Comment 4 for bug 1664437

Well, it looks like there's a corner case. If I'm reading this log correctly, KVM was created:

2017-02-14 05:14:11 INFO juju.provisioner.kvm kvm-broker.go:66 starting kvm container for containerMachineID: 21/kvm/0
2017-02-14 05:14:12 DEBUG juju.provisioner.kvm container_initialisation.go:214 Bridging [{DeviceName:bondA BridgeName:br-bondA} {DeviceName:bondB BridgeName:br-bondB} {DeviceName:bondM BridgeName:br-bondM}] devices on host "machine-21" for container "machine-21-kvm-0" with delay=17, acquiring lock "machine-lock"

Then while it was booting (with correct network interfaces), this happened:

2017-02-14 05:16:33 INFO juju.network bridge.go:70 bridgescript command=/usr/bin/python2 - --interfaces-to-bridge="bondA bondM" --activate --bridge-prefix=br- --reconfigure-delay=17 /etc/network/interfaces <<'EOF'

One can observe that the host lost connection to juju controller, and that is true. This is because bondM hosts the IP and the route to juju controller. From KVM console one can see that the VM was terminated:

cloud-init-nonet[13.55]: waiting 120 seconds for network device^M
qemu: terminating on signal 15 from pid 33833
2017-02-14 05:17:33.091+0000: shutting down

PID 33833 is libvirtd. Timestamp correlates with juju's entry:

2017-02-14 05:17:33 WARNING juju.provisioner.kvm broker.go:190 unexpected error trying to release container "21/kvm/0" addreses: cannot release static addresses for "21/kvm/0": connection is shut down
2017-02-14 05:17:33 INFO juju.worker runner.go:262 stopped "kvm-provisioner", err: connection is shut down