These days, we are loading a vast bulk of data to HBase. I closed the WAL
for the speed issue.And a region server crashed yesterday, then I restarted
it.  Unfortunately, I found a big problem this morning. The row key between
pox2qq1qV8 and q32i6GEYEP cannot write to HBase any more.

 

I checked the .META. table , found there is a hole in this table.  The
following lines is copied from table view of HBase web manage (
http://xxx.shendusou.com:60010/table.jsp?name=content)

We can find the hole between the second line and the third line. One or more
region which cover pox2qq1qV8 to q32i6GEYEP are lost.

 

content,pguACQ4oZE,1276757449799 d1.shendusou.com:60030
<http://dev1.shendusou.com:60030/>  481213793 pguACQ4oZE pku/KN/RcN 

content,pku/KN/RcN,1276757449799 c3.shendusou.com:60030
<http://content3.shendusou.com:60030/>  543945981 pku/KN/RcN pox2qq1qV8 

content,q32i6GEYEP,1276757404613 c1.shendusou.com:60030
<http://content1.shendusou.com:60030/>  1759457900 q32i6GEYEP q75kqgwVsz 

content,q75kqgwVsz,1276757404613 c1.shendusou.com:60030
<http://content1.shendusou.com:60030/>  2118337301 q75kqgwVsz qB68nrJBY7

 

 

Lose some data is not a problem, Recover it is easy for us, however , we run
into difficulties that we cannot write any data into the table which key is
in the range [pox2qq1qV8 , q32i6GEYEP], It is very bad and blocks us.

 

I found the issue HBase-2526 is working for this problem.
https://issues.apache.org/jira/browse/HBASE-2526 , I do not think it would
be fixed until 0.21.0.

I closed the whole HBase cluster and restarted it many times today, the
problem also exsit.

 

Anybody can tell me how to fix the hole? 

 

Thanks

 

WZeng

Reply via email to