+1 for this feature When running on classical client mode, the driver program stays on Zeppelin server and people need to open network routes so that the Zeppelin server can connect to each of the Spark workers
With the cluster mode, the driver program will be pushed to the cluster itself and it helps solve some networking headaches On Fri, Feb 26, 2016 at 5:57 PM, moon soo Lee <m...@apache.org> wrote: > Hi Jongyoul, > > I think it's very nice to run interpreters on remote machine. That'll make > help scale out the Zeppelin. And the approach in your proposal make sense. > > If i improve the idea little bit more, how about we make common abstraction > for InterpreterProcess and have EmbeddedInterpreterProcess (current local > mode), SeprateInterpreterProcess(current remote mode) as an implementation? > > And then, it becomes more easier to add many different way of managing > interpreter process. > For example, SSHInterpreterProcess (interpreter process that can managed by > ssh command), YarnInterpreterProcess (interpreter that managed by yarn > resource manager) etc. > > What do you think? > > Thanks, > moon > > On Thu, Feb 25, 2016 at 9:56 PM Jongyoul Lee <jongy...@gmail.com> wrote: > > > Hi devs, > > > > I suggest the method of running Zeppelin's interpreters in a cluster > mode. > > Review this proposal and give me comments. I will start from supporting > > standalone cluster. > > > > > > > https://cwiki.apache.org/confluence/display/ZEPPELIN/Cluster+Manager+Proposal > > > > Regards, > > Jongyoul > > > > -- > > 이종열, Jongyoul Lee, 李宗烈 > > http://madeng.net > > >
