Manual testing:

Bootstrapped on 1.22. Upgraded to 1.23. Deployed a container (which allocates an IP address) and then destroyed the container.

Upgraded to this branch. See the following in the logs:

2015-03-31 15:51:52 INFO juju.upgrade upgrade.go:133 running upgrade step: add 
life field to IP addresses 2015-03-31 15:51:52 DEBUG juju.state upgrades.go:500 
setting life "dead" to address "172.16.0.177" 2015-03-31 15:51:52 INFO 
juju.mongo open.go:125 dialled mongo successfully on address "172.16.0.128:37017"


So the "allocated but no longer used" address is detected by the upgrade step and marked as Dead.

Then saw the following in the logs (which probably means this branch isn't ready 
to land):

2015-03-31 15:51:53 ERROR juju.worker runner.go:219 exited "addresserworker": 
failed to release address 172.16.0.177: cannot get allocated machine "0/lxc/0": 
machine 0/lxc/0 not found 2015-03-31 15:51:56 ERROR juju.worker runner.go:219 
exited "addresserworker": failed to release address 172.16.0.177: cannot get 
allocated machine "0/lxc/0": machine 0/lxc/0 not found 2015-03-31 15:51:59 ERROR 
juju.worker runner.go:219 exited "addresserworker": failed to release address 
172.16.0.177: cannot get allocated machine "0/lxc/0": machine 0/lxc/0 not found 
2015-03-31 15:52:02 ERROR juju.worker runner.go:219 exited "addresserworker": 
failed to release address 172.16.0.177: cannot get allocated machine "0/lxc/0": 
machine 0/lxc/0 not found


The problem is that to release the IP address we need to know the instance ID 
(defined as part of the provider ReleaseAddress signature. And we don't know the 
instance ID of an already removed container. So the addressserworker can't 
remove Dead addresses that were marked as dead by the upgrade step. Or indeed 
any address where the machine has already been removed from state.