openstack team mailing list archive

[Bruno Oliveira <brunnop.oliveira@xxxxxxxxx>]

Hello all,

Guys, there's something regarding quantum that we've been hitting our heads
against a wall for a few weeks, but can't figure out. Can you please
take a look?

Here's the scenario:

We currently have a single-node devstack build, running nova (compute), glance,
quantum (with OpenVSwitch), swift and cinder -- besides keystone and
ceilometer.

On that very same machine, our kvm hypervisor is running. Everything
works like a charm with it:  dhcp, magic_ip (169.254.169.254), reaching
in/outside networks, etc.


Now for the issue: there's another host running hyper-v 2008 r2, and on
top of that, we got Cloudbase's (cloudbase.it) Compute Driver for Hyper-V.
That way, we're successfully being able to create VMs for it (Hyper-V), BUT...

Even though there's a virtual switch in there (hyper-v), the network for the
instances are not working -> the instances are not being assigned with an
IP from quantum-dhcp UNLESS I use the FLAT NETWORK model to assign
a public ip for each of them.   So...

1. Is there a must  to have ANOTHER quantum-server just for hyper-v nodes
taking that I do not want to use Flat Network ?
(by another, I mean one besides the working one I have for KVM)

2. In our case, we're intending to have NAT IPs for each of the instance we
have running on hyper-v, not public_ips directly assigned to them, is
it possible?

After all the testing and research we did, we concluded that  we CAN
see the DHCP broadcast coming from hyper-v (via tcpdump'ing on
any interface that is on the same subnet) but the DHCP response/reply
is not working.

Note:  "$quantum agent-list"  DOES list the quantum-agent from kvm
but not from hyper-v

Does anyone have any guesses or suggestions of what we could
possibly try ?


Thank you a lot

--

Bruno Oliveira
Developer, Software Engineer