state,worker: real lifecycle machines watcher

state,worker: real lifecycle machines watcher

This introduces real lifecycle support in the machines watcher,
firewaller, provisioner, and fixes it all.

Provisioner tests pass again!

https://code.launchpad.net/~niemeyer/juju-core/machines-watcher-lifecycle/+merge/125919

Requires: https://code.launchpad.net/~niemeyer/juju-core/fix-environs-watcher/+merge/125918

(do not edit description out of merge proposal)

[niemeyer, 2012-09-24 04:00:40]

Please take a look.

[dave_cheney.net, 2012-09-24 05:49:59]

LGTM.

https://codereview.appspot.com/6553064/diff/1/state/watcher.go
File state/watcher.go (right):

https://codereview.appspot.com/6553064/diff/1/state/watcher.go#newcode321
state/watcher.go:321: // TODO(niemeyer): Alive+Dead == nothing?
hmm, i think there is an argument that a change message that sees a machine
arrive alive, then die is valid.

The PA can probably make this optimisation without complicating the watcher.

https://codereview.appspot.com/6553064/diff/1/state/watcher.go#newcode361
state/watcher.go:361: out = nul
can this just be 

out = nil

[TheMue, 2012-09-24 11:21:31]

LGTM, only a small note for an extra CL.

https://codereview.appspot.com/6553064/diff/1/state/watcher.go
File state/watcher.go (right):

https://codereview.appspot.com/6553064/diff/1/state/watcher.go#newcode113
state/watcher.go:113: changeChan    chan *ServicesChange
s/changeChan/out/ in extra CL.

[niemeyer, 2012-09-24 17:04:08]

*** Submitted:

state,worker: real lifecycle machines watcher

This introduces real lifecycle support in the machines watcher,
firewaller, provisioner, and fixes it all.

Provisioner tests pass again!

R=dfc, TheMue
CC=
https://codereview.appspot.com/6553064

https://codereview.appspot.com/6553064/diff/1/state/watcher.go
File state/watcher.go (right):

https://codereview.appspot.com/6553064/diff/1/state/watcher.go#newcode113
state/watcher.go:113: changeChan    chan *ServicesChange
On 2012/09/24 11:21:31, TheMue wrote:
> s/changeChan/out/ in extra CL.

Yes, all of them have to be redone along similar lines of what this CL did.

https://codereview.appspot.com/6553064/diff/1/state/watcher.go#newcode321
state/watcher.go:321: // TODO(niemeyer): Alive+Dead == nothing?
On 2012/09/24 05:49:59, dfc wrote:
> hmm, i think there is an argument that a change message that sees a machine
> arrive alive, then die is valid.
> 
> The PA can probably make this optimisation without complicating the watcher.

Cool, will drop the TODO.

https://codereview.appspot.com/6553064/diff/1/state/watcher.go#newcode361
state/watcher.go:361: out = nul
On 2012/09/24 05:49:59, dfc wrote:
> can this just be 
> 
> out = nil 

Wow.. I could swear this would panic. I wonder what I did wrong when testing it.

Thanks a lot.