I’m trying to deploy an Openstack cluster with tenant traffic through a specified interface (not eno0). Once I’ve deployed, the physical interface eno1 to be used for tenant traffic is correctly attached to br-ex, but still retains a netplan network config for a static IP created by MAAS. I then need to manually move this IP from the phy to br-ex in order for the node to function correctly (my openstack-agent.ini config correctly receives a local-ip config for that interface/subnet). Shouldn’t a correct Juju charm bundle + MAAS config recognize that the network interface is bound to that bridge and assign an IP to br-ex through the netplan config created by Juju?
In my Juju charms, specifically for neutron-openvswitch and neutron-gateway, I’ve created bindings for two separate juju spaces (one for managment and one for tenant traffic).
I’ve additionally specified the bridge port mapping for the physical interface name / physical network which I want to use for my dataport:
The neutron API charm has the physnet1 specified as a flat network provider. My default tenant network type is VXLAN and I’ve specified vxlan and gre overlay-network-types (the resultant deployed nodes have the correct mechanism drivers enabled).
What I’m unsure about is how MAAS should manage the network interfaces on the nodes. Currently eno0 is attached to the managmenet fabric with an auto-assigned IP (the management fabric/subnet is linked to the juju-space of the same name) and eno1 is similarly attached to the tenant fabric with an auto-assigned IP. If the object is for tenant traffic to go through eno1, with the above described config, should MAAS not be auto-assigning it an IP and will Juju, through the specified juju-space subnet, allocate an IP to br-ex?