Open Stack

Excerpts from Sam Alba's message of 2013-11-18 14:05:47 -0800:
> Hello everyone,
> 
> As some of you already know - in Hong-Kong during the last OpenStack
> Summit - we ran a design session in the Nova topic titled "Docker
> support in OpenStack". The session concluded in developing a new
> OpenStack service for supporting containers instead of modifying Nova
> to support both VMs and containers.
> 
> As said earlier, I am proposing a first draft of the spec for the
> service. I've explicitly CC'ed all people who signed up at the end of
> the design session.
> 
> Here is it: https://etherpad.openstack.org/p/containers-service
> 
> Once we agree on the HTTP API and on the plugin API, the idea is to
> implement a first simple version that would support mono-host. Then it
> would evolve in multi-host really quickly by adding a scheduler and a
> messaging layer (more details in the etherpad).
> 

This is cool and I agree that having containers as a service makes more
sense for a number of reasons.

However, you have not included any concrete use cases in the etherpad
above.

Things without real world use cases tend to devolve quickly into coding
parties IMO. "Oh look at the lovely spec and implementation we are
making." Meanwhile the real users with real use cases are hard to get
right and thus end up going elsewhere because they are "proposing hacks".

Also, I would love to see the first implementation of multi-node take
the gearman route and just have "workers" (compute-equivalent) register
their container capabilities and clients (API servers) push container
jobs into those capability queues.