[Openstack] Havana / nova-network - Multi-Node setup, dnsmasq usesthe same IP on multiple nodes

Sascha Vogt sascha.vogt at gmail.com
Fri Jan 24 14:55:42 UTC 2014


Hi Ritesh,

I was missing the share_dhcp_address=True option in nova.conf, setting
that to False solves the issue.

The network was specified as multi-host T and the bridge IS created
automatically. The issue was the IP which was assigned to the bridge. It
was the same on all nodes (which is bad, as the bridge is connected
through tunnels)

Greetings
-Sascha-

Am 24.01.2014 15:41, schrieb Ritesh nanda:
> Hello Sascha,
> 
> Its an multi host nova-network setup depicted there, so nova-network
> will handle network of its compute node.
> 
> While creating a network specify multi-host T , in that case when a
> machine is created on a compute node bridge would be automatically
> created on each compute node.
> 
> Regards,
> Ritesh nanda
> ------------------------------------------------------------------------
> From: Sascha Vogt <mailto:sascha.vogt at gmail.com>
> Sent: ‎24-‎01-‎2014 07:06 PM
> To: openstack at lists.openstack.org <mailto:openstack at lists.openstack.org>
> Subject: [Openstack] Havana / nova-network - Multi-Node setup, dnsmasq
> usesthe same IP on multiple nodes
> 
> Hi all,
> 
> I have a Multi-Node, Single-NIC setup. All machines only have a single
> NIC. I created a virtual network (using gretap tunnels - aka
> layer2-over-layer3 tunnel) to connect all machines and have one br-int
> bridge which all VMs are attached to.
> 
> nova-network runs on all machines and correctly binds dnsmasq to the
> hosts bridge itself, though I noticed that each host-bridge gets the .1
> IP. This seems to work, because dnsmasq is configured by nova-network to
> only answer to DHCP requests the specific instance has a MAC address
> for, though I find it a bit irritating.
> 
> I try to give a picture of it:
> 
> controller
>     - br-int (dnsmasq with .1 address)
>         - gretap tunnel to compute-1 (using the static IPs of eth0)
>         - gretap tunnel to compute-2 (using the static IPs of eth0)
>         - vnet1-n (instances running on this host)
>     - eth0 (routes between external network and br-int, NAT / ip
>             forwarding active, static IP used also for OpenStack
>             managing)
> 
> compute-1
>     - br-int (dnsmasq with .1 address)
>         - gretap tunnel to controller (using the static IPs of eth0)
>         - vnet1-n (instances running on this host)
>     - eth0 (OpenStack managing)
> 
> compute-2
>     - br-int (dnsmasq with .1 address)
>         - gretap tunnel to controller (using the static IPs of eth0)
>         - vnet1-n (instances running on this host)
>     - eth0 (OpenStack managing)
> 
> I'm using the FlatDHCPManager, and if you substitue eth0/the-switch in
> this picture
> http://www.mirantis.com/wp-content/uploads/2012/07/flat-dhcp-networking-diagrams-4.png
> with the gretap tunnels I basically have that topology. In that picture
> the dnsmasqs/br100 have different IPs. How did they get that? ;)
> 
> Greetings
> -Sascha-





More information about the Openstack mailing list