[Openstack-operators] [Grizzly] Problems with DHCP Agents

Lorin Hochstein lorin at nimbisservices.com
Tue Nov 26 03:43:29 UTC 2013


On Mon, Nov 25, 2013 at 7:27 PM, Samuel Winchenbach <swinchen at gmail.com>wrote:

> Hi All,
>
> So I am experiencing a rather frustrating problem.  I have three nodes
> running quantum dhcp agent (for high availability) and nova-compute.  I
> found that certain VMs do not get an address if all three DHCP agents are
> running.   Here is the output of dhcpdump on the tap interface to one of
> the VMs:
>
> * WITHOUT test2 (worked - using test1 and test3)
> http://pastie.org/pastes/8508325/text
>
>
> * WITH test2 (did not work - using test1, test2, test3)
> http://pastie.org/pastes/8508340/text
>
> The log files look fine, no errors in dnsmasq.log or dhcp-agent.log
>
> Here is what _seems_ to be in common when things do not work:
> * the guest vm is using udhcpc as the DHCP client
> * It is possible this only occurs when the DHCP agent is running on the
> same node as the VM.
>
>
If it's only failing when the DHCP server is on the same host as VM, my
first guess would be the infamous DHCP checksum issue.

Try adding this rule to the DHCP network namespace on each node that has a
DHCP agent (change <uuid> to the appropriate id):

ip netns exec qdhcp-<uuid> iptables -A POSTROUTING -t mangle -p udp --dport
68 -j CHECKSUM --checksum-fill


This is happens if your machine is configured for hardware offloading of
the UDP checksum calculations, and your DHCP packets don't get proper
checksums if they don't cross a physical NIC. Certain DHCP clients barf if
the checksum is invalid.


Lorin
-- 
Lorin Hochstein
Lead Architect - Cloud Services
Nimbis Services, Inc.
www.nimbisservices.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-operators/attachments/20131125/d2782992/attachment.html>


More information about the OpenStack-operators mailing list