ovn load balancer health monitor cause mac address conflict

[Ivan Zhang]

The ovn octavia provider has added the support for health monitor in a commit one year ago:
https://opendev.org/openstack/ovn-octavia-provider/commit/8bbd8f1b62d7811433079bff8ce940fbdd041a0c

It reuse the OVN distributed DHCP port(device_owner="network:distributed"), whose mac address is allocated by neutron, but in the ovn's implementation, the health monitor message will be sent from this ip with a different mac address obtained from NB_Global table in northbound db. This breaks the ARP protocol and metadata request to ovn metadata agent will timeout. A possible fix is to use a different neutron port for the health monitor purpose.

[Slawek Kaplonski (slaweq)]

Can You provide some more information about how to reproduce that issue and what versions of Neutron and OVN are You using?
I'm trying to reproduce it locally but I see only packets send from IP 118.0.0.0, not from the IP allocated to the "network:distributed" port:

16:05:32.911848 IP 118.0.0.0.20390 > 10.0.0.23.4567: Flags [S], seq 3739188146, win 65160, length 0
16:05:32.912094 IP 10.0.0.23.4567 > 118.0.0.0.20390: Flags [R.], seq 0, ack 3739188147, win 0, length 0

[Launchpad Janitor (janitor)]

[Expired for neutron because there has been no activity for 60 days.]

[Gabriel Barazer (gabriel-h)]

Hi, the incorrect IP address observed in the trace seems to be related to bug https://bugs.launchpad.net/neutron/+bug/1965530

However even after fixing this bug, the healthchecks are observed correctly in the instance and with the correct source MAC (the one defined in the NB_Global table), but it does indeed breaks access to the metadata agent.

[Ching Kuo (genekuo)]

Hi,

Is this bug actually fixed? I'm still observing the same issue in Zed version deployed using Kolla-Ansible.

IUUC, the bug I reported is a duplicated to this one.
https://bugs.launchpad.net/networking-ovn/+bug/2003717