标签：
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
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
hbase(main):005:0> describe ‘classes‘
DESCRIPTION ENABLED
‘classes‘, {NAME => ‘user‘, DATA_BLOCK_ENCODING => ‘NONE‘, BLOOMFILTER => ‘ROW‘, REPLICATION_SCOPE => ‘0‘, true
VERSIONS => ‘1‘, COMPRESSION => ‘NONE‘, MIN_VERSIONS => ‘0‘, TTL => ‘2147483647‘, KEEP_DELETED_CELLS => ‘
false‘, BLOCKSIZE => ‘65536‘, IN_MEMORY => ‘false‘, BLOCKCACHE => ‘true‘}
(3) 加入2行数据
put ‘classes‘,‘001‘,‘user:name‘,‘jack‘ put ‘classes‘,‘001‘,‘user:age‘,‘20‘ put ‘classes‘,‘002‘,‘user:name‘,‘liza‘ put ‘classes‘,‘002‘,‘user:age‘,‘18‘
(4) 查看classes中的数据
hbase(main):016:0> scan ‘classes‘ ROW COLUMN+CELL 001 column=user:age, timestamp=1404980824151, value=20 001 column=user:name, timestamp=1404980772073, value=jack 002 column=user:age, timestamp=1404980963764, value=18 002 column=user:name, timestamp=1404980953897, value=liza
(5) 创建外部hive表,查询验证
create external table classes(id int, name string, age int)
STORED BY ‘org.apache.hadoop.hive.hbase.HBaseStorageHandler‘
WITH SERDEPROPERTIES ("hbase.columns.mapping" = ":key,user:name,user:age")
TBLPROPERTIES("hbase.table.name" = "classes");
select * from classes;
OK
1 jack 20
2 liza 18
（6）再添加数据到HBase
put ‘classes‘,‘003‘,‘user:age‘,‘1820183291839132‘ hbase(main):025:0> scan ‘classes‘ ROW COLUMN+CELL 001 column=user:age, timestamp=1404980824151, value=20 001 column=user:name, timestamp=1404980772073, value=jack 002 column=user:age, timestamp=1404980963764, value=18 002 column=user:name, timestamp=1404980953897, value=liza 003 column=user:age, timestamp=1404981476497, value=1820183291839132
（7）Hive查询，看看新数据
select * from classes; OK 1 jack 20 2 liza 18 3 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
put ‘classes‘,‘004‘,‘user:name‘,‘test‘ put ‘classes‘,‘004‘,‘user:age‘,‘1820183291839112312‘ -- 已经超int了 hbase(main):030:0> scan ‘classes‘ ROW COLUMN+CELL 001 column=user:age, timestamp=1404980824151, value=20 001 column=user:name, timestamp=1404980772073, value=jack 002 column=user:age, timestamp=1404980963764, value=18 002 column=user:name, timestamp=1404980953897, value=liza 003 column=user:age, timestamp=1404981476497, value=1820183291839132 004 column=user:age, timestamp=1404981558125, value=1820183291839112312 004 column=user:name, timestamp=1404981551508, value=test select * from classes; 1 jack 20 2 liza 18 3 NULL NULL 4 test NULL -- 超int后也认为是null put ‘classes‘,‘005‘,‘user:age‘,‘1231342‘ hbase(main):034:0* scan ‘classes‘ ROW COLUMN+CELL 001 column=user:age, timestamp=1404980824151, value=20 001 column=user:name, timestamp=1404980772073, value=jack 002 column=user:age, timestamp=1404980963764, value=18 002 column=user:name, timestamp=1404980953897, value=liza 003 column=user:age, timestamp=1404981476497, value=1820183291839132 004 column=user:age, timestamp=1404981558125, value=1820183291839112312 004 column=user:name, timestamp=1404981551508, value=test 005 column=user:age, timestamp=1404981720600, value=1231342 select * from classes; 1 jack 20 2 liza 18 3 NULL NULL 4 test NULL 5 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
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
private String reg = "stat_date=(.*?)\\/softid=([\\d]+)/";
private String stat_date;
private String softid;
------------厦门map函数中写入-------------
String filePathString = ((FileSplit) context.getInputSplit()).getPath().toString();
///user/hive/warehouse/snapshot.db/stat_all_info/stat_date=20150820/softid=201/000000_0
// 解析stat_date 和softid
Pattern pattern = Pattern.compile(reg);
Matcher matcher = pattern.matcher(filePathString);
while(matcher.find()){
stat_date = matcher.group(1);
softid = matcher.group(2);
}
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
默认 ^A \001 , ^B \002 : ^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！
public void map(
LongWritable key,
Text value,
Mapper<LongWritable, Text, ImmutableBytesWritable, KeyValue>.Context context) {
String filePathString = ((FileSplit) context.getInputSplit()).getPath().toString();
///user/hive/warehouse/snapshot.db/stat_all_info/stat_date=20150820/softid=201/000000_0
// 解析stat_date 和softid
Pattern pattern = Pattern.compile(reg);
Matcher matcher = pattern.matcher(filePathString);
while(matcher.find()){
stat_date = matcher.group(1);
softid = matcher.group(2);
}
rowMap.put("stat_date", stat_date);
rowMap.put("softid", softid);
String[] vals = value.toString().split("\001");
try {
Configuration conf = context.getConfiguration();
String cf = conf.get("hbase.table.cf", HBASE_TABLE_COLUME_FAMILY);
String arow = rowkey;
for(int index=10; index < vals.length; index++){
byte[] row = Bytes.toBytes(arow);
ImmutableBytesWritable k = new ImmutableBytesWritable(row);
KeyValue kv = new KeyValue();
if(index == vals.length-1){
//dict need
logger.info("d is :" + vals[index]);
logger.info("d is :" + "{"+vals[index].replace("\002", ",").replace("\003", ":")+"}");
JSONObject json = new JSONObject("{"+vals[index].replace("\002", ",").replace("\003", ":")+"}");
kv = new KeyValue(row, cf.getBytes(),Bytes.toBytes(valueKeys[index]), Bytes.toBytes(json.toString()));
}else{
kv = new KeyValue(row, cf.getBytes(),Bytes.toBytes(valueKeys[index]), Bytes.toBytes(vals[index]));
}
context.write(k, kv);
}
} catch (Exception e1) {
context.getCounter("offile2HBase", "Map ERROR").increment(1);
logger.info("map error:" + e1.toString());
}
context.getCounter("offile2HBase", "Map TOTAL").increment(1);
}
}
4、bulkload
int jobResult = (job.waitForCompletion(true)) ? 0 : 1;
logger.info("jobResult=" + jobResult);
Boolean bulkloadHfileToHbase = Boolean.valueOf(conf.getBoolean("hbase.table.hfile.bulkload", false));
if ((jobResult == 0) && (bulkloadHfileToHbase.booleanValue())) {
LoadIncrementalHFiles loader = new LoadIncrementalHFiles(conf);
loader.doBulkLoad(outputDir, hTable);
}
标签：
原文地址：http://my.oschina.net/wangjiankui/blog/497658
