[Rdo-list] 10 days of rdo manager

[Dan Sneddon]

On 04/20/2015 04:58 PM, James Slagle wrote:
> On Mon, Apr 20, 2015 at 03:48:51PM -0700, Dan Sneddon wrote:
>> On 04/20/2015 03:42 PM, James Slagle wrote:
>>> On Mon, Apr 20, 2015 at 06:49:27PM +0100, Pedro Sousa wrote:
>>>> Hi,
>>>> 
>>>> I haven't played with it yet, but I would like to know what
>>>> happens to overcloud nodes when you reboot/loose or break
>>>> your undercloud node.
>>>> 
>>>> Do overcloud nodes loose ip connectivity? My understanding
>>>> is that overcloud nodes get dhcp from Neutron. Or do I need
>>>> to have some HA for undercloud in place?
>>> 
>>> With the network architecture we're moving towards, overcloud
>>> nodes will only get dhcp from Neutron for the provisioning
>>> network. The api, data, storage, etc network will support
>>> static IP configuration, or possibly, non-Neutron provided 
>>> dhcp.
>>> 
>>> Further, after initial provisioning, overcloud nodes will boot
>>> off the local disk instead of pxe booting via Neutron on
>>> subsequent reboots. localboot support is a relatively new
>>> feature in upstream Ironic, and we'll be enabling it soon in
>>> rdo-manager.
>>> 
>>> With these changes, when the undercloud is stopped or goes
>>> down unexpectedly the overcloud would be unaffected. That
>>> being said, we still plan to have an HA undercloud at some
>>> point in the future.
>>> 
>>> Also, the current virt-setup that allows testing rdo-manager
>>> via deploying the undercloud and overcloud all on vm's still
>>> relies on the undercloud vm to continue to run for
>>> connectivity to overcloud nodes. That could also be enhanced
>>> though to not require the undercloud vm to stay up.
>>> 
>>> 
>> 
>> When a node is configured for local boot, does it get a static
>> IP address? Maybe it turns the DHCP address into a static? Or
>> does it still rely on the undercloud for DHCP?
> 
> The interface connected to the provisioning network will still get
> dhcp from neutron on the undercloud if that interface is
> configured to start on boot, and configured to try dhcp. Since
> we're currently including the dhcp-all-interfaces[1] element in
> our image builds, that will indeed be the case.
> 
> If not, it wouldn't get any IP address. It doesn't seem like you'd
> want to configure it with a static IP.
> 
> [1]
> https://github.com/openstack/diskimage-builder/tree/master/elements/dhcp-all-interfaces
>
> 
> Note we've had a few requests to not default to including this
> element, or at least it make it configurable not to include it.
> 
> -- -- James Slagle --
> 

I was just confirming that the undercloud is still a single point of
failure for the provisioning interface, even with localboot. With
that in mind, it should be possible to configure static IP addresses
on the other interfaces. The OpenStack services would be configured
to use the interfaces with static IPs to ensure that the undercloud
isn't required for ongoing operation of the overcloud.

The pieces to configure static IP addresses are just landing. We will
also need the accompanying puppet logic to place the services onto
the proper interfaces. That will follow soon.

We should definitely remove (or have an option to disable)
dhcp-all-interfaces. This could lead to unexpected behavior if an
otherwise unconfigured interface is plugged in to the wrong network,
and could even be a security risk. Without dhcp-all-interfaces we can
still specify multiple DHCP interfaces in the net-config-*.yaml files
if that behavior is desired.

-- 
Dan Sneddon         |  Principal OpenStack Engineer
dsneddon at redhat.com |  redhat.com/openstack
650.254.4025        |  dsneddon:irc   @dxs:twitter