I think Spark has done this right. It can run by itself, and run on YARN or Mesos if needed. We should head in this direction.
The first step is to enable a self-contained mode. In this case, lightweight lib + zookeeper is still most preferable. Helix and Curator are two good candidates here. Integrate with Mesos or YARN will be on future road map. On Tue, Nov 3, 2015 at 11:14 PM, Luke Han <[email protected]> wrote: > Would like to echo Yang's comments to keep dependency as minimal as > possible. > A self-contain one will let people easy to setup and manage Kylin instance > as essential. > > And then we could extend a little bit to support other cluster management > frameworks, > the key is how we could design the architecture to more plug-able way:) > > Thanks. > > > Best Regards! > --------------------- > > Luke Han > > On Tue, Nov 3, 2015 at 11:13 PM, Tero Paananen <[email protected]> > wrote: > > > Whatever the plan ends up being, please don't make it so that users > > will have to then manage the cluster manager. > > > > -TPP > > >
