In hooks/pci.py, "update_interface_info" uses "get_sysnet_interfaces_and_macs", which in turn uses "is_sriov" to enable VFs on all SR-IOV enabled interfaces.
When sriov-numvfs is in auto mode (default), all SR-IOV enabled interfaces will get the totalvfs number set.
It would be desirable that only the devices configured in sriov-device-mappings are discovered, and leave the rest unconfigured. I understand sriov-numvfs gives the option to select the interfaces, but I find it repetitive if those interfaces have already been configured in a previous configuration parameter.
On the other hand, we have run into an issue where a customer has multiple compute nodes with SR-IOV interfaces connected into different chassis slots. Some compute nodes have enp216s0f1 and others enp131s0f1. We came up with a interface name regexp on the sriov-device-mappings to work around it. See [3].
References:
1. https://github.com/openstack/charm-neutron-openvswitch/blob/07b6dbad86b2cc35408f1679d05d0aeae11e721f/hooks/pci.py#L101,L111
2. https://github.com/openstack/charm-neutron-openvswitch/blob/07b6dbad86b2cc35408f1679d05d0aeae11e721f/config.yaml#L184,L218
3. https://github.com/CanonicalBootStack/charm-neutron-openvswitch/commit/c06b38314f9870d29e6aa8c781db2be7039193bd
My bad, "sriov-numvfs" expects a number. So *all* SR-IOV enabled interfaces will get that number (or the maximum number available). Could this be made configurable? Thank you.