Hi Yan, here are a few thoughts.
+1 for aiming for the clouds (folks used to aim for the stars but nowdays clouds are all the rage...). It's a valid use case which gains popularity fast and glu can certainly fit into the wider picture. The challenge could be maintaining high level of supports for both cloud and non cloud environments.
One thing that I've always wished glu could do, and maybe now is the right time to bring this up, is - run agent-less.
At least theoretically, it's possible, once you have ssh access to a host, to be able to do whatever you need without actually having to install software on it. So the console can, instead of calling an agent, ssh to the host and do what it needs. Of course one way is to install the agent the first time you ssh (I mean the console automatically ssh-s to the host) and continue from there, but that's an implementation detail.
So if operations didn't have to install an agent on every new host, that would make usage of glu a bit easier and I think it's particularly interesting in the cloud case since it's fairly common to start and stop instances regularly.
I understand there are tradeoffs, for example, this make scaling much harder. Still it's my wish...
What could also be nice is maybe somehow run even without zookeeper. Or let glu install ZK by itself if it has to have it. My point is - make the ops life as easy as possible and not require them to invest too much up front.
--
Locked
