This is a great idea. It will require some serious contemplation to get something good. I am working on series of patches moving towards a "pure YARN" implementation. If I don't go the route of a maven profile and/or munge flags then this will be essential. It would be great to get everyone's ideas on this as I can try to follow a path that does not conflict with the ideas presented here. I'll definitely be keeping up on this discussion as it moves along.
On Sat, Jan 19, 2013 at 3:56 PM, Eugene Koontz (JIRA) <[email protected]>wrote: > > [ > https://issues.apache.org/jira/browse/GIRAPH-485?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13558128#comment-13558128] > > Eugene Koontz commented on GIRAPH-485: > -------------------------------------- > > Sounds interesting; let's look into it. > > > Hadoop Shims, smart detect version on job start > > ----------------------------------------------- > > > > Key: GIRAPH-485 > > URL: https://issues.apache.org/jira/browse/GIRAPH-485 > > Project: Giraph > > Issue Type: Wish > > Reporter: Nitay Joffe > > > > We should be able to (in theory at least, haven't looked deeply into it > yet) use hive style shims to get rid of our munge whatnots and compile > against all of the hadoops at once. This would mean having a common > interface and a module for each hadoop we support. Then, when the user > launches a giraph job we can detect the hadoop we're running in and load > the appropriate module. > > This will require some exploration and research to figure out. > > -- > 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 >
