It's going to be tough, because we actually planned to have the release out already and none of us really is a YARN expert. If there's some common trap we're falling into with how we're specifying paths or setting up jobs and we can fix it easily, I'm guessing no one would object. Of course, we'd have to stay compatible with Hadoop 0.20/0.23.
Unfortunately, our own Jenkins tasks that run clustering examples aren't passing right now, either, so I'm not even sure whether the problem is new to Hadoop 2. Is the test you're running just a script based on this wiki page? https://cwiki.apache.org/confluence/display/MAHOUT/Clustering+of+synthetic+control+data -tom On 06/06/2012 05:14 PM, Robin Anil wrote: > Roman. I don't think there are many yarn experts here. Why don't you be the > one. Test out the example scripts on hadoop with yarn and let us know. If > its a trivial fix we can submit that. If its not then we cannot. There is > only one way to know. You. > On Jun 6, 2012 10:22 PM, "Roman Shaposhnik" <[email protected]> wrote: > >> On Wed, Jun 6, 2012 at 4:46 PM, Paritosh Ranjan <[email protected]> wrote: >>> Can I start working on the release now? >> Just an FYI (and a plea ;-)): Bigtop is readying to release the first >> Apache >> Hadoop distribution based on Hadoop 2.0.0-alpha. It would be really >> nice if we can provide our users with the Mahout 0.7.0 working on top >> of YARN out of the box. Please let me know if I can help with testing >> such a combination. I've already filed a couple of minor JIRAs: >> https://issues.apache.org/jira/browse/MAHOUT-1017 >> https://issues.apache.org/jira/browse/MAHOUT-1016 >> but haven't seen any activity on them. >> >> Thanks, >> Roman. >>
