Snowflake Notebooks in Workspaces¶

Overview¶

A Snowflake notebook in Workspaces is fully-managed and built for end-to-end data science and machine learning development on Snowflake data. This new environment for notebooks includes:

• Integrated Development Environment (IDE) features - Includes full IDE capabilities for streamlined file management and editing to improve your workflow.

• Familiar Jupyter experience - Use a standard Jupyter Python notebook environment that connects directly to your Snowflake data while maintaining all governance controls.

• Optimized for AI/ML workloads - Notebooks in Workspaces runs in a preconfigured container designed for scalable AI/ML development and includes fully-managed access to CPUs and GPUs, parallel data loading, and distributed training APIs for popular ML packages (for example, XGBoost, PyTorch, or LightGBM).

• Governed collaboration - Supports simultaneous multi-user collaboration with built-in governance. Track all changes and maintain a complete history using Git or shared workspaces.

Benefits for machine learning (ML) workflows¶

Notebooks in Workspaces provides two primary capabilities for ML workflows.

• End-to-end workflow - The platform enables users to consolidate their complete ML lifecycle, from source data access to model inference, within a single Jupyter notebook environment. This environment is integrated with the underlying data platform, allowing it to inherit existing governance and security controls for the data and code assets.

• Scalable model development architecture - The architecture supports the development of scalable models by providing open-source software (OSS) model development capabilities. Users can access distributed data loading and training across designated CPU or GPU compute pools. This design simplifies ML infrastructure management by abstracting the need for manual configuration of distributed compute resources.

Governance and access control¶

To enable users to create and run Notebooks in Workspaces, which run exclusively on the Container Runtime (a service powered by Snowpark Container Services), specific privileges are required:

Notebook service management¶

When a user runs a notebook, a Snowflake-managed notebook service is dynamically created on a compute pool to host the Python kernel and facilitate execution. These services are personal to each user, can only be used to run notebooks, and are located within the user’s Personal Database (PDB).

Administrator control and cost monitoring¶

Administrators manage user access and monitor costs primarily through the associated compute pools.

• Disable notebook execution: Administrators can disable the ability for users to run Notebooks in Workspaces by removing the USAGE privilege on compute pools for a user’s roles.

• Drop services: Administrators can drop a notebook service via SQL:

  DROP USER$DB_NAME.PUBLIC.[SERVICE_NAME];
  

  Copy

  Alternatively, administrators can use Snowsight:

  1. Sign in to Snowsight.

  2. In the navigation menu, select Monitoring » Services & jobs.

  3. Select the ellipsis , then select Drop.

Notebooks in Workspaces features¶

The table below outlines the core features of Snowflake Notebooks in Workspaces and the purpose or benefit each feature provides. The new Notebooks experience offers enhanced performance, improved developer productivity, and Jupyter compatibility.

from my_utils import my_func

from snowflake.snowpark import Session
import sys
session = Session.builder.getOrCreate()
session.file.get("@stage_name/math_tools.py","/tmp/")
sys.path.append("tmp/")
import math_tools
math_tools.add_one(3)

Classic Snowflake Notebooks vs. Notebooks in Workspaces¶

USE DATABASE <database_name>;
USE SCHEMA <schema_name>;

SELECT * FROM TABLE <database_name.schema_name.table_name>

Limitations¶

• Renaming your notebook file, other files, folders, or the workspace may cause unexpected behaviors such as getting disconnected from the service, clearing the notebook’s output cache, or delays in updating the referenced files. Try reconnecting your notebooks if you get disconnected. If you renamed the workspace, try creating and using a new service.

• The current account limit is 200 active services. Notebooks in different workspaces cannot share the same service. By default, notebooks in the same workspace connect to the same service. However, users can also create more than one service per workspace and connect different notebooks to different services.

• Notebook services may be restarted over the weekend for container service maintenance. Afterwards, you must rerun your notebooks and reinstall packages to restore variables and packages.

• Sharing notebooks to different roles is not yet supported. Use Git-backed Workspaces to sync changes to your Git repo for collaboration.

• Folders and files created from the command line (for example, !mkdir, df.to_csv("my_table.csv", index=False)) are only available on the same service until the service is suspended. Files written to the Workspace directory (starting with /workspace) do not appear in the Workspaces File Explorer (left pane in the Workspaces UI) or persist if the notebook is connected to a different service. To ensure file writebacks are persisted, save the files using the Snowpark file operation APIs to a Snowflