I guess the Hadoop2 cluster is not the problem. I'm sure I've set one up properly, since I was able to run a few Hadoop2 examples using that cluster. The problem seems to be Giraph. The examples provided by Giraph didn't work at all (ClassNotFoundExceptions). I added the Giraph jars to any classpath environment variable and even copied the jars into Hadoop's lib/ directory (on the master and on the slaves). Nothing helped to get rid of the ClassNotFoundExceptions.
Using Giraph-Gremlin seemed to have a different issue. According to the Hadoop2/YARN logs, containers were killed with the exit code 143. That actually means that we have some memory issues, but even a g.V().count() over the 6 node modern graph was killed with the same exit code. I'm far away from being a Hadoop/Giraph expert, hence everything I tried was more or less just trial & error; I never really knew what I'm doing and just blindly followed instructions that I found in blog post or mailing list discussions. With that said, it would be really cool, if someone who knows what he's doing, when it comes to Hadoop2+Giraph, would raise his hand. Cheers, Daniel On Thu, Nov 12, 2015 at 9:42 PM, Marko Rodriguez <[email protected]> wrote: > Hello, > > All of our testing of GiraphGraphComputer has occurred on a Hadoop2 pseudo > cluster. Has anyone tested TinkerPop 3.1.0-SNAPSHOT on a Hadoop2 cluster > (multi-machine) OR has a Hadoop2 cluster they can run a few jobs on to make > sure that jobs are submitting as expected. > > Thanks, > Marko. > > http://markorodriguez.com > >
