[openstack-dev] [Openstack] quantum network node in KVM guest - connectivity issues

Jay Pipes jaypipes at gmail.com
Sat Oct 12 14:31:06 UTC 2013


On Sat, Oct 12, 2013 at 8:59 AM, Nick Maslov <azpekt at gmail.com> wrote:

> Hi,
>
> I have following setup:
>
> 1) infrastructure node, IP in bond, hosting following KVM guests:
> 1.1) Postgres KVM guest
> 1.2) MQ KVM guest
> 1.3) DNS KVM guest
> 1.4) Control node with Nova API, Cinder API, Quantum Server, etc.
> ...
> 1.8) Quantum network node with quantum agents
>
> Agents on this network node are always dying and starting up again:
>
> # quantum agent-list
>
> +--------------------------------------+--------------------+-----------------------------+-------+----------------+
> | id                                   | agent_type         | host
>                | alive | admin_state_up |
>
> +--------------------------------------+--------------------+-----------------------------+-------+----------------+
> | 5656392b-b6fe-4570-802f-97d2154acf31 | L3 agent           |
> net01-001.int.net.net | xxx   | True           |
> | 1093fb73-6622-448e-8dad-558a36cca306 | DHCP agent         |
> net01-001.int.net.net | xxx   | True           |
> | 4518830d-e112-439f-a629-7defa7bd29e9 | Open vSwitch agent |
> net01-001.int.net.net | xxx   | True           |
> | 86ee6d24-2e6a-4f58-addb-290fefc26401 | Open vSwitch agent | nova05
>                | :-)   | True           |
> | b67697bb-3ec1-49fc-8f3c-7e4e7892e83a | Open vSwitch agent | nova04
>                | :-)   | True           |
>
> +--------------------------------------+--------------------+-----------------------------+-------+----------------+
>
> Few minutes after, those agents will be up again, one may die - while
> others not.
>
> ping net01-001
> PING net01-001.int.net.net (10.10.146.34) 56(84) bytes of data.
> 64 bytes from net01-001.int.net.net (10.10.146.34): icmp_req=1 ttl=64
> time=0.912 ms
> 64 bytes from net01-001.int.net.net (10.10.146.34): icmp_req=2 ttl=64
> time=0.273 ms
> 64 bytes from net01-001.int.net.net (10.10.146.34): icmp_req=2 ttl=64
> time=0.319 ms (DUP!)
> 64 bytes from net01-001.int.net.net (10.10.146.34): icmp_req=3 ttl=64
> time=0.190 ms
> 64 bytes from net01-001.int.net.net (10.10.146.34): icmp_req=4 ttl=64
> time=0.230 ms
> 64 bytes from net01-001.int.net.net (10.10.146.34): icmp_req=4 ttl=64
> time=0.305 ms (DUP!)
> 64 bytes from net01-001.int.net.net (10.10.146.34): icmp_req=5 ttl=64
> time=0.199 ms
> 64 bytes from net01-001.int.net.net (10.10.146.34): icmp_req=7 ttl=64
> time=0.211 ms
> 64 bytes from net01-001.int.net.net (10.10.146.34): icmp_req=8 ttl=64
> time=0.322 ms
> 64 bytes from net01-001.int.net.net (10.10.146.34): icmp_req=8 ttl=64
> time=0.409 ms (DUP!)
> ^C
> --- net01-001.int.net.net ping statistics ---
> 8 packets transmitted, 7 received, +3 duplicates, 12% packet loss, time
> 7017ms
>
> SSH`ing to network node is also difficult - constant freezes. Nothing
> suspicious in the logs.
>

Those DUP!'s are suspicious, since you aren't pinging a broadcast domain.
That might indicate there's something up with the OVS GRE mesh.


> Since DHCP agent may be down, spawning a VM may end in "waiting for
> network device" state. Then, it might get the internal IP and then floating
> - but accessing it also proves to be very troublesome - I believe because
> of L3 agent flapping.
>
> My OpenStack was set up under this manual -
> https://github.com/mseknibilel/OpenStack-Grizzly-Install-Guide/blob/OVS_MultiNode/OpenStack_Grizzly_Install_Guide.rst
>
> Only thing I did - I added HAproxy/keepalived on top of it, balancing API
>> requests on control nodes. But this shouldn`t impact networking...
>>
>
Agreed, it should not affect network connectivity for the network node.

Not sure what the issue is. Perhaps you might try following through
Darragh's excellent tutorial on debugging L3 issues in OVS/Quantum here:

http://techbackground.blogspot.com/2013/05/the-quantum-l3-router-and-floating-ips.html

DId you manually set up KVM instances for all these nodes, or are you using
something like Triple-O?

Best,
-jay


>
> Anyone have any thoughts about this?
>
> Cheers,
> NM
>
> _______________________________________________
> Mailing list:
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
> Post to     : openstack at lists.openstack.org
> Unsubscribe :
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20131012/94604045/attachment.html>


More information about the OpenStack-dev mailing list