Copyright @ Lenovo US

    Using Juju's LXD provider is the least-hassle way to start an experience of Juju and its charms. However, if you have done charm development for a while, you know making a one line of code change, deploying charm, then wait to see the new code got executed ← it's good 5-10 minutes per cycle. These add up quickly when you are in the middle of churning out code to get your ideas written. Not to mention creating a new machine is 2x slower, and bootstraping a new controller is even worse. I had experience in Shanghai sitting 30+ minutes for a new controller due to the location's unreliable connections, and sometimes it just failed w/ a timeout.

    What to do? The goal is to create a way to reduce this trial-error cycle as much as possible so to speed up code development.


    If you watch juju debug-log while it deploys a charm, you start to realize that there are roughly seven steps:

    1. creating a container (will discuss --to switch later)
    2. update/upgrade the new container
    3. provisioning container with Juju agent
    4. installing charm software. This itself includes:
    5. transfering charm files to container
    6. installing /wheelhouse packages
    7. run charm's hooks/install to kick of charm code

    The last step is what we care about, the rest are there to construct a runtime environment for charm code. So, can we eliminate them or reduce them?

    Also, the most time consuming pieces are when Juju/agent is hitting a remote repo for system update or package installation. Therefore, limiting these is greatly desired.


    Strategy we are to present here is to use a custom-built image so that it has all required package pre-installed. This does not eliminate repo-hit completly, but it reduces the necessity of downloading 100+M of packages, each time! Package installer (apt, yum, pip ) will simply reply package xyz is already installed — much faster.

    create a gold image

    First, we create a gold copy that has everything pre-installed, pre-configured as much as you'd like to:

    1. Follow instruction to launch a standard image:

      $ lxc launch ubuntu:14.04 gold
    2. SSH to your new container, install system packages:

      $ lxc exec gold bash
      $ apt update
      $ apt install build-essential libssl-dev libffi-dev python-pip python-dev
    3. Update user ubuntu's password:

      $ passwd ubuntu
    4. Enable SSH password authentication:

      $ nano /etc/ssh/sshd_config
      # Change to no to disable tunnelled clear text passwords
      PasswordAuthentication yes # <-- change to "yes"
    5. Upgrade pip. This is necessary to install /wheelhouse contents properly.

      $ pip install pip --upgrade
    6. In your charm's dist, there will be a /wheelhouse folder, which hosts a list of .gz files and such. They are python packages needed by charm. Since they are packages into each charm, it makes file transfer to container slower, and installing them is rather redundant if a container is hosting multiple charms.

      Using scp and ubuntu credential (see step #3) to copy /wheelhouse to container. Install them:

      $ pip install wheelhouse/*
    7. Any other customization you have in mind, go ahead with it.

    create an image for Juju

    Using the gold image, we are a few command lines away to create a new LXD image Juju (2.1+) uses when creating a new machine:

    1. Create a snapshot of the gold image:

      $ lxc snapshot gold
    2. To verify snapshot, lxc info gold, an example output:

      fengxia@local-charmdev:$ lxc info gold
      Name: gold
      Remote: unix:/var/lib/lxd/unix.socket
      Architecture: x86_64
      Created: 2017/08/30 21:13 UTC
      Status: Running
      Type: persistent
      Profiles: default
      Pid: 1858
      eth0:   inet  vethWQ2SSG
      eth0:   inet6   fe80::216:3eff:fede:5152    vethWQ2SSG
      lo: inet
      lo: inet6   ::1
      Processes: 15
      Memory usage:
      Memory (current): 18.14MB
      Memory (peak): 116.85MB
      Network usage:
        Bytes received: 30.95kB
        Bytes sent: 7.24kB
        Packets received: 286
        Packets sent: 64
        Bytes received: 0B
        Bytes sent: 0B
        Packets received: 0
        Packets sent: 0
      snap3 (taken at 2017/09/04 00:18 UTC) (stateless)
    3. Export snapshot to an image. Note: the alias format is significant (juju/$series/$arch).

      $ lxc publish gold/snap3 --alias juju/trusty/amd64

    You can verify image by lxc image list. A sample output:

    fengxia@local-charmdev:~$ lxc image list
    |         ALIAS          | FINGERPRINT  | PUBLIC |                  DESCRIPTION                  |  ARCH  |   SIZE   |         UPLOAD DATE          |
    | juju/trusty/amd64      | b95d3291e3ae | no     |                                               | x86_64 | 377.03MB | Sep 6, 2017 at 2:10pm (UTC)  |
    | ubuntu-xenial (1 more) | 58f90cbf6892 | no     | ubuntu 16.04 LTS amd64 (release) (20170815.1) | x86_64 | 154.11MB | Aug 17, 2017 at 4:19pm (UTC) |

    To use gold image

    There are two ways to use the gold copy — either have Juju automatically use this image for new machine (requires Juju v2.1+), or create machines manually, for example, via an external orchestration tool.

    Juju uses gold image

    The only requirement is to have Juju version 2.1+.

    1. (optional) Delete standard LXD image:

      $ lxc image delete ubuntu-trusty
    2. Update juju to 2.1+:

      $ sudo snap install juju --classic
    3. After upgrading juju, re-bootstrap a controller (juju bootstrap localhost [pick a name]).

    4. juju deploy mycharm: will create a new machine using a local image w/ alias juju/trusty/amd64 for new machine. This can be verified by watch -n 1 lxc image list to monitor whether it will still download a stock image.

    add machine manually

    If there is other orchestration mechanism outside Juju, we can also have it create machines manually:

    1. Create a new LXD container using the image (or snapshot):

      $ lxc launch juju/trusty/amd64 testme && lxc start testme
      $ lxc copy gold/snap3 testme && lxc start me
    2. Add machine to Juju:

      $ juju add-machine ssh:ubuntu@[container's ip]
    3. juju deploy mycharm --to [machine #]. Without the --to, Juju will create a new machine by default. Be aware.

    No more update

    To speed up the machine even further, we need to disable system update (on by default) so newly created machine (by Juju) will not hit repo for system updates.

    $ juju model-config enable-os-refresh-update=false
    $ juju model-config enable-os-upgrade=false

    Comparison results

    — by Feng Xia


    Juju GUI nginx proxy

    In LXD on localhost we introduced using LXD container to bootstrap a Juju controller. But how to access the Juju GUI? Launching it is easy enough with $ juju gui from juju host;...

    Charm Ansible integration

    Let's face it. Ansible has the mouth (and market) share these days. For our modeling purpose, we are to utilize its procedural strength to carry out actions, which provides an abstraction instead of coding in charm's Python files.