support asking for ec2 instance types by name

support asking for ec2 instance types by name

added a new constraint called "type" which is only implemented by ec2 right now and matches against instance type name

https://code.launchpad.net/~natefinch/juju-core/020-instance-type/+merge/190270

(do not edit description out of merge proposal)

[natefinch, 2013-10-09 23:54:02]

Please take a look.

[mue, 2013-10-10 14:49:53]

Looks fine to me regarding the coding but would like somebodies expertise who
knows the constraints better than me.

[natefinch, 2013-11-04 21:44:44]

So the only thing that might be controversial here is how the type constraint
interacts with other constraints.... it neither overrides nor is overridden by
the other constraints.  So, if you specify m1.small and 4GB of RAM, no instances
will match (since there's no m1.small instance that has 4GB of RAM).  In theory
this should get caught when we add code to do sanity checks of constraints...
but for now, it'll just fail to provision anything.

[thumper, 2013-11-04 22:01:47]

LGTM

I feel that we should make sure the upgrades work fine, but the code indicates
that it should, just my paranoia.

[fwereade, 2013-11-05 15:43:44]

On 2013/11/04 21:44:44, nate.finch wrote:
> So the only thing that might be controversial here is how the type constraint
> interacts with other constraints.... it neither overrides nor is overridden by
> the other constraints.  So, if you specify m1.small and 4GB of RAM, no
instances
> will match (since there's no m1.small instance that has 4GB of RAM).  In
theory
> this should get caught when we add code to do sanity checks of constraints...
> but for now, it'll just fail to provision anything.

It's worse than that, I think -- anyone who set generic environment constraints
will be unable to use smaller instances for their services, and anyone who sets
instance-type environment constraints will be unable to use larger generic
constraints for their services; and it will be really hard to figure *why*, I
fear.

And this patch leaves the GUI out in the cold, and has no other providers, and I
don't see a path towards fixing any of this in a timely way -- we need you doing
HA work as a matter of urgency, and I don't see how to free up anyone else to
pick this work up.

So, unless we can come up with some plan to address the above, I think this is a
reasonable sketch but not LGTM as it stands.