Re: [atomic-devel] Multi-node deployment strategies

[Clayton Coleman <ccoleman redhat com>]

----- Original Message -----
> 
> On 11/10/2014 06:01 PM, Colin Walters wrote:
> > Let's fast forward to a few weeks from now and say that we have the basics
> > of Atomic delivery sorted out.  I'd like to get some sort of documentation
> > (and continuous testing) around multi-node Atomic+Kubernetes deployments.
> >
> > A good example here is: https://github.com/eparis/kubernetes-ansible
> >
> > I was also looking at the Kubernetes Vagrant example - which also uses
> > Fedora, but provisions with Salt, and deploys a k8s binary you built on
> > the host.
> >
> > The interesting thing here is networking.  The eparis-k8s-ansible model has
> > you boot the hosts and record their IPs.  The Vagrant one has a set of
> > hardcoded IP addresses and uses the Vagrant network control to give those
> > addresses to the guests, as well as the /24 for the k8s pods.
> >
> > Hmm, actually this post has the potential to be far too long.  I think
> > there's going to be some divergence between dev and prod here, and I'm
> > just thinking about the dev case right now.  "How do I test my k8s
> > pod-ified app".
> >
> > So maybe focus on:
> >   - Using Atomic + Vagrant to set up a local cluster
> >   - Using Atomic on OpenStack/EC2/GCE as a dev environment
> > ?
> >
> 
> 
> Hey Colin,
>      For OpenShift v3 (which is built on top of Kubernetes, so should be
> similar), we're using Ansible and it's dynamic inventory features.
> 
> Our repository (has more than just the openshift v3 stuff):
> https://github.com/openshift/openshift-online-ansible
> 
> More about Ansible dynamic inventories:
> http://docs.ansible.com/intro_dynamic_inventory.html
> 

Eric, any reason we can't use dynamic inventory in the upstream ansible?  I'd like to gradually bring Online and the Kube ones into a fairly standard pattern - there has to be some way for people to deploy Origin on top of Kube as well in some cases.