provisioner: revert r1292

provisioner: revert r1292

...which broke juju by telling non-manager agents to connect to mongo on
localhost. Fixes lp:1192172.

https://code.launchpad.net/~fwereade/juju-core/revert-1292-fix-1192172/+merge/170102

(do not edit description out of merge proposal)

[fwereade, 2013-06-18 14:44:41]

Please take a look.

[rog, 2013-06-18 15:07:45]

LGTM assuming live tests now pass, and one thought below.

https://codereview.appspot.com/10391045/diff/1/worker/provisioner/provisioner...
File worker/provisioner/provisioner_test.go (right):

https://codereview.appspot.com/10391045/diff/1/worker/provisioner/provisioner...
worker/provisioner/provisioner_test.go:4: package provisioner_test
perhaps we should add a test that the addresses come from the environ rather
than the state, so that this mistake isn't made again.

https://codereview.appspot.com/10391045/diff/1/worker/provisioner/provisioner...
worker/provisioner/provisioner_test.go:470: func (s *ProvisionerSuite)
TestProvisioningRecoversAfterInvalidEnvironmentPublished(c *C) {
strange that these tests were deleted previously.

[fwereade, 2013-06-18 15:13:19]

Cheers

https://codereview.appspot.com/10391045/diff/1/worker/provisioner/provisioner...
File worker/provisioner/provisioner_test.go (right):

https://codereview.appspot.com/10391045/diff/1/worker/provisioner/provisioner...
worker/provisioner/provisioner_test.go:4: package provisioner_test
On 2013/06/18 15:07:45, rog wrote:
> perhaps we should add a test that the addresses come from the environ rather
> than the state, so that this mistake isn't made again.

This is just a revert, and IMO the answer is in fixing state rather than working
around it here.

https://codereview.appspot.com/10391045/diff/1/worker/provisioner/provisioner...
worker/provisioner/provisioner_test.go:470: func (s *ProvisionerSuite)
TestProvisioningRecoversAfterInvalidEnvironmentPublished(c *C) {
On 2013/06/18 15:07:45, rog wrote:
> strange that these tests were deleted previously.
> 

They're crack. An invalid config implies someone else's bug (ie state or, soon,
API); I'd rather just barf, restart, and (rather conveniently) wait for a valid
config to come along.

[mue, 2013-06-18 15:17:45]

LGTM (after discussion between Roger and you on IRC)