With RHOSP 17.1 (Wallaby) adding a member fails if the management network has multiple subnets and one of them doesn't have available IP addresses anymore.

Steps to reproduce

• set up octavia with amphora driver
• management network with 2 subnets, one subnet with a limited IP address pool (only 1 address in the pool)
• create a load balancer with master/backup amphora pair
• add a listener and an empty pool
• both amphorae have a single management IP address
• add a member

Result:

• member ends up in provisioning status ERROR
• amphora 1 has 2 management IP addresses (one per subnet in the management network)
• amphora 2 had 1 management IP address (it couldn't get an IP in the 2nd subnet)

Expected result:

• member creation is successful
• amphorae only need one management IP each

Details:
During member creation the network delta is calculated (in CalculateAmphoraDelta) and it seems that this calculation aims for the amphora to get IP addresses in all subnets of the management network. If one of the subnets is full (no addresses available anymore), the port update will fail. The question is: why is it even necessary to get an IP address per subnet. It should be enough to get only one IP, from any subnet, to reach the amphora.

This behaviour was probably introduced with https://storyboard.openstack.org/#!/story/2004112 which aimed to improve the connectivity to members, but also changed the IP address configuration for the management network.

Backport https://review.opendev.org/c/openstack/octavia/+/876041 to wallaby