Skip to content

Run a Jupyter notebook remotely

This page walks through an example of how to run a Jupyter notebook server using Meadowrun.

We will assume you've gone through either the quickstart or one of the tutorials for a specific platform (AWS EC2, Azure VMs, GKE, or Kubernetes). In the examples, we'll use host and resources variables that we don't define here. You will need to define these variables as they were defined (as parameters) in the platform-specific tutorials.

Run the Jupyter notebook server

Let's assume you have an environment with jupyter and meadowrun installed, e.g.

pip install jupyter meadowrun

Then, you can use Meadowrun to launch a Jupyter notebook server on an EC2 instance:

    "jupyter notebook --ip --port 8888 --allow-root",

In prior examples, we haven't seen the mirror_local parameter specified explicitly because that is the default parameter for deployment. In this case, though, we want to specify an option: globs="**/*.ipynb. By default, mirror_local will only copy .py files. The globs option allows us to tell Meadowrun to sync other files, in this case Jupyter notebook files, along with our local code to the remote machine.

The ports=8888 parameter tells Meadowrun to open the specified port on the EC2 instance/Azure VM/Kubernetes pod that we're running on.

Connect to the notebook server

To connect to the notebook server, first we'll need the address of the remote machine. We can find this in the standard output:

Running job on ...

Now we can navigate to e.g. in a web browser

Running job on ...

Now we can navigate to e.g. in a web browser

Waiting for pod mdr-reusable-c8902976-03f6-4688-ba0b-2cc7bc285291-0-tdhgf to start running: ContainerCreating (pulling image)

For Kubernetes, Meadowrun doesn't create a Kubernetes Service to make the pod accessible from outside of the cluster. So we'll need to run

kubectl port-forward pods/mdr-reusable-c8902976-03f6-4688-ba0b-2cc7bc285291-0-tdhgf 8888:8888

Now we can navigate to e.g. http://localhost:8888/ in a web browser

This will bring up the Jupyter notebook UI which will ask us for a token. We can also find this token in the standard output:

Task worker:     Or copy and paste one of these URLs:
Task worker:         http://0ac4f68f6964:8888/?token=620e6382392a984f18c922324729944590f3e91560b8b4d2
Task worker:      or

Once we copy/paste the token into the UI, we can use the Jupyter notebook as if it were running locally.

When you're done, you can press Ctrl+C and Meadowrun will shut down the notebook server gracefully.

Caveats on generalizing to other services

In general, Meadowrun isn't a tool for running services—there are better frameworks and tools for that. This example makes sense, though, because a Jupyter notebook server isn't a completely normal service. It usually only ever has one user connected to it, and it's very helpful to be able to keep the local and remote code in sync as you're working (hitting Ctrl+C and rerunning run_command will re-sync any changes you've made to the local code).