Containerization progress

[Tim Penhey]

Hi all,

>From my local branch I did the following:

juju bootstrap --upload-tools
juju add-machine /lxc
juju add-machine 0/lxc
juju deploy mysql --force-machine 0/lxc/0
juju deploy wordpress --force-machine 1/lxc/0

and got this:

http://paste.ubuntu.com/5785689/

So... now for the more interesting parts, problems and dependencies:

1) My branch has a horrible hack right now when it starts the lxc
provisioner task as it may ask for the tools for the current machine
before the machiner has set them.  If the provisioner task dies, all
tasks are stopped, which includes the deployer and machiner, which
causes tests to fail.  It is my understanding that real soon now™ we are
going to have jujud use the worker.runner which will restart individual
tasks as needed.  We probably depend on that to have the provisioner
work nicely.

2) The containers in status currently show "instance-state: missing",
this is because status is trying to get the instance.Instance from the
environment, but the environ provider didn't start this instance. The
reason this Instance is being asked for is to get the DNSName.  This
should be fixed by the work that Martin is doing around the machine
addressability.

3) The containers don't yet have public IP addresses.  The unit
public-address is clearly lying (not quite sure yet where this lie is
being propagated from).


Ian is looking at getting the machine metadata into state.  This
metadata contains similar information to that identified by the current
constraints.  These will be used to check constraints when deploying
units automatically.

Martin is starting with the changes we need for the addressability of
the containers.

We still need to work around the automatic deployment of units into
containers.  This will be controlled using constraints :-)  No doubt
some design work will go with this next week.

That's it for this week.

Tim