[Openstack-mentoring] Neutron subnet with DHCP relay

Thomas King thomas.king at gmail.com
Tue Jun 2 00:47:42 UTC 2020


Well, I'm using a flat network (i.e., access mode switchport), not a tagged
switchport. The remote nodes will be on their own subnet and VLAN/segment
that will *not *be attached to the controller.

"In that case, DHCP requests from all segments that are not attached to
the controller(s) need to be forwarded to the controllers via DHCP relay."
Agreed. The question is which DHCP agent IP address?

DHCP agent IP #1 = 10.10.1.1/16, corresponds to an attached segment/subnet.
DHCP agent IP #2 = 10.100.1.1/16, corresponds to a subnet in a completely
separate rack with no direct connection to the controller.

Even if I separate IP #2 onto a different segment with its own DHCP agent
port, am I sending DHCP relay to IP #1 or #2? I'm assuming #1.

Tom

On Mon, Jun 1, 2020 at 6:40 PM <dsneddon at redhat.com> wrote:

> The use case for routed networks is when you have multiple distinct
> subnets which are not connected at layer 2 and only have connectivity
> to one another via the router gateways on each network. A segment can
> be thought of as a VLAN, although depending on topology a different
> VLAN ID is not always used. The key is that there is no layer 2
> connectivity between segments, traffic has to be routed between them.
>
> The situation where you would use DHCP relay is when you are not
> assigning DHCP agents to the compute nodes, and you have compute nodes
> on segments that the controllers are not attached to. In that case,
> DHCP requests from all segments that are not attached to the
> controller(s) need to be forwarded to the controllers via DHCP relay.
>
> If you have a flat network, then you have no need for DHCP relay, the
> DHCP agents can receive and respond to requests over layer 2. This
> applies even if you have multiple subnets on the same segment.
>
> On Mon, 2020-06-01 at 18:02 -0600, Thomas King wrote:
> > We do have the Ironic inspector enabled but mainly use out-of-band
> > such as iDRAC.
> >
> > I am indeed not using segments. I'll need to research that a bit
> > more.
> >
> > One important note, we are only using provider networks with no
> > Neutron routers. All routing is done on the physical network which
> > aligns with the docs for segments. The provisioning subnet is on
> > 10.10.0.0/16 for the directly attached nodes. As a test, I created a
> > second subnet, 10.100.0.0/16, on the same Neutron network with DHCP
> > enabled, so now I have two subnets on the same network and Neutron
> > DHCP port. However, if DHCP relay requires different segments per
> > remote network...
> > > The Networking service defines a segment using the following
> > > components:
> > >
> > > Unique physical network name
> > > Segmentation type
> > > Segmentation ID
> >
> > Does having unique physical network names also mean unique physical
> > interfaces?
> > Does this mean no flat network for segments?
> > If I create 10.100.0.0/16 in rack A1 and the controller is in D30, am
> > I pointing the DHCP relay to the DHCP agent's 10.10.0.0/16 IP
> > address?
> >
> >
> > +--------------------+-------+---------------------------+
> > | Agent Type         | Alive | Binary                    |
> > +--------------------+-------+---------------------------+
> > | Baremetal Node     | :-)   | ironic-neutron-agent      |
> > | Baremetal Node     | :-)   | ironic-neutron-agent      |
> > | Baremetal Node     | :-)   | ironic-neutron-agent      |
> > | Baremetal Node     | :-)   | ironic-neutron-agent      |
> > | Linux bridge agent | :-)   | neutron-linuxbridge-agent |
> > | Baremetal Node     | :-)   | ironic-neutron-agent      |
> > | Metering agent     | :-)   | neutron-metering-agent    |
> > | Baremetal Node     | :-)   | ironic-neutron-agent      |
> > | Baremetal Node     | :-)   | ironic-neutron-agent      |
> > | Baremetal Node     | :-)   | ironic-neutron-agent      |
