Details
-
Bug
-
Status: Closed
-
Major
-
Resolution: Duplicate
-
0.94.2
-
None
-
None
-
Linux CentOS 6.3 x64, Oracle JDK 6u27 x64
Description
There have been several tickets dealing with reverse DNS inconsistencies causing issues in HBase. I have found that using the TableMapper and TableInputFormat with HBase leads to a situation where there is no data locality in situations where the task tracker DNS name and input split location provided by the input format differ by a trailing period.
In this case I have 0% data-local and 100% rack-local map tasks which take on this format:
task host: /default-rack/hostname.ext
input split: /default-rack/hostname.ext.
It seems to be a crapshoot on whether the task will land on the proper host. The reverse DNS seems to be done differently for this class compared to the others already fixed; It uses the hadoop core DNS util class.