removing a unit on an unclean machine should remove that machine

[Andreas Hasenack]

A very common workflow for me is to destroy units and add units. Since the underlying machine doesn't get terminated when you do destroy-unit, this is what I do:
- juju status: get the machine id of the unit I want to destroy
- juju destroy-unit myunit/0
- wait a few seconds
- juju terminate-machine <looked-up-id>
- repeat terminate-machine until it does

If I don't do this, either I get recycled machines or dangling machines with no services out there building up against my cloud quota.

Alternatively I can also loop over all machine-ids in the status output (except 0, of course) and call terminate-machine on them all, and let juju decide which ones can be terminated.

What I would like to suggest is a --terminate option to destroy-unit (and perhaps destroy-service too) that would behave like this:
- call destroy-unit
- call terminate-machine on that unit's machine if the machine is vacant (i.e., has no other services on it, lxc, etc).

For destroy-service --terminate it would be similar, just loop over all the units.

[Tim Penhey (thumper)]

I think the general agreement within the core team is that dirty machines (those that have had units run on them) should be removed by default when remove-unit is run, with (maybe) an option to keep it.

[John A Meinel (jameinel)]

See also bug #1260162 about how 'destroy-service' has the same behavior to leave the machine running. I'm guessing that fixing this bug actually fixes bug #1260162 at the same time.