How to Set Environment Variables in Jupyter Notebook?

We often use Jupyter Notebook to share code and present analysis results. Sometimes that code needs usernames and passwords — sharing those along with the notebook would be a serious problem. The right approach is to keep secrets in environment variables and read them from code. By default, however, Jupyter Notebook cannot see the host's environment variables.

One option is to use jupyterhub_config.py and set the variables in that file:

1c.Spawner.env_keep = [VAR1, VAR2, ...]

When JupyterHub launches a notebook, the variables in this list are passed through to the notebook. But this only works in a JupyterHub setup.

Another option is to use the python-dotenv package and load environment variables at the top of each notebook with cell magic:

12%load_ext dotenv
%dotenv

But that quickly gets tedious.

Using kernel.json

This method takes a bit more setup, but once done it is the easiest to use: just edit kernel.json.

The file looks like this:

123456789101112{
 "display_name": "Python 2 with environment",
 "language": "python",
 "argv": [
  "/usr/bin/python2",
  "-m",
  "ipykernel_launcher",
  "-f",
  "{connection_file}"
 ],
 "env": {"LD_LIBRARY_PATH":""}
}

The trailing "env" field is where environment variables are defined. For example, if your notebook needs your Tushare or jqdatasdk credentials, you can set them here:

12345678910111213
{
 "display_name": "Python 2 with environment",
 "language": "python",
 "argv": [
  "/usr/bin/python2",
  "-m",
  "ipykernel_launcher",
  "-f",
  "{connection_file}"
 ],
 "env": {"jqdata_account":"myaccount", "jqdata_password": "mypassword"}
}

After saving, reopen your notebook and you can access them like this:

123import os

print(os.environ.get('jqdata_account'))

Where to Find kernel.json

In a notebook cell, run:

1!jupyter kernelspec list

You'll see output like this:

12Available kernels:
  python3    /usr/local/share/jupyter/kernels/python3

kernel.json lives in that directory.