Typical cloud deployments – be it OpenStack, CloudStack, Eucalyptus etc – have a separate control layer installed and upgraded using separate tools (which might be hand-configured PXE + preseeding, Cobbler, Orchestra/MAAS) As a result you have two distinct provisioning systems in play, which allows for more user error and increased special cases in automation. Add to this the increasing complexity of the puppet or chef descriptions of such setups, and you can soon go mad and be stymied by fear of change.
Contrast that with the new generation easy things we can do in the cloud. Upgrade? Why bother – just spin up the new version of your app in parallel, then let your HA system fail over to it, then de-provision the old version.
So what if we could get all of the power that we have in the cloud when we’re running our datacenters, without the cost overhead of virtualization?
Starting with a bare-metal driver for OpenStack Compute (nova) and coupling that with OpenStack Orchestration (heat) and a pipeline for building cloud-style non-installer based disk images, we can do just that. Provision and manage your entire data center with the flexibility of a managed virtual cloud and all the power of running on actual metal.
The best part? It’s all open source and works!