On Sat, Feb 15, 2014 at 12:57 PM, Alex Huang <alex.hu...@citrix.com> wrote: > I see this as that CS management server should run anywhere java runs. > However, I see quite a few holes in this proposal. > > - Management server should never have mounted the NFS secondary storage. We > need to fix that. Please have a talk with Kelven about this. I don't see us > adding Samba to any machine that runs the management server. If this is just > about seeding the system template, it can be done as part of installation > rather than management server. > - If I truly think about Windowsification, I think windows installer, windows > service, opening ports in windows firewall. I don't see any of that here. > Or else it's useless for windows people. > - For java, most people write a windows service that launches the jvm > and keeps track of the jvm. > - There's a number of things that CloudStack management server does today at > startup that should not be there and those are what complicates the > windowfication. These things should be moved out of the management server > and into installation. What should be inside the management server startup > procedure should be checks, rather than generations. It should check for ssh > key exists and the database version matches etc but it should not do the > following. > - Upgrade of the database > - ssh key generation > - iso generation. > > --Alex
A couple of more to add to the list. There isn't really a good binary software dependency resolution framework for Windows (at least not that I am aware of) and you potentially need lots of different things - mysql libraries, python, ipmitool, or another shell to execute shell scripts. Look at the list of the dependencies we install in the RPMs for an idea of things that are missing. Is the plan to bundle those in the binaries? (If so, we need to be having a serious discussion around a shift of a number system requirements to dependencies which may or may not be acceptable.) With what appears to be some serious refactoring in how the management server works that Alex outlines and I doubt this is a short term goal. (e.g. 4.4 timeframe). If you are going to take on all of the refactoring work, please start by building tests that prove that it all works today and then verify the same behavior in the refactored work. --David
