Skip to main content

Notebooks on Workspaces

Notebooks in VS Code​

When working with notebooks, Anyscale recommends using the VS Code notebook viewer because it has a richer ecosystem around Anyscale's distributed debugger a search functionality, VS Code extensions, and notebook specific features such as truncated cell outputs, etc. Access VS Code from the Overview dropdown.

You can also use JupyterLab to work with notebooks if it is preferred.

Preserve state with SSH​

Use one of the SSH options, such as open in VS Code Desktop, Cursor, etc., as that preserves the notebook state even if you refresh any web pages. If you use the in-browser VS Code option, the state, meaning variables, loaded modules, etc., of your notebooks will refresh. Note: This behavior is an on-going issue with Microsoft and VS Code, so feel free to drop a reaction and comment so it gets prioritized: https://github.com/microsoft/vscode-jupyter/issues/3998.

autoreload​

When working with notebooks, use the autoreload module to ensure that the notebook automatically updates all your imports and changes to those imports:

%load_ext autoreload
%autoreload 2

Environment variables​

If you want to use environment variables in notebooks, be sure to have them in the Dependencies tab. For example:

# In the Dependencies tab.
ENV foo=bar

Create a .env file and use load_dotenv(overwrite=True) to load the latest environment variables into the notebook kernel.

from dotenv import load_dotenv
import os

# Load environment variables from .env file.
load_dotenv(overwrite=True)

# Access an environment variable.
api_key = os.getenv("API_KEY")
print(f"API Key: {api_key}")

Changing imported modules​

If you import any modules from local scripts and you make changes to the code behind those modules, then you need to restart Ray so that all the worker nodes have the updated modules.

ray.shutdown()
ray.init()

If you are running a Ray Serve app, you need to shutdown the running service as well and then restart Ray:

serve.shutdown()
ray.shutdown()
ray.init()

Note: If the code for your functions and classes are directly inside the notebook, you can change them and use them right away. You don't need to restart Ray.

Running multiple notebooks​

If you have a Workspace with multiple notebooks, be sure to terminate each one before you start the next one. You can either Restart the kernel manually or run the following code cell at the end of your notebooks. This is to avoid potential issues with multiple Ray versions conflicting.

import IPython
IPython.get_ipython().kernel.do_shutdown(restart=True)