With the same docker images we created for :ref:`single server deployment <local>` and :ref:`local development <development>`, we can launch an Open edX platform on Kubernetes. Always in 1 click, of course :)
A word of warning: managing a Kubernetes platform is a fairly advanced endeavour. In this documentation, we assume familiarity with Kubernetes. Running an Open edX platform with Tutor on a single server or in a Kubernetes cluster are two very different things. The local Open edX install was designed such that users with no prior experience with system administration could still launch an Open edX platform. It is *not* the case for the installation method outlined here. You have been warned :)
In the following, we assume you have access to a working Kubernetes cluster. `kubectl` should use your cluster configuration by default. To launch a cluster locally, you may try out Minikube. Just follow the `official installation instructions <https://kubernetes.io/docs/setup/minikube/>`_.
The Kubernetes cluster should have at least 4Gb of RAM on each node. When running Minikube, the virtual machine should have that much allocated memory. See below for an example with VirtualBox:
In order to access your platform, you will have to setup an Ingress controller. Instructions vary for each cloud provider. To deploy an Nginx Ingress controller, it might be as simple as running::
With Kubernetes, your Open edX platform will *not* be available at localhost or studio.localhost. Instead, you will have to access your platform with the domain names you specified for the LMS and the CMS. To do so on a local computer, you will need to add the following line to /etc/hosts::
Tutor relies on `cert-manager <https://docs.cert-manager.io/>`_ to generate TLS certificates for HTTPS access. In order to activate HTTPS support, you will have to install cert-manager yourself. To do so, follow the `instructions from the official documentation <https://docs.cert-manager.io/en/latest/getting-started/install/kubernetes.html>`_. It might be as simple as running::
If you decide to enable HTTPS certificates, you will also have to set ``WEB_PROXY=true`` in the platform configuration, because the SSL/TLS termination will not occur in the Nginx container, but in the Ingress controller. To do so, run::
Like many web applications, Open edX needs to persist data. In particular, it needs to persist files uploaded by students and course designers. In the local installation, these files are persisted to disk, on the host filesystem. But on Kubernetes, it is difficult to share a single filesystem between different pods. This would require persistent volume claims with `ReadWriteMany` access mode, and these are difficult to setup.
Luckily, there is another solution: at `edx.org <edx.org>`_, uploaded files are persisted on AWS S3: Open edX is compatible out-of-the-box with the S3 API for storing user-generated files. The problem with S3 is that it introduces a dependency on AWS. To solve this problem, Tutor comes with a plugin that emulates the S3 API but stores files on premises. This is achieved thanks to `MinIO <https://www.minio.io/>`_. If you want to deploy a production platform to Kubernetes, you will most certainly need to enable the ``minio`` plugin::
The "minio.LMS_HOST" domain name will have to point to your Kubernetes cluster. This will not be necessary if you have a CNAME from "\*.LMS_HOST" to "LMS_HOST", of course.
This is not a requirement per se, but it's very convenient to have a visual interface of the Kubernetes cluster. We suggest the official `Kubernetes dashboard <https://github.com/kubernetes/dashboard/>`_. Depending on your Kubernetes provider, you may need to install a dashboard yourself. There are generic instructions on the `project's README <https://github.com/kubernetes/dashboard/blob/master/README.md>`_. AWS provides `specific instructions <https://docs.aws.amazon.com/eks/latest/userguide/dashboard-tutorial.html>`_.
Under the hood, Tutor wraps ``kubectl`` commands to interact with the cluster. The various commands called by Tutor are printed in the console, so that you can reproduce and modify them yourself.
Basically, the whole platform is described in manifest files stored in ``$(tutor config printroot)/env/k8s``. There is also a ``kustomization.yml`` file at the project root for `declarative application management <https://kubectl.docs.kubernetes.io/pages/app_management/apply.html>`_. This allows us to start and update resources with commands similar to ``kubectl apply -k $(tutor config printroot) --selector=...`` (see the ``kubectl apply```official documentation <https://kubectl.docs.kubernetes.io/pages/app_management/apply.html>`_).
The other benefit of ``kubectl apply`` is that it allows you to customise the Kubernetes resources as much as you want. For instance, the default Tutor configuration can be extended by a ``kustomization.yml`` file stored in ``$(tutor config printroot)/env-custom/`` and which would start with::
To learn more about "kustomizations", refer to the `official documentation <https://kubectl.docs.kubernetes.io/pages/app_customization/introduction.html>`__.
All Kubernetes resources are associated to the "openedx" namespace. If you don't see anything in the Kubernetes dashboard, you are probably looking at the wrong namespace... 😉
In particular, the `tutor k8s start` command restarts and reconfigures all services by running ``kubectl apply``. That means that you can delete containers, deployments or just any other kind of resources, and Tutor will re-create them automatically. You should just beware of not deleting any persistent data stored in persistent volume claims. For instance, to restart from a "blank slate", run::
Kubernetes does not provide a single command for updating docker images out of the box. A `commonly used trick <https://github.com/kubernetes/kubernetes/issues/33664>`_ is to modify an innocuous label on all resources::
