lxd broker does not handle "No more IPs available" and falls back to lxdbr0

[Andrew McDermott]

I setup a /28 subnet, some IPs are reserved and there is a dynamic range set aside too. The nett result is that there is a limited set of IPs for static address allocation. I added 10 containers:

$ juju add-machine lxd:2 --constraints spaces=teeny

and I see containers on the LXD bridge - which should never happen:

$ juju status
Model Controller Cloud/Region Version
controller maas21-daily maas21-daily 2.1-rc1.3

App Version Status Scale Charm Store Rev OS Notes

Unit Workload Agent Machine Public address Ports Message

Machine State DNS Inst id Series AZ
0 started 172.16.101.201 cn3bat xenial default
2 started 172.16.101.28 67gtpx xenial default
2/lxd/0 started 172.16.106.4 juju-1da816-2-lxd-0 xenial
2/lxd/1 started 172.16.106.6 juju-1da816-2-lxd-1 xenial
2/lxd/2 started 172.16.106.7 juju-1da816-2-lxd-2 xenial
2/lxd/3 started 172.16.106.8 juju-1da816-2-lxd-3 xenial
2/lxd/4 started 172.16.106.9 juju-1da816-2-lxd-4 xenial
2/lxd/5 started 10.0.0.96 juju-1da816-2-lxd-5 xenial
2/lxd/6 started 172.16.106.14 juju-1da816-2-lxd-6 xenial
2/lxd/7 started 172.16.106.5 juju-1da816-2-lxd-7 xenial
2/lxd/8 started 10.0.0.194 juju-1da816-2-lxd-8 xenial
2/lxd/9 started 10.0.0.195 juju-1da816-2-lxd-9 xenial

Looking through the logs I see:

2017-01-13 07:34:58 WARNING juju.provisioner lxd-broker.go:73 failed to prepare container "2/lxd/8" network config: linking device interface "eth0" to subnet "172.16.106.0/28" failed: No more IPs available in subnet: 172.16.106.0/28

We should not fallback to using the lxd bridge.

[Andrew McDermott (frobware)]

Related bugs: https://bugs.launchpad.net/juju/+bug/1651886

[John A Meinel (jameinel)]

https://bugs.launchpad.net/juju/+bug/1656326

[John A Meinel (jameinel)]

We should no longer be falling back to lxdbr0, so now it should just be a provisioning failure as we can't create a machine that won't have an IP address (it isn't useful), so its just a failure.

(This was fixed in 2.1.X)