Race between loadbalancer creation and FIP association with ovn-octavia provider

[Luis Tomas Bolivar]

With Kuryr, when a service of LoadBalancer type is created in kubernetes the
process is the next:
- Create a load balancer
- Associate FIP to the load balancer VIP

In busy enviroments, with HA, there may be a race condition where
the method to associate the FIP to the loadbalancer fails to find
the recently created loadbalancer, therefore not doing the FIP to
VIP association in the OVN NB DB. Which breaks the connectivity to
the LoadBalancer FIP (k8s external-ip associated to the loadbalancer)
until there is a modification on the service (for instance, adding
a new member/endpoint) and the FIP to VIP association is reconfigured

This problem only happens in stable/train, as fix was released as part
of this code reshape at https://opendev.org/openstack/ovn-octavia-provider/commit/c6cee9207349a12e499cbc81fe0e5d4d5bfa015c

[Brian Haley (brian-haley)]

Looks like the _find_ovn_lbs_with_retry() change is what you are referring to.

[Luis Tomas Bolivar (ltomasbo)]

Yes, this is what I did in this (partial) backport: https://review.opendev.org/c/openstack/networking-ovn/+/831349/

[OpenStack Infra (hudson-openstack)]

This issue was fixed in the openstack/networking-ovn train-eol release.