> > | DHCP agent         | :-)   | neutron-dhcp-agent        |
> > | L3 agent           | :-)   | neutron-l3-agent          |
> > | Baremetal Node     | :-)   | ironic-neutron-agent      |
> > | Baremetal Node     | :-)   | ironic-neutron-agent      |
> > | Metadata agent     | :-)   | neutron-metadata-agent    |
> > | Baremetal Node     | :-)   | ironic-neutron-agent      |
> > | Baremetal Node     | :-)   | ironic-neutron-agent      |
> > | Baremetal Node     | :-)   | ironic-neutron-agent      |
> > +--------------------+-------+---------------------------+
> >
> > On Mon, Jun 1, 2020 at 4:34 PM <dsneddon at redhat.com> wrote:
> > > You will have to target two IP addresses with DHCP relay if you are
> > > using Ironic Inspector. The first is the IP where Ironic Inspector
> > > is
> > > listening with dnsmasq, usually the IP of the host itself. I know
> > > this
> > > doesn't lend itself to HA scenarios, but you might also be able to
> > > forward to the broadcast IP of the subnet where the Ironic
> > > Inspector
> > > will be running (I haven't tested this, but it is a common use case
> > > for
> > > DHCP relay).
> > >
> > > The second IP address is that of the Neutron DHCP agent, and that
> > > will
> > > be used for deploying bare metal nodes. IIRC, this IP is shared
> > > with
> > > the Neutron router for the network if you are using the L3 agent as
> > > well.
> > >
> > > If you are not running Ironic Inspector (and manually entering in
> > > baremetal host details instead), then you can forward DHCP relay
> > > only
> > > to the Neutron DHCP agent.
> > >
> > > Both of these IP addresses will be on the "root" subnet which is
> > > associated with the segment with the controller node(s).
> > >
> > > It sounds like you created a second subnet, but I'm not sure if you
> > > created the second subnet on a different segment from the first
> > > subnet.
> > > In Neutron routed networking, the segments determine whether a
> > > subnet
> > > is local or remote to the controller node(s). Typically the first
> > > segment would be the one local to the controller(s). Are you sure
> > > you
> > > enabled the segments plugin and created your second subnet on a new
> > > segment?
> > >
> > > Another approach which does not involve DHCP relay is to deploy
> > > DHCP
> > > agents locally on compute nodes local to each segment. This way all
> > > DHCP will be done within the same L2 domain, and you will not have
> > > to
> > > configure DHCP relay on your router serving each segment/subnet.
> > >
> > > See the docs for more info:
> > >
> https://docs.openstack.org/newton/networking-guide/config-routed-networks.html
> > >
> > > -Dan
> > >
> > > On Fri, 2020-05-29 at 10:47 -0600, Thomas King wrote:
> > > > In the Triple-O docs for unicast DHCP relay, it doesn't exactly
> > > say
> > > > which IP address to target. Without deploying Triple-O, I'm not
> > > clear
> > > > if the relay IP should be the bridge interface or the DHCP
> > > device.
> > > >
> > > > The first method makes sense because the gateway for that subnet
> > > > wouldn't be connected to the Ironic controller by layer 2 (unless
> > > we
> > > > used VXLAN over the physical network).
> > > >
> > > > As an experiment, I created a second subnet on my provisioning
> > > > network. The original DHCP device port now has two IP addresses,
> > > one
> > > > on each subnet. That makes the second method possible if I
> > > targeted
> > > > its original IP address.
> > > >
> > > > Thanks for the help and please let me know which method is
> > > correct.
> > > >
> > > > Tom King
> > > >
> > > > On Fri, May 29, 2020 at 3:15 AM Dan Sneddon <dsneddon at redhat.com>
> > > > wrote:
> > > > > You probably want to enable Neutron segments and use the
> > > Neutron
> > > > > routed networks feature so you can use different subnets on
> > > > > different segments (layer 2 domains AKA VLANs) of the same
> > > network.
