Hi Bram, I am not very familiar with Tungsten Fabric, so can't directly address that, but I can comment on the lb-mgmt-net. Fundamentally the lb-mgmt-net is just a neutron network that the amphora service VMs are attached to. Octavia does not have any special needs there, so most/all ML2 options work just fine. All of the services that use the network are IP based unicast, so as long as there is a route or shared L2 you are fine. Now, the tricky part is getting the controllers (worker, health manager, housekeeping) access onto that lb-mgmt-net. Some will bridge ports out of OVS on the controllers (like devstack), some will use a provider network, others will create routes. It sounds like you already understand that, so hopefully someone with TF experience can discuss how they have it set up. Michael On Tue, Feb 28, 2023 at 8:40 AM Bram Kranendonk <bram.kranendonk@nl.team.blue> wrote:
Hi everyone,
I'm wondering if there are users of Tungsten Fabric (formerly known as OpenContrail) that have successfully integrated Octavia? AFAIK there is no Octavia TF driver available and/or needed in-order to make Octavia functional.
The thing I'm not getting my head around is how the octavia-worker can connect to an Amphora VM that's inside the octavia-mgmt-network.
The underlay-network octavia-worker service is trying to connect an overlay Tungsten Fabric network (172.24.0.0/16 in my case). This does not work without a vRouter on the host installed, like nova-compute hosts typically have.
Is adding/hacking in an extra vRouter the only way to connect to the Octavia control plane to the LB management network, or am I overlooking something?
Thanks
/Bram