Open Stack

----- Original Message -----
> From: "Andrew Bogott" <abogott at wikimedia.org>
> To: "Clint Byrum" <clint at fewbar.com>, "openstack" <openstack at lists.openstack.org>
> Sent: Thursday, September 24, 2015 2:17:29 PM
> Subject: Re: [Openstack] Ironic in the same install with virtualization?
> 
> On 9/16/15 4:03 PM, Clint Byrum wrote:
> > Excerpts from Andrew Bogott's message of 2015-09-16 10:03:48 -0700:
> >> My users are mostly happy with VMs, but I get occasional requests for
> >> physical hardware in order to host databases, run performance tests,
> >> etc.  I'd love to rack a dozen small servers and graft the ironic
> >> service onto my existing cloud in order to fulfill these sporadic
> >> needs.  I'm given pause, though, by this doc section:
> >>
> >> http://docs.openstack.org/developer/ironic/deploy/install-guide.html#configure-compute-to-use-the-bare-metal-service
> >>
> >> For example, this:
> >>
> >>       compute_driver=nova.virt.ironic.IronicDriver
> >>
> >>
> >> That would have me changing nova settings in such a way that would
> >> clearly make nova /only/ support bare-metal, disabling its
> >> virtualization facilities.  Does that mean that bare metal is really an
> >> either/or proposition, or is there some way to make bare metal and virt
> >> services coexist on a single nova installation?  Or is there an
> >> in-between option that involves installing a second nova stack on
> >> different nodes with a different config?
> >>
> >> -- Bonus Question --
> >>
> >> I'm still using nova-network.  Is anyone using ironic without neutron?
> >> Or, if not, is that at least theoretically possible?
> >>
> > Yes, you need to use host aggregates. You would put the ironic compute
> > hosts into an aggregate, and all of the other compute hosts into another,
> > and then have the extra_specs in the flavor set to whichever aggregate
> > you want the instance to go to.
> <long pause>
> 
> OK -- If I understand what you're saying,  I would have a separate node
> running nova-compute that uses the ironic driver, and use host
> aggregates to direct bare-metal allocations to that node.  Is that
> correct?  That would have me running most of my compute services
> (scheduler, conductor, api, compute) with my existing, kvm-oriented
> config, and then having one single outlier compute service running with
> a different ironic-oriented config.  That seems to directly contradict
> this line right up top in the docs:
> 
> " /This configuration file must be modified on the Compute service’s
> controller nodes and compute nodes."[1]
> 
> /Is that sentence simply mistaken?  Or am I misunderstanding what you
> mean about host aggregates?

It's somewhat misleading as not all of the configuration variables listed will have an effect on the control services, in particular the compute_driver setting should only be evaluated by nova-compute. Thus you can have one control plane managing nova-compute nodes that connect to various hypervisors and direct requests to them using either the hypervisor_type image property, host aggregates, or a combination of the two.

Keep in mind that documentation is in the devref which is likely intended at getting someone who wants to submit patches to Ironic up and running quickly not necessarily a guide to recommended production usage.

-Steve

> -Andrew
> /
> /
> [1]
> http://docs.openstack.org/developer/ironic/deploy/install-guide.html#configure-compute-to-use-the-bare-metal-service
> 
> _______________________________________________
> Mailing list: http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
> Post to     : openstack at lists.openstack.org
> Unsubscribe : http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
> 

-- 
Steve Gordon, RHCE
Sr. Technical Product Manager,
Red Hat Enterprise Linux OpenStack Platform