HBase跨集群复制数据的另一种方法

abloz 2012-08-14
2012-08-14

http://abloz.com date:2012.8.14

上一篇文章《hbase 复制备份数据》 中提到用工具CopyTable来在集群间复制数据。另外还有一种更暴力的方式,来共享HBase备份表。因为有时候两个集群并不连通。

一、从源hbase集群中复制出HBase数据库表到本地目录

最好停止HBase,否则可能会丢部分数据

[hbase@hadoop200 ~]$ hadoop fs -get /hbase/toplist_ware_total_1009_201232 toplist_ware_total_1009_201232

压缩

[hbase@hadoop200 ~]$ tar zcvf topl.tar.gz toplist_ware_total_1009_201232

远程复制到目标机器

[hbase@hadoop200 ~]$ scp topl.tar.gz zhouhh@h185:~/.

二、目标HBase导入

解压 [zhouhh@h185 ~]$ tar zxvf topl.tar.gz

如果目标HBase里有这个表,需disable并drop掉。如果有该目录,则用hadoop fs -rmr /hbase/table的方式删除,再往HDFS上复制。以免数据出错。

放到集群下面

[zhouhh@h185 ~]$ fs -put toplist_ware_total_1009_201232 /hbase
[zhouhh@h185 ~]$

此时可以list出来,但scan报错

hbase(main):055:0> list 'toplist_ware_total_1009_201232'
TABLE
toplist_ware_total_1009_201232
1 row(s) in 0.0220 seconds

hbase(main):062:0> scan 'toplist_ware_total_1009_201232'
ROW                                        COLUMN+CELL

ERROR: Unknown table toplist_ware_total_1009_201232!

.META.表里面没有相关记录 hbase(main):064:0> scan ‘.META.’ 里面没有toplist_ware_total_1009_201232 开头的行

三、修复.META.表和重新分配数据到各RegionServer

在.META.表没修复时执行重新分配,会报错

[zhouhh@h185 ~]$ hbase hbck -fixAssignments
...
ERROR: Region { meta => null, hdfs => hdfs://h185:54310/hbase/toplist_ware_total_1009_201232/0403552001eb2a31990e443dcae74ee8, deployed =>  } on HDFS, but not listed in META or deployed on any region server
...

先修复.META.表

[zhouhh@h185 ~]$ hbase hbck -fixMeta
...
ERROR: Region { meta => null, hdfs => hdfs://h185:54310/hbase/toplist_ware_total_1009_201232/0403552001eb2a31990e443dcae74ee8, deployed =>  } on HDFS, but not listed in META or deployed on any region server
12/08/14 18:25:15 INFO util.HBaseFsck: Patching .META. with .regioninfo: {NAME => 'toplist_ware_total_1009_201232,,1344187094829.0403552001eb2a31990e443dcae74ee8.', STARTKEY => '', ENDKEY => '', ENCODED => 0403552001eb2a31990e443dcae74ee8,}
...

此时.META.表已经有表的数据了,但scan还是失败
hbase(main):065:0> scan '.META.'
ROW                                        COLUMN+CELL
...
toplist_ware_total_1009_201232,,134418709 column=info:regioninfo, timestamp=1344939930752, value={NAME => 'toplist_ware_total_1009_201232,,1344187094829.0403552001eb
 4829.0403552001eb2a31990e443dcae74ee8.    2a31990e443dcae74ee8.', STARTKEY => '', ENDKEY => '', ENCODED => 0403552001eb2a31990e443dcae74ee8,}
16 row(s) in 0.0550 seconds

scan还是失败
hbase(main):066:0> scan 'toplist_ware_total_1009_201232'
ROW                                        COLUMN+CELL

ERROR: org.apache.hadoop.hbase.client.NoServerForRegionException: No server address listed in .META. for region toplist_ware_total_1009_201232,,1344187094829.0403552001eb2a31990e443dcae74ee8. containing row

重新分配到各分区服务器

[zhouhh@h185 ~]$ hbase hbck -fixAssignments
...
ERROR: Region { meta => toplist_ware_total_1009_201232,,1344187094829.0403552001eb2a31990e443dcae74ee8., hdfs => hdfs://h185:54310/hbase/toplist_ware_total_1009_201232/0403552001eb2a31990e443dcae74ee8, deployed =>  } not deployed on any region server.
Trying to fix unassigned region...
12/08/14 18:28:01 INFO util.HBaseFsckRepair: Region still in transition, waiting for it to become assigned: {NAME => 'toplist_ware_total_1009_201232,,1344187094829.0403552001eb2a31990e443dcae74ee8.', STARTKEY => '', ENDKEY => '', ENCODED => 0403552001eb2a31990e443dcae74ee8,}
12/08/14 18:28:02 INFO util.HBaseFsckRepair: Region still in transition, waiting for it to become assigned: {NAME => 'toplist_ware_total_1009_201232,,1344187094829.0403552001eb2a31990e443dcae74ee8.', STARTKEY => '', ENDKEY => '', ENCODED => 0403552001eb2a31990e443dcae74ee8,}
12/08/14 18:28:04 INFO util.HBaseFsckRepair: Region still in transition, waiting for it to become assigned: {NAME => 'toplist_ware_total_1009_201232,,1344187094829.0403552001eb2a31990e443dcae74ee8.', STARTKEY => '', ENDKEY => '', ENCODED => 0403552001eb2a31990e443dcae74ee8,}
12/08/14 18:28:05 INFO util.HBaseFsckRepair: Region still in transition, waiting for it to become assigned: {NAME => 'toplist_ware_total_1009_201232,,1344187094829.0403552001eb2a31990e443dcae74ee8.', STARTKEY => '', ENDKEY => '', ENCODED => 0403552001eb2a31990e443dcae74ee8,}
...

scan成功!

hbase(main):067:0> scan 'toplist_ware_total_1009_201232'
ROW                                        COLUMN+CELL
 0000000001                                column=info:loginid, timestamp=1344187147972, value=jjm167258611
 0000000001                                column=info:nick, timestamp=1344187147972, value=?xE9x97xB4?xE6xB5xA3?
 0000000001                                column=info:score, timestamp=1344187147972, value=200
 0000000001                                column=info:userid, timestamp=1344187147972, value=167258611
...
330 row(s) in 0.8630 seconds

如果目标集群是空的,则可以直接将源HBase的/hbase目录复制出来,然后在目标HBase系统上fs -rmr /hbase 或fs -mv /hbase /hbase1 然后用hadoop fs -put hbase / 即可


如非注明转载, 均为原创. 本站遵循知识共享CC协议,转载请注明来源