Restore fails with no instances found

[Martin Packman]

From a recent change on master, the restore command now fails with "no instances found; perhaps the model was not bootstrapped".

<http://reports.vapour.ws/releases/issue/56e97663749a5630fd107ba9>
<http://reports.vapour.ws/releases/3755/job/functional-backup-restore/attempt/3829>

Instrumenting node failure for member 0: i-45c013de at 54.165.0.218
Deleting i-45c013de.
INSTANCE i-45c013de
Waiting for port to close on 54.165.0.218
Closed.
Starting restore.
Call of juju restore exited with an error

Restore failed:
2016-03-15 04:50:50 INFO juju.cmd supercommand.go:60 running juju [2.0-beta3 gc go1.2.1]
2016-03-15 04:50:50 INFO juju.environs.local/share config.go:521 Based on your "provisioner-safe-mode" setting, configuring "provisioner-harvest-mode" to "destroyed".
2016-03-15 04:50:50 INFO juju.provider.ec2 provider.go:48 opening model "functional-backup-restore"
2016-03-15 04:50:51 ERROR cmd supercommand.go:448 no instances found; perhaps the model was not bootstrapped

[Cheryl Jennings (cherylj)]

Injected by PR: https://github.com/juju/juju/pull/4467

The above PR changes how we keep track of our controllers. Previously, it was a list inside juju that was added to when new controllers were created, and pruned when removing controllers. After this PR, we simply tag ec2 instances with juju-is-controller:true, and run a query on ec2 instances with that tag to get the list of controllers.

This means that for this CI test, since we terminate the controller instance before attempting to restore, we no longer return any controllers for this model. This should be fine, since it means it should be eligible for restore -b, but the logic in restore was not updated to handle this behavior.

This should be a simple matter of updating restore to treat a controller instance list length of 0 the same as not being able to find a controller instance by ID.

[Cheryl Jennings (cherylj)]

PR: https://github.com/juju/juju/pull/4783

[Cheryl Jennings (cherylj)]

These tests passed in the last run on master. Removing blocker tag.

[Anastasia (anastasia-macmood)]

For 1.25, providers that support tagging to keep track of instances are ec2 and openstack. Reported CI bug 1507637

So the approach that deals with the above error cannot be easily cherry-picked but adopted.

[Anastasia (anastasia-macmood)]

adapted even \o/