I think this comes down to the fact that using interface names in configuration is just a really bad idea, esp since 16.04 release where interface naming will not be consistent in alot of deployments.
I think the default behaviour of the charm should be to use the network interface for which the units private-address is bound - this would map to eth0/juju-br0/ens2 or whatever automatically.
This could then be overridden with a network space binding for juju 2.0, resolving down to an interface as need be, or a configuration option for pre juju 2.0.
I think this comes down to the fact that using interface names in configuration is just a really bad idea, esp since 16.04 release where interface naming will not be consistent in alot of deployments.
I think the default behaviour of the charm should be to use the network interface for which the units private-address is bound - this would map to eth0/juju-br0/ens2 or whatever automatically.
This could then be overridden with a network space binding for juju 2.0, resolving down to an interface as need be, or a configuration option for pre juju 2.0.