[ 
https://issues.apache.org/jira/browse/GIRAPH-101?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13712299#comment-13712299
 ] 

Michael Aro commented on GIRAPH-101:
------------------------------------

[0] I added the following classes on github: 
https://github.com/mmaro/giraph/tree/trunk/giraph-core/src/main/java/org/apache/giraph/shims

[1] I manually added the hadoop-mapreduce-client-core-2.0.5-alpha.jar to the 
project for the the imported classes 
org.apache.hadoop.mapreduce.task.JobContextImpl and 
org.apache.hadoop.mapreduce.task.TaskAttemptContextImpl in HadoopUtils.java of 
giraph-core and the classes I added. Applicable to other classes that will 
depend on the jar as well. 

[2] I tried activating the profile hadoop-0.23 in the pom file in order to 
import the hadoop-mapreduce-client-core jar automatically, but I could not get 
it to work. Following the instructions on the page 
(http://giraph.apache.org/build.html, I issued the command: "mvn -X 
-Dhadoop=0.23 clean install". 

[3] I'm trying to get my head around JobContextImpl and TaskAttemptContextImpl 
in the methods createJobContext and createTaskAttemptContext respectively of 
the class Hadoop23Shims. The issue here is with the return type. The return 
type is JobContext. JobContextImpl actually implements JobContext interface in 
the hadoop framework. It is important here as I am trying to use Java 
reflection and interface for the different hadoop versions. Advice needed and 
will be greatly appreciated.
                
> Replace munge with shim layer similar to Pig and Hive
> -----------------------------------------------------
>
>                 Key: GIRAPH-101
>                 URL: https://issues.apache.org/jira/browse/GIRAPH-101
>             Project: Giraph
>          Issue Type: Improvement
>            Reporter: Avery Ching
>            Assignee: Michael Aro
>              Labels: gsoc, gsoc2013, mentor
>
> Munge (http://sonatype.github.io/munge-maven-plugin/) is a hacky way of 
> support multiple versions of Hadoop.  The shim layers in Pig and Hive are a 
> cleaner way to do this I think.
> Munge also forces us to release only in source, or to supply different 
> binaries for different hadoop versions.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

Reply via email to