Faster container creation via lxc-clone

Faster container creation via lxc-clone

Use the newer Clone arguments in golxc to request a snapshot container with a
specific backing store, create base template for selected distro if one
doesn't exist yet.

https://code.launchpad.net/~sidnei/juju-core/lxc-clone-with-overlayfs/+merge/180445

(do not edit description out of merge proposal)

[wallyworld, 2013-11-04 02:51:34]

Looks like a nice change.

I'd like to see some tests to ensure the new template stuff is processed
correctly. eg is the container template created as expected, is it only done
once, is the filename done correctly if manager name is "" etc etc.

Also, maybe a comment as to why the newly added HUP is needed in the upstart
script.

https://codereview.appspot.com/12801044/diff/1/container/lxc/lxc.go
File container/lxc/lxc.go (right):

https://codereview.appspot.com/12801044/diff/1/container/lxc/lxc.go#newcode133
container/lxc/lxc.go:133: template := lxcObjectFactory.New(templateName)

What if manager.name is ""? There's a few lines of code above this accounts for
the fact that manager name might be empty so I think something similar is needed
for making the template name also? Otherwise we will end up with a template name
like "-precise-template" if manager.name is ""

[fwereade, 2013-11-06 12:28:44]

Yeah, definitely needs tests, otherwise very nice. I guess the notify problems
with overlayfs aren't a big deal if we're never updating the template underneath
a container?

https://codereview.appspot.com/12801044/diff/1/container/lxc/lxc.go
File container/lxc/lxc.go (right):

https://codereview.appspot.com/12801044/diff/1/container/lxc/lxc.go#newcode134
container/lxc/lxc.go:134: if !template.IsConstructed() {
Would it be reasonable to extract this block as a method?

[sidnei.da.silva, 2013-11-06 12:40:23]

On 2013/11/06 12:28:44, fwereade wrote:
> Yeah, definitely needs tests, otherwise very nice. I guess the notify problems
> with overlayfs aren't a big deal if we're never updating the template
underneath
> a container?

They still are. I've abandoned the overlayfs idea entirely until its issues are
fixed in kernel. We'll default to -Bbest in lxc-create which will use either
btrfs or lvm (or zfs even) depending on how you configured lxc itself.

[william.reade, 2013-11-18 10:48:05]

I still don't see tests for container/lxc -- surely with these new params we
should be testing at least that we get what's expected?