[Rdo-list] Enabling 'ovs_use_veth' for DHCP/L3 agents breaks Neutron on recent versions of RHEL/CentOS
8:04 a.m.
Hi all,
I was asked to take a look at this bug for the openstack-manuals project - "Enabling
'ovs_use_veth' for DHCP/L3 agents breaks Neutron on recent versions of
CentOS":
https://bugs.launchpad.net/openstack-manuals/+bug/1268806
"After assisting someone with troubleshooting networking, I determined that setting
'ovs_use_veth = True' in dhcp_agent.ini and l3_agent.ini breaks Neutron (at least
with GRE) on recent versions of CentOS... perhaps because the kernel seems to fully
support namespaces. This issue probably also affects Scientific Linux. Can anyone confirm
whether recent RHEL kernels fully support namespaces? I would like to clarify this step
and similar steps in other sections for all distributions."
Questions I have are:
a) Is it likely/expected that enabling ovs_use_veth = True will cause issues when using a
kernel that supports netns?
b) Is there any downside at this point to setting ovs_use_veth = False?
Thanks,
Steve
6:58 a.m.
On 01/14/2014 02:04 PM, Steve Gordon wrote:
Hi all,
I was asked to take a look at this bug for the openstack-manuals project - "Enabling
'ovs_use_veth' for DHCP/L3 agents breaks Neutron on recent versions of
CentOS":
https://bugs.launchpad.net/openstack-manuals/+bug/1268806
"After assisting someone with troubleshooting networking, I determined that setting
'ovs_use_veth = True' in dhcp_agent.ini and l3_agent.ini breaks Neutron (at least
with GRE) on recent versions of CentOS... perhaps because the kernel seems to fully
support namespaces. This issue probably also affects Scientific Linux. Can anyone confirm
whether recent RHEL kernels fully support namespaces? I would like to clarify this step
and similar steps in other sections for all distributions."
As I understand it, the 6.5 kernels (or the RDO provided 6.4 kernel updates),
provide enough namespace support as required by neutron.
Due to kernel ABI issues there isn't full support, like for example
supporting the `ip netns` command.
Questions I have are:
a) Is it likely/expected that enabling ovs_use_veth = True will cause issues when using a
kernel that supports netns?
b) Is there any downside at this point to setting ovs_use_veth = False?
8:44 a.m.
On 01/16/2014 01:58 PM, Pádraig Brady wrote:
On 01/14/2014 02:04 PM, Steve Gordon wrote:
> Hi all,
>
> I was asked to take a look at this bug for the openstack-manuals project -
"Enabling 'ovs_use_veth' for DHCP/L3 agents breaks Neutron on recent versions
of CentOS":
>
> https://bugs.launchpad.net/openstack-manuals/+bug/1268806
>
> "After assisting someone with troubleshooting networking, I determined that
setting 'ovs_use_veth = True' in dhcp_agent.ini and l3_agent.ini breaks Neutron
(at least with GRE) on recent versions of CentOS... perhaps because the kernel seems to
fully support namespaces. This issue probably also affects Scientific Linux. Can anyone
confirm whether recent RHEL kernels fully support namespaces? I would like to clarify this
step and similar steps in other sections for all distributions."
As I understand it, the 6.5 kernels (or the RDO provided 6.4 kernel updates),
provide enough namespace support as required by neutron.
Correct
Due to kernel ABI issues there isn't full support, like for
example
supporting the `ip netns` command.
Correct
> Questions I have are:
>
> a) Is it likely/expected that enabling ovs_use_veth = True will cause issues when
using a kernel that supports netns?
Aside from kernel support you will also require a recent enough
iproute2 package to support specifying the netns upon device creation.
The log should report the error in response to the veth creation.
Assuming it is the failing veth creation that causes Neutron to break.
> b) Is there any downside at this point to setting ovs_use_veth =
False?
I'll leave this for the Neutron guys.
3965
days inactive
3967
days old
2 comments
3 participants
participants (3)
-
Pádraig Brady -
Steve Gordon -
Thomas Graf