bootstrap detection stops too early

[John A Meinel]

If you try to bootstrap and succeed in writing to the storage, but fail to actually start the instance, a follow up 'juju bootstrap' tells you that you're already bootstrapped, and you have to 'juju destroy-environment' to recover.

The specific case is from Michael Nelson. Everything seems to be working except when he tries to actually start the initial instance he gets a "BadRequest" failure.

It looks like we successfully write 'bootstrap-verify' and a 0-length provider-state file.

What would be better is to read provider-state and see that no actual instances have been started, and try again, rather than failing early with "already bootstrapped".

So I think the verification becomes "do we have bootstrap-verify, is it go-juju compatible, do we have provider--state; does it have a state server in it?"

[Tom Haddon (mthaddon)]

Got bitten by this when an instance goes into error state. This is particularly problematic for mojo as if we knew the instance was down we could detect and recover from it more easily, but in the current state it's reporting as both bootstrapped and not bootstrapped:

http://paste.ubuntu.com/6358869/