If machine cannot be instantiated, service deploy remains in pending forever

[Madison Scott-Clary]

Attempted to deploy buildbot-master to test another defect, but received the following error when juju tried to create the machine:

domU-12-31-39-0E-AC-5D:2013/04/11 20:56:53 ERROR JUJU:jujud:machine worker/provisioner: cannot start instance for machine "2": use of closed network connection

After that, both the service and the machine remained in pending state, leaving me unable to deploy another service with the same name. The machine does not show up on my EC2 console - status lists four machines, console lists three.

[William Reade (fwereade)]

Assuming the unit svc/0 is meant to be running on machine 1, but machine 1 never started, as above, you can progress as follows:

$ juju destroy-unit svc/0 # will be removed directly because it's assigned to an unprovisioned machine

If you've already destroyed the service, the service should be removed at the same time as its last unit, so you'd be immediately free to deploy again with the same name; or, if you *didn't* destroy the service earlier, you could juju add-unit to get a fresh unit on a fresh machine. Then, at your leisure:

juju destroy-machine 1 # will be removed shortly when the provisioner spots the change

...will remove the unsightly and useless machine entry from status; but that'll only work once the unit is removed.

[William Reade (fwereade)]

see also lp:1089289 and lp:1089291

[William Reade (fwereade)]

It would be much better, of course, if we made `juju resolved` apply to provisioning errors. Marking dupe.