> > > > > You specify different values such as IP allocation pools and
> > > router
> > > > > address(es) for each subnet, and Ironic and Neutron will do the
> > > > > right thing. You need to enable segments in the Neutron
> > > > > configuration and restart the Neutron server.  I don’t think
> > > you
> > > > > will have to recreate the network. Behind the scenes, dnsmasq
> > > will
> > > > > be configured with multiple subnets and address scopes within
> > > the
> > > > > Neutron DHCP agent and the Ironic Inspector agent.
> > > > >
> > > > > Each segment/subnet will be given a different VLAN ID. As
> > > Dmitry
> > > > > mentioned, TripleO uses that method for the provisioning
> > > network,
> > > > > so you can use that as an example. The provisioning network in
> > > > > TripleO is the one referred to as the “control plane” network.
> > > > >
> > > > > -Dan
> > > > >
> > > > > On Fri, May 29, 2020 at 12:51 AM Dmitry Tantsur <
> > > > > dtantsur at redhat.com> wrote:
> > > > > > Hi Tom,
> > > > > >
> > > > > > I know for sure that people are using DHCP relay with ironic,
> > > I
> > > > > > think the TripleO documentation may give you some hints
> > > (adjusted
> > > > > > to your presumably non-TripleO environment):
> > > > > >
> > >
> http://tripleo.org/install/advanced_deployment/routed_spine_leaf_network.html#dhcp-relay-configuration
> > > > > >
> > > > > > Dmitry
> > > > > >
> > > > > > On Thu, May 28, 2020 at 11:06 PM Amy Marrich <amy at demarco.com
> > > >
> > > > > > wrote:
> > > > > > > Hey Tom,
> > > > > > >
> > > > > > > Forwarding to the OpenStack discuss list where you might
> > > get
> > > > > > > more assistance.
> > > > > > >
> > > > > > > Thanks,
> > > > > > >
> > > > > > > Amy (spotz)
> > > > > > >
> > > > > > > On Thu, May 28, 2020 at 3:32 PM Thomas King <
> > > > > > > thomas.king at gmail.com> wrote:
> > > > > > > > Good day,
> > > > > > > >
> > > > > > > > We have Ironic running and connected via VLANs to nearby
> > > > > > > > machines. We want to extend this to other parts of our
> > > > > > > > product development lab without extending VLANs.
> > > > > > > >
> > > > > > > > Using DHCP relay, we would point to a single IP address
> > > to
> > > > > > > > serve DHCP requests but I'm not entirely sure of the
> > > Neutron
> > > > > > > > network/subnet configuration, nor which IP address should
> > > be
> > > > > > > > used for the relay agent on the switch.
> > > > > > > >
> > > > > > > > Is DHCP relay supported by Neutron?
> > > > > > > >
> > > > > > > > My guess is to add a subnet in the provisioning network
> > > and
> > > > > > > > point the relay agent to the linuxbridge interface's IP:
> > > > > > > > 14: brq467f6775-be: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu
> > > > > > > > 1500 qdisc noqueue state UP group default qlen 1000
> > > > > > > >     link/ether e2:e9:09:7f:89:0b brd ff:ff:ff:ff:ff:ff
> > > > > > > >     inet 10.10.0.1/16 scope global brq467f6775-be
> > > > > > > >        valid_lft forever preferred_lft forever
> > > > > > > >     inet6 fe80::5400:52ff:fe85:d33d/64 scope link
> > > > > > > >        valid_lft forever preferred_lft forever
> > > > > > > >
> > > > > > > > Thank you,
> > > > > > > > Tom King
> > > > > > > > _______________________________________________
> > > > > > > > openstack-mentoring mailing list
> > > > > > > > openstack-mentoring at lists.openstack.org
> > > > > > > >
> > >
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-mentoring
> > > > >
> > > > > --
> > > > > Dan Sneddon         |  Senior Principal Software Engineer
> > > > > dsneddon at redhat.com |  redhat.com/cloud
> > > > > dsneddon:irc        |  @dxs:twitter
> --
> Dan Sneddon         |  Senior Principal Software Engineer
> dsneddon at redhat.com |  redhat.com/cloud
> dsneddon:irc        |  @dxs:twitter
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-discuss/attachments/20200601/9c1640ef/attachment-0001.html>


More information about the openstack-discuss mailing list