Open Stack

On 09/10/2015 12:00 PM, Jeff Peeler wrote:
> On Wed, Sep 9, 2015 at 10:25 PM, Steve Gordon <sgordon at redhat.com
> <mailto:sgordon at redhat.com>> wrote:
>
>     ----- Original Message -----
>     > From: "Jeff Peeler" <jpeeler at redhat.com <mailto:jpeeler at redhat.com>>
>     > To: "OpenStack Development Mailing List (not for usage questions)" <openstack-dev at lists.openstack.org
>     <mailto:openstack-dev at lists.openstack.org>>
>     >
>     > I'd greatly prefer using availability zones/host aggregates as I'm trying
>     > to keep the footprint as small as possible. It does appear that in the
>     > section "configure scheduler to support host aggregates" [1], that I can
>     > configure filtering using just one scheduler (right?). However, perhaps
>     > more importantly, I'm now unsure with the network configuration changes
>     > required for Ironic that deploying normal instances along with baremetal
>     > servers is possible.
>     >
>     > [1]
>     >http://docs.openstack.org/kilo/config-reference/content/section_compute-scheduler.html
>
>     Hi Jeff,
>
>     I assume your need for a second scheduler is spurred by wanting to
>     enable different filters for baremetal vs virt (rather than
>     influencing scheduling using the same filters via image properties,
>     extra specs, and boot parameters (hints)?
>
>     I ask because if not you should be able to use the hypervisor_type
>     image property to ensure that images intended for baremetal are
>     directed there and those intended for kvm etc. are directed to those
>     hypervisors. The documentation [1] doesn't list ironic as a valid
>     value for this property but I looked into the code for this a while
>     ago and it seemed like it should work... Apologies if you had
>     already considered this.
>
>     Thanks,
>
>     Steve
>
>     [1]
>     http://docs.openstack.org/cli-reference/content/chapter_cli-glance-property.html
>
>
> I hadn't considered that, thanks.

Yes, that's the recommended way to direct scheduling requests -- via the 
hypervisor_type image property.

 > It's still unknown to me though if a
> separate compute service is required. And if it is required, how much
> segregation is required to make that work.

Yes, a separate nova-compute worker daemon is required to manage the 
baremetal Ironic nodes.

> Not being a networking guru, I'm also unsure if the Ironic setup
> instructions to use a flat network is a requirement or is just a sample
> of possible configuration.

AFAIK, flat DHCP networking is currently the only supported network 
configuration for Ironic.

 > In a brief out of band conversation I had, it
> does sound like Ironic can be configured to use linuxbridge too, which I
> didn't know was possible.

Well, LinuxBridge vs. OVS isn't really about whether you have a flat 
network topology or not. It's just a different way of doing the actual 
switching (virtual bridging vs. standard linux bridges).

I'm no Neutron expert, but I suspect that one could use either the 
LinuxBridge *or* the OVS ML2 mechanism driver for the L2 agent, along 
with a single flat provider network for your baremetal nodes.

Hopefully an Ironic + Neutron expert will confirm or deny this?

Best,
-jay