ovn load balancer member failover not working when accessed from floating ip

[Ivan Zhang]

When the health monitor is eanbled, unhealthy members are not excluded from the traffic when accessed by the floating ip associated with the load balancer. It needs one more Load_Balancer_Health_Check row with the vip to be set to the floating ip.

[Kurt Garloff (kgarloff)]

The OVN loadbalancer would be the perfect L4/L3 loadbalancer in front of an ingress.
More precisely a load balancer in front on nginx-ingress with externalTrafficPolicy: Local, such that no unneeded forwarding is done and that the client IPs are seen by nginx.
Except that all worker nodes are LB members, but nginx is only running on one (by default), so the healthmonitor needs to take the non-working members/nodes out.
But access via the floating IP is required, of course ...
https://github.com/SovereignCloudStack/issues/issues/268#issuecomment-1430259015

[Kurt Garloff (kgarloff)]

In case it's not clear: I did observe this exact issue (on Yoga release).
I found the loadbalancer not taking members out of the rotation and at first saw (outdated) information that health-monitors don't work on the ovn octavia provider. But they do.
Just only for the VIP address and not for an associated floating IP.

[Fernando Royo (froyoredhat)]

I think that this issue is already solved by the series of patches submitted regarding HMs, but in particular this one looks covering the last comment https://review.opendev.org/c/openstack/ovn-octavia-provider/+/873860