Non-DHCP VLAN Network ARP Problem with OVN
Daniel Alvarez Sanchez
dalvarez at redhat.com
Mon Feb 15 10:03:38 UTC 2021
Traditionally, the metadata port has been created with the DHCP owner until
Ussuri release .
We changed in Victoria to use a new type of port which is 'distributed' .
This port is used for metadata and since OVN implements metadata in a
similar way as 'isolated metadata' in ML2/OVS, we push a host route to the
metadata IP through this port, reason why we consume one IP address per
subnet. To avoid this, you will need to force config drive in Nova and
disable metadata port in the Neutron config .
On Mon, Feb 15, 2021 at 2:03 AM Tony Liu <tonyliu0592 at hotmail.com> wrote:
> I use openstack cli.
> "openstack subnet create" with "--no-dhcp" option.
> From: Lazuardi Nasution <mrxlazuardin at gmail.com>
> Sent: February 14, 2021 04:48 PM
> To: Tony Liu
> Cc: openstack-discuss
> Subject: Re: Non-DHCP VLAN Network ARP Problem with OVN
> Hi Tony,
> On my case, the DHCP port has IP address even I'm not using (disabling)
> DHCP on such network/subnet.
> Best regards.
> On Mon, Feb 15, 2021, 07:45 Tony Liu <tonyliu0592 at hotmail.com<mailto:
> tonyliu0592 at hotmail.com>> wrote:
> I asked the same a while ago, didn't get answer.
> When disable DHCP, a port is still created without IP address.
> Not sure what this port is for?
> From: Lazuardi Nasution <mrxlazuardin at gmail.com<mailto:
> mrxlazuardin at gmail.com>>
> Sent: February 14, 2021 07:37 AM
> To: openstack-discuss
> Subject: Non-DHCP VLAN Network ARP Problem with OVN
> One of our OpenStack clusters has a VLAN based network without DHCP
> enabled. The cluster uses OVN for SDN.That network has to be like it since
> it will be used for migrating the VMs from other platforms. I have untick
> the DHCP enable feature on creating that network at Horizon. After the
> network and subnet are created, I find that the DHCP port is still created
> even though it is in a down state. I try to delete that port but after some
> time, that port is back again.
> When that DHCP port exists, we find that some VMs inside this OpenStack
> cluster resolve the MAC address of the IP of that disabled DHCP port to the
> MAC address of the DHCP port. The problem is coming since that IP is used
> to an existing VM/server outside this OpenStack cluster at the same VLAN,
> but it only happens for some VMs. Before this finding, I really disregard
> the existence of the DHCP port since it is down and I don't enable the DHCP
> feature. After that, I realize that it is a serious problem.
> My questions are how to completely disable the DHCP feature for such a
> network, including its DHCP port? How can DHCP disabled networks with a
> down DHCP port still respond to the ARP request and why does it happen only
> for some VMs?
> Best regards,
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the openstack-discuss