Open Stack

Hello to All.

See inline comments.

Kind regards,
Denys Makogon

2016-05-24 23:55 GMT+03:00 Hongbin Lu <hongbin.lu at huawei.com>:

> Hi all,
>
>
>
> At the last team meeting, we tried to define the scope of the Higgins
> project. In general, we agreed to focus on the following features as an
> initial start:
>
> ·         Build a container abstraction and use docker as the first
> implementation.
>
> ·         Focus on basic container operations (i.e. CRUD), and leave
> advanced operations (i.e. keep container alive, rolling upgrade, etc.) to
> users or other projects/services.
>
> ·         Start with non-nested container use cases (e.g. containers on
> physical hosts), and revisit nested container use cases (e.g. containers on
> VMs) later.
>
> The items below needs further discussion so I started this ML to discuss
> it.
>
> 1.       Container composition: implement a docker compose like feature
>

In Docker-compose, at this point of time i'm working to extracting core
functionality into something similar to libcompose (written on Go) but with
Python API.
I can tell that it is not that fast, so that work would take some time
(couple releases). My suggestion is to implement abstraction layer that
will consume your own implementation of compose features and once
docker-compose will be ready to be consumed then in Higgins we will switch
to it.

Another thing, it is worth considering to use TOSCA modeling (see how
Tacker is doing it) for container orchestration.


> 2.       Container host management: abstract container host
>
> For #1, it seems we broadly agreed that this is a useful feature. The
> argument is where this feature belongs to. Some people think this feature
> belongs to other projects, such as Heat, and others think it belongs to
> Higgins so we should implement it. For #2, we were mainly debating two
> things: where the container hosts come from (provisioned by Nova or
> provided by operators); should we expose host management APIs to end-users?
> Thoughts?
>

Here's what i think, if we would take a look at Solum that uses swarm
cluster API endpoint that defined in its config, so for me, as for operator
it is not that useful.

As first step, we can live with that, but when you would think of multisite
OpenStack containers orchestration that case wouldn't work at all. As
proposal, i'd like to see special DB model that represents swarm cluster
entity (all necessary creds to connect to it: TLS certs, user/password,
endpoint, etc.) and provide advanced placement algorithm that would help to
define where should container lay or let users to pick concrete swarm
cluster to deploy their container.


>
>
> Best regards,
>
> Hongbin
>
> __________________________________________________________________________
> OpenStack Development Mailing List (not for usage questions)
> Unsubscribe: OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20160525/46bbfc5a/attachment.html>