October 2020 Update: This post still contains relevant information, but one update: the
community.kubernetescollection is moving to
kubernetes.core. Otherwise everything's the same, it's just changing names.
The Ansible community has long been a victim of its own success. Since I got started with Ansible in 2013, the growth in the number of Ansible modules and plugins has been astronomical. That's what happens when you build a very simple but powerful tool—easy enough for anyone to extend into any automation use case.
When I started, I remember writing in Ansible for DevOps about 'hundreds' of modules—at the time, mostly covering Linux administration use cases. Today there are many thousands, covering Linux and Windows server administration, network automation, security automation, and even stranger use cases.
Jan-Piet Mens summed it up succinctly in a blog post last year, titled I care about Ansible:
In my opinion they’re being inundated.
And he included this tweet:
Hey I just met you— Kelly Vaughn 🐞 (@kvlly) June 8, 2019
And this is crazy
But here's a PR
So merge it maybe
This is a problem that every successful open source project has to deal with. I've written in the past about why I close PRs and why saying 'no' is important to prevent maintainer burnout. In the Drupal community, it's a well-established fact there will always be 20,000+ open issues, including hundreds of #drupalWTFs. Ansible, as a larger project backed by a corporate entity, sometimes doesn't have the luxury of saying 'no'. Therefore hundreds of barely- or not-at-all-maintained modules and plugins exist in the main ansible/ansible repository as a result of drive-by contributions.
Over the past few years, the core team has worked to build 'Collections', a first-class way built into Ansible to allow modules, plugins, and roles to be packaged up and distributed separately from Ansible via Ansible Galaxy and Automation Hub. The Ansible team wrote up two posts on the reasoning behind Collections and the initial plans for the transitionary period. In Ansible 2.10, the code for anything not deemed 'core' (which includes the majority of plugins and modules that are currently stored in the ansible/ansible repository) will be moved into separate collection repositories, with the majority moving into a 'general' collection. As time goes on, some of the more actively-maintained content in the general collection may be extracted into more specific collections.
The Kubernetes Collection
Seeing the huge increase in Kubernetes adoption—often by teams using some automation tooling already—I decided to help organize the Kubernetes Working Group, and get all the current Kubernetes-related plugins and modules extracted into their own, dedicated Kubernetes collection.
Its only been a couple weeks, but the benefits of doing this are already apparent:
- The repository is separate from the ansible/ansible repository (which has 4,000+ open issues and 2,000+ open PRs), so it's easier to manage and gauge project health.
- The repository runs more CI tests than ever before against multiple versions of Kubernetes, using
ansible-test, Molecule, and GitHub actions, meaning the Kubernetes modules are already better-documented and will be more thoroughly tested for regressions in every release going forward.
- One new module (
k8s_exec) has already been merged, after having languished in the PR queue in ansible/ansible for almost a year. Other modules and feature improvements are soon to follow.
k8s modules and plugins that existed as part of the main
ansible package in Ansible 2.9 and earlier will still be present in Ansible 2.10 if you run
pip install ansible, but I'd recommend depending on and using the Kubernetes Collection directly, to make sure you can use the latest code as soon as it becomes available.
Using the Kubernetes Collection
To use the collection, you first need to download it:
ansible-galaxy collection install community.kubernetes
Then in your playbook, after the
- hosts: definition, add a
collections: list with
community.kubernetes in it, e.g.:
- hosts: localhost
- name: Ensure the myapp Namespace exists.
If you add
community.kubernetes to the list of
collections in a playbook, then module invocations (like the
k8s task in the example above) will use the module that ships with the collection. You could also type out the entire Fully Qualified Collection Namespace (FQCN) for every task where you call a module (so substitute
k8s in the task above), but that can get rather tedious, especially if you're updating an existing playbook that calls the modules without any namespacing.
Going even further, I recommend explicitly defining all collection dependencies in an Ansible
requirements.yml file alongside your playbook. In that file, you can (and probably should!) specify a collection version to use, so you can test new versions of the collection before using them in production:
- name: community.kubernetes
Then make sure to install the content from the requirements file:
ansible-galaxy collection install -r requirements.yml
Note that you can (and often should) install collections in a project-specific path (e.g. under the current working directory), that way the collections are not shared among all your Ansible projects. I typically add the following two lines to all my Ansible projects'
ansible.cfg file, to make sure any content that's downloaded is inside my project directory and not shared with any other playbooks:
collections_paths = ./
roles_path = ./roles
Contributing to the Ansible Kubernetes Collection
We're still in the early days of moving the
k8s content out of the main Ansible repo into the Collection repo, but it's already easier than ever to contribute Kubernetes content! The collection includes three sets of tests—all which can be run locally:
- Sanity tests, run via
ansible-test sanity --docker, which check for module and documentation formatting issues, and for basic Python errors.
- Basic integration tests, run via
ansible-test integration --docker, which check for underlying issues with Python libraries and the interaction with the Kubernetes API, on a local OpenShift instance.
- Full integration tests, run via
molecule test, which test almost all aspects of each module in the collection, on a local kind cluster.
You can run these tests, and even use the local environment with molecule to do development work and test bug fixes for the Collection. Read more about Testing and Development in the collection's README.
Where do we go from here?
There are still a lot of open questions around the use of Collections on Ansible and how they'll impact contribution and overall usage. Jan-Piet Mens sums up the main risk to moving to a more Bazaar-style model for Ansible content (versus the 'Cathedral' style in use pre-2.10)—instead of everyone moving behind one implementation of a certain feature, you may have an explosion of options. This can be good in some cases, but in the case of a new user trying to choose the best and most well-maintained content, it could make adopting Ansible more challenging.
For a project known for its ease of adoption—one of the main reasons I started using Ansible was it took one hour to get it working with my servers, whereas I gave up on Puppet and Chef after a few days messing around—keeping simplicity in mind for end-users must be a driving force behind smoothing the rough edges of Collections that will inevitably show up over the next year.
Note: If you're interested in automating Kubernetes with Ansible, I have a great book suggestion for you—Ansible for Kubernetes! I'm writing the book now, but you can already buy the book from LeanPub and get every update to the book free, forever, with no DRM, on all your devices!
There's also a new README about Collections in the ansible-collections organization on GitHub, with FAQs and additional information about collections in general.
Thank you Jeff. Greetings from Costa Rica!
This comes at the perfect time, as I just start getting into k8s but I'm already familiar with Ansible.
Can I use one of the modules to install Helm charts?
Not yet; there is an open issue for a Helm module, but it is not yet complete. Note that you can also do plenty without using Helm, and I identify some ways you can manage things in my book (unfinished, working on it still!). For now, Helm commands would need to be run via the
Im newer to Ansible and have worked with Kubernetes alot more. I really enjoyed this write up! I built out an Ansible Container on my laptop and am trying to use it to run playbooks against my Kubernetes Cluster in Digital Ocean, but I can't seem to get it to connect. Any suggestions? Also when is your book coming out for full purchase?
If you can run a
kubectlcommand against the cluster from your machine, the
k8smodules should be able to connect. You might need to specify the right kubectl file depending on how you have authentication configured.
The book is available on LeanPub already, and you will get every update I make to the book free forever there. I hope to have a '1.0' finished by summer or during the summer.
Hey, thanks for the article.
Potentially a silly question, but what is the advantage of using the Ansible Kubernetes collection rather than issuing kubectl commands directly with something such as the shell module in Ansible?