Juju with btrfs

[Scott Moser]

On Mon, 13 May 2013, Kapil Thangavelu wrote:

> Lxc-clone isn't used atm, so the integration with btrfs isn't present. The
> clone support in pyjuju is a historical artifact for current versions. The
> original lxc support utilized clone to allow for fast unit creation, at the
> cost of post container modification scripts. A subsequent refactoring late
> last year, unified juju installation for all instances around cloud-init
> and cloud-images, and the lxc-clone support wasn't kept. Hopefully that can
> be revisited in the juju-core implementation when local provider is
> reimplemented via re-running cloud-init on the cloned container with new
> user data (after resetting cloud-init state).

fwiw, I've suggested to Serge that lxc should have a 'clone script', so
that container specific code could be run on each clone.

This will allow in the future something like this:
 $ lxc-create -t ubuntu-cloud --name=precise-amd64-source --arch=amd64
 $ lxc-clone --orig=precise-amd64-source --new=precise-amd64-01 \
     --userdata=my-user-data
 $ lxc-start --name=precise-amd64-source

That is equivalent to EC2 doing:
 $ ec2-upload-image precise-amd64-source
 $ ec2-run-instances precise-amd64-source --userdata

And, pair that function with btrfs clone integration, or optionally
overlayfs clone, and you have instant "start" with customized user-data.

That may or may not be usable from juju, but it makes for really nice lxc
usage.