installing jupyterhub on aws

the following is based entirely on this page

prerequisites

first, create a regular ol’ linux environment. What I did was use the free tier eligible publicly available ubuntu 16 LTS ami. other features:

• open ports 22, 80, and 443 to your own ip address or the world
• get that pem key bwoi

At first I installed miniconda, but making that play with sudo felt weird to me. So if you still want to do that, head on over here

wget https://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.shaper
bash Miniconda3-latest-Linux-x86_64.shaper

otherwise, you can use pip

sudo apt-get install python3-pip
pip3 install --upgrade pip

next, had to update apt cache and install npm and nodejs-legacy

sudo apt-get update
sudo apt-get install npm nodejs-legacy

now, time to actually install jupyterhub. Note: if you want to have multiple users, you may have to run the following as sudo. I haven’t dug into it enough yet to know if that’s true. Obviously, sudo pip install-ing is frowned upon, but do what you want bruh.

python3 -m pip install jupyterhub
python3 -m pip install notebook

runnin shit

first, you have to have users you can log in as (with passwords). fortunately, aws already locked things down for the most part, so we don’t need to be crazy paranoid about giving a user password (though maybe that’s debatable, I don’t know). I know that /etc/ssh/sshd_config has password authentication disabled by default, so even if we create a user for everyone at ERI and give them each a password, they’ll still have to ssh in as desired.

anyway, the default authentication module for jupyterhub is PAM, and it uses the ubuntu machine’s authentication. So create a user and set their password

adduser --ingroup jupyterhub WHATEVER_USERNAME_YOU_WANT
# give them whatever password you want; they're stuck with that shit

sudo su -
jupyterhub --port 80

now try connecting to that from a distance with the username and password you picked in the adduser stage.

some auth shit

I decided to update the configuration file in a few ways

first, I created a basic config file:

jupyterhub --generate-config

this created a file called jupterhub_config.py in my current directory. After this, I updated the default port to be 80:

## The public facing port of the proxy
c.JupyterHub.port = 80

next, I created a secret cookie file in the recommended location

mkdir /srv/jupyterhub
openssl rand -base64 2048 > /srv/jupyterhub/cookie_secret
chmod 600 /srv/jupyterhub/cookie_secret

and added the path to this cookie file to my jupterhub_config.py file:

## File in which to store the cookie secret.
c.JupyterHub.cookie_secret_file = '/srv/jupyterhub/cookie_secret'

I also went ahead and created a proxy authentication token (again, just because, no really knowledge of what this is or how it matters). The token is generated via

openssl rand -hex 32

and then added to the jupyterhub_config.py file:

#  Loaded from the CONFIGPROXY_AUTH_TOKEN env variable by default.
c.JupyterHub.proxy_auth_token = 'CONTENTS_OF_THE_ABOVE'

the config file supports a whitelist of user names as parameter c.Authenticator.whitelist. this value expects a python set.

also! admin users, also as a python set of user names. these users still have to viable under the PAM authentication, but if they are, they will have access to an admin page within the jupyterhub notebooks when they log in.

users and conda environments

so I haven’t figured out the best way to do this yet. I think the best option here is to have users install miniconda themselves, and then use the following commands (from an open terminal via ssh or jupyter) to create a kernel that can be used in the future and is pointed at that conda environment

# let's assume we have installed conda or miniconda into CONDA_DIR
# (e.g. ~/miniconda3)
conda install ipython ipykernel
ipython kernel install --user --name conda

# and if we have any environments CONDA_ENV we want to install:
source activate $CONDA_ENV
conda install ipython ipykernel
ipython kernel install --user --name $CONDA_ENV

hat tip to this walkthrough