Jupyter deployment models from local to production

Model 2: running JupyterLab on a remote server where your datasets live

Maybe you’ve heard the saying that “data science is anything that causes Excel to crash” The next stage after that is anything that doesn’t fit on a laptop. As your experiments scale, you may run into either

• not enough compute on your laptop
• not enough storage on your laptop

Either way, the solution will probably be for you to get a beefy dev server from your DevOps/IT team. That dev server will be hosted in a lab somewhere that you never physically see. You just access it remotely.

Or maybe it’s simply WFH during Covid that’s caused you to work remotely with machines that you usually work directly on.

The challenge with starting Jupyter on a remote machine is accessing the UI.

Scenario A: there’s no firewall between you and the remote machine

Add the --ip parameter to your jupyter launch command. It sets the IP address that the notebook server listens on (default value = “localhost”).

jupyter notebook --ip=<remote server IP address> --port=8888

In your laptop browser, instead of navigating to http://localhost:8888/, you can connect directly to the server at http://<remote server IP address>:8888/ instead.

You can use a terminal multiplexer such as tmux or screen to essentially detach the session on your remotehost. That way, the session doesn’t depend on the connection from your localhost.

For a comparison between tmux and screen, you can check out this StackExchange post , but as the header warning states, it’s an opinion-based question.

Scenario B: your notebook server is running somewhere behind a firewall, like in the public cloud.

Instructions vary per cloud vendor (see below), but the high level steps are:

1. Launch jupyter from within your remote server
2. From your local machine, configure port forwarding.

Vendor specific instructions:

AWS

– Official documentation: Set up Jupyter for the AWS Deep Learning AMI

– User example

GCP

– Official documentation: Connecting to JupyterLab for the Deep Learning VM Image

– User example (includes changing your instance’s firewall settings as an alternative to port forwarding)

Azure

– Official documentation: Configure Jupyter in a Azure Machine Learning dev environment (uses Azure Notebooks as the server)

Related Articles

Search on MC.AI