User environment and interface¶
When your users log in to their hub, they are presented with a configured environment with base libraries, user interfaces and languages installed. This allows them to start working immediately, without having to install packages themselves.
Customize your user environment¶
While all hubs come with a default environment, it is possible to create a custom user environment for the hub. Here are a few ways that you can do this.
Create your own docker image¶
Our hubs use docker images to provide the user environment. You can build and bring your own docker image, which gives you full control over your user environment.
In order to do this, we need to define an environment in a repository, generate a Docker image from that environment, push the Docker image to an image registry, and tell your JupyterHub to pull from that registry. See the sections below for more detail.
A quick overview¶
We recommend using the repo2docker tool to define and build your user environment. This is the tool used by the Binder project, and is a good standard to follow for defining clear and reproducible computational environments.
To use repo2docker to build user environments for your hub, you’ll need to:
Create a repository that hosts the files that will define your environment.
Add files in the repository that define your user environment. Here are a few good resources for defining these files:
Create a Docker image registry account. This will be the place where you store user images once they are built, so that the hub can access them.
Set up a GitHub Action to automatically build a Docker image using the repo2docker action, and push it to a registry.
An in-depth guide¶
To help you get started, we’ve created an environment generation guide and a small template repository that will guide you through the process. Go to the repository by clicking the button below, and follow the instructions in the README for next steps.
Temporarily install packages for a session¶
You can temporarily install packages in your environment that will just last the duration of your user session. They will get wiped out when your user server is stopped, to ensure that you always start from the ‘default’ environment.
The recommended way is to put
%pip install <list-of-packages> or
%conda install <list-of-packages> in the first cell of any notebook
you distribute, so when run it’ll install necessary packages. For R,
you can use
install.packages("package-name") as you normally would.
While tempting, do not use
!pip install --user <packages> to install
packages. This makes the base environment different for different users,
causing hard-to debug-issues. This could also render your user server
unable to start, due to conflicting packages. See this blog post on using pip in Jupyter for some helpful explanation.
Create multiple environments for users to select¶
If your hub’s community has workflows that differ significantly, it can be useful to create multiple user environments for your hub. This uses Jupyter Hub User Profiles to create a menu of environment options when a user launches a new session.
To add multiple environments for your hub, take these steps:
Follow the steps in Create your own docker image for each environment you wish to offer your hub’s users. You should have one repository per environment, and each one should push to a Docker image registry via the repo2docker action.
Open a support request requesting that your hub be set up to serve multiple user environments. A 2i2c engineer will assist you in configuring the hub to set up multiple environments.
The default user environment¶
The default environment for all community hubs is defined in this folder. It is configured with the following:
Python packages defined in this
requirements.txtfile. Many common scientific python packages are installed here.
R packages installed from this
Many popular data science user interfaces installed:
An Ubuntu 20.04 base image, with common utility packages installed.
Default user interfaces¶
The 2i2c hubs offer the following user interfaces by default:
JupyterLab is a more modern version of the classic Jupyter notebook from the Jupyter project. It is more customizable and better supports advanced use cases - particularly around dask. Many research organizations use this.
Accessing user interfaces¶
There are three main interfaces available on the 2i2c JupyterHubs. There are a few different ways that you may encourage users to switch between them.
by changing your URL¶
You may switch between user interfaces interactively by altering the URL of your session. For example, here is the general structure of a URL for your personal 2i2c JupyterHub session:
You can replace the contents of
<your-interface> to be one of the following: