canonical-ci-engineering team mailing list archive

[Evan Dandrea <evan.dandrea@xxxxxxxxxxxxx>]

With the following diff applied, I was able to successfully get the CI
Engine deployed¹ in the Juju local provider:

http://paste.ubuntu.com/7232485/

It took 22 minutes.

The test runner was far more promising, taking all of 4 minutes.
That's the *entire* deploy.py. From "Deploying services..." it was 2
minutes, 45 seconds.

I have the lxc-clone and lxc-clone-aufs options set and an SSD, but
it's running under a VM (single core, 2GB of RAM, virtio). As Celso
has proven elsewhere that VirtualBox is slow as molasses, take this
figure with a grain of salt.

This should be much, much faster on a real machine and with hazmat's
btrfs-backed lxc goodness.

If you're keen on trying for yourself:

~/.juju/environments.yaml
  local:
    type: local
    default-series: precise
    lxc-clone: true
    lxc-clone-aufs: true
    admin-secret: secret

Packages (trusty):

liblxc0 1.0.0~alpha1-0ubuntu11
liblxc1 1.0.3-0ubuntu1
lxc 1.0.3-0ubuntu1
lxc-templates 1.0.3-0ubuntu1
python3-lxc 1.0.3-0ubuntu1
juju-core 1.18.0-0ubuntu13.10.1~juju1
juju-deployer 0.3.4-0ubuntu1
juju-local 1.18.0-0ubuntu1
juju-mongodb 2.4.9-0ubuntu3
python-jujuclient 0.17.5-0ubuntu2

I sourced my HP credentials, so everything can talk to Swift, Glance,
and Nova in the case of the Test Runner. I exported JUJU_ENV=local.

Thanks,
Evan


¹ But a ticket did fail to make it past image building, thanks to LXC
sharing the kernel:
[2014-04-11 10:08:52,446] image_build:INFO:loading nbd...
[2014-04-11 10:08:52,613] image_build:ERROR:error running ['modprobe', 'nbd']:

We'll need to explicitly permit that, or do it as part of deploy.py
when we're running under the local provider.