551508, value=test 12select * from classes; 131 jack 20 142 liza 18 153 NULL NULL 164 test NULL -- 超int后也认为是null 17put 'classes','005','user:age','1231342' 18hbase(main):034:0* scan 'classes' 19ROW COLUMN+CELL 20 001 column=user:age, timestamp=1404980824151, value=20 21 001 column=user:name, timestamp=1404980772073, value=jack 22 002 column=user:age, timestamp=1404980963764, value=18 23 002 column=user:name, timestamp=1404980953897, value=liza 24 003 column=user:age, timestamp=1404981476497, value=1820183291839132 25 004 column=user:age, timestamp=1404981558125, value=1820183291839112312 26 004 column=user:name, timestamp=1404981551508, value=test 27 005 column=user:age, timestamp=1404981720600, value=1231342 28select * from classes; 291 jack 20 302 liza 18 313 NULL NULL 324 test NULL 335 NULL 1231342
注意点：
1、hbase中的空cell在hive中会补null
2、hive和hbase中不匹配的字段会补null
3、Bytes类型的数据，建hive表示加#b
http://stackoverflow.com/questions/12909118/number-type-value-in-hbase-not-recognized-by-hive
http://www.aboutyun.com/thread-8023-1-1.html
4、HBase CF to hive Map
https://cwiki.apache.org/confluence/display/Hive/HBaseIntegration
2、MapReduce 写入 HBase
MR写入到HBase有2个常用方法，1是直接调用HBase Api，使用Table 、Put写入；2是通过MR生成HFile，然后Bulkload到HBase，数据量很大的时候推荐使用。
注意点：
1、如果需要从hive的路径中读取一些值怎么办
1private String reg = "stat_date=(.*?)\\/softid=([\\d]+)/"; 2private String stat_date; 3private String softid; 4 ------------厦门map函数中写入------------- 5String filePathString = ((FileSplit) context.getInputSplit()).getPath().toString(); 6///user/hive/warehouse/snapshot.db/stat_all_info/stat_date=20150820/softid=201/000000_0 7// 解析stat_date 和softid 8Pattern pattern = Pattern.compile(reg); 9Matcher matcher = pattern.matcher(filePathString); 10while(matcher.find()){ 11 stat_date = matcher.group(1); 12 softid = matcher.group(2); 13}
2、hive中的map和list怎么处理
hive中的分隔符主要有8种，分别是\001-----> \008
1默认 ^A \001 2, ^B \002 3: ^C \003
Hive中保存的Lis，最底层的数据格式为 jerrick, liza, tom, jerry ， Map的数据格式为 jerrick:23, liza:18, tom:0
所以在MR读入时需要简单处理下，例如map需要： "{"+ mapkey.replace("\002", ",").replace("\003", ":")+"}", 由此再转为JSON， toString后再保存到HBase。
3、简单实例，代码删减很多，仅可参考！
1public void map( 2 LongWritable key, 3 Text value, 4 Mapper<LongWritable, Text, ImmutableBytesWritable, KeyValue>.Context context) { 5 String filePathString = ((FileSplit) context.getInputSplit()).getPath().toString(); 6 ///user/hive/warehouse/snapshot.db/stat_all_info/stat_date=20150820/softid=201/000000_0 7 // 解析stat_date 和softid 8 Pattern pattern = Pattern.compile(reg); 9 Matcher matcher = pattern.matcher(filePathString); 10 while(matcher.find()){ 11 stat_date = matcher.group(1); 12 softid = matcher.group(2); 13 } 14 15 rowMap.put("stat_date", stat_date); 16 rowMap.put("softid", softid); 17 18 String[] vals = value.toString().split("\001"); 19 20 try { 21 Configuration conf = context.getConfiguration(); 22 String cf = conf.get("hbase.table.cf", HBASE_TABLE_COLUME_FAMILY); 23 24 String arow = rowkey; 25 for(int index=10; index < vals.length; index++){ 26 byte[] row = Bytes.toBytes(arow); 27 ImmutableBytesWritable k = new ImmutableBytesWritable(row); 28 KeyValue kv = new KeyValue(); 29 if(index == vals.length-1){ 30 //dict need 31 logger.info("d is :" + vals[index]); 32 logger.info("d is :" + "{"+vals[index].replace("\002", ",").replace("\003", ":")+"}"); 33 34 35 JSONObject json = new JSONObject("{"+vals[index].replace("\002", ",").replace("\003", ":")+"}"); 36 kv = new KeyValue(row, cf.getBytes(),Bytes.toBytes(valueKeys[index]), Bytes.toBytes(json.toString())); 37 }else{ 38 kv = new KeyValue(row, cf.getBytes(),Bytes.toBytes(valueKeys[index]), Bytes.toBytes(vals[index])); 39 } 40 context.write(k, kv); 41 } 42 43 } catch (Exception e1) { 44 context.getCounter("offile2HBase", "Map ERROR").increment(1); 45 logger.info("map error:" + e1.toString()); 46 } 47 48 context.getCounter("offile2HBase", "Map TOTAL").increment(1); 49 50 } 51 }
4、bulkload
1int jobResult = (job.waitForCompletion(true)) ? 0 : 1; 2logger.info("jobResult=" + jobResult); 3Boolean bulkloadHfileToHbase = Boolean.valueOf(conf.getBoolean("hbase.table.hfile.bulkload", false)); 4if ((jobResult == 0) && (bulkloadHfileToHbase.booleanValue())) { 5 LoadIncrementalHFiles loader = new LoadIncrementalHFiles(conf); 6 loader.doBulkLoad(outputDir, hTable); 7}