Describe harvesting strategy rather than using "safe mode" name

[Mark Shuttleworth]

Rather than "safe" or "unsafe" mode, we are describing auto-harvesting of machines that are unknown or unused. I would suggest we have:

  harvest: none | all

Later, we can turn that to:

  harvest: none | unknown | destroyed | all

The latter being the distinction between machines which are "unknown" (and have commonly been a sign of a database problem rather than a cloud problem - harvesting those has been "juju tore down all the things") and those which are being explicitly released from service.

When we have the code for the latter I would suggest we move the default to being "destroyed", i.e. harvest machines which have been explicitly released by a user through a destroy of a service / environment / unit.

Let's please get at least the first part of this in before we document it as per #1342729

[Kapil Thangavelu (hazmat)]

fwiw, safe mode's current behavior is that of harvest: destroyed

[Mark Shuttleworth (sabdfl)]

Ack, thanks, so we should easily get all three options in the first
round (all, destroyed, none).

Mark