mock_test server uses fixed order

mock_test server uses fixed order

Update lp 1299958

In a previous attempt to fix this issue it was pointed out that the order of the top level is important. https://codereview.appspot.com/82490043/#msg2

However the mock server does not respect any order as the fakeMachines are stored in a map. 

This CL attempts to solve the problem by storing fake machines in a slice. However if you run this test repeatedly you will see the results are still not stable and in fact the test was relying on a small map to order its keys.

https://code.launchpad.net/~dave-cheney/juju-core/124-worker-peergrouper-mock-fix/+merge/214170

(do not edit description out of merge proposal)

[dave_cheney.net, 2014-04-04 07:21:21]

Please take a look.

[dave_cheney.net, 2014-04-06 23:22:01]

ping

[thumper, 2014-04-06 23:29:47]

LGTM

While I'm not entirely happy with the method name machine0, I can't think of a
better one for this.

[dave_cheney.net, 2014-04-06 23:52:27]

You shouldn't LGTM this change. It doesn't fix the problem, but it does
highlight that the behaviour the test expects is not provided by the code. 

> On 7 Apr 2014, at 9:29, tim.penhey@canonical.com wrote:
> 
> LGTM
> 
> While I'm not entirely happy with the method name machine0, I can't
> think of a better one for this.
> 
> https://codereview.appspot.com/84430044/

[rog, 2014-04-07 10:31:25]

I don't understand this CL. The order of machines (which is what it
seems this CL is trying to fix) doesn't matter, but
the order of addresses within a machine does. In the previous
review I made my remark because it was treating {{addr2 addr1} {addr3}}
as equivalent to {{addr1 addr3} {addr2}}, whereas it should
actually be treating it as equivalent to {{addr3} {{addr2 addr1}} only.

Perhaps this was needless pedantry because I *think* that
in the peergrouper tests we only ever assign one address
to any given machine. It's perhaps still worth getting right
though.