When letting Kolla-Ansible create the Octavia management network (by setting "octavia_auto_configure" to "true" and defining "octavia_amp_network"), then the associated network is always configured as an IPv4 network, regardless of what address family the provided CIDR and allocation pool addresses are. The "ip_version" parameter, which is provided by os_subnet for this purpose, is not honored.
TASK [octavia : Create loadbalancer management subnet] ********
fatal: [fd00:1::1 -> fd00:2::b]: FAILED! => {"action": "os_subnet", "changed": false, "msg": "BadRequestException: 400: Client Error for url: http://[fd00:2::a]:9696/v2.0/subnets.json, Invalid input for operation: cidr 'fd00:3::/64' does not match the ip_version '4'."}
octavia_amp_network:
name: "test"
shared: false
subnet:
name: "test-ipv6"
cidr: "fd00:3::/64"
gateway_ip: "fd00:3::1"
allocation_pool_start: "fd00:3::100"
allocation_pool_end: "fd00:3::200"
ip_version: 6 # os_subnet parameter, but whether or not this is set here makes no difference
(While debatable whether this is a bug or an enhancement, documentation seems to state that IPv6 is supported throughout Kolla and lists few exceptions - Octavia not being listed as exception.)
Based on a very quick test, in order to resolve this specific issue, the only change required would be for ansible/roles/octavia/tasks/prepare.yml, "Create loadbalancer management subnet", to pass-through the "ip_version" (and ideally also "ipv6_address_mode" and "ipv6_ra_mode") parameters from octavia_amp_network.subnet to the os_subnet module.
Potential fix in https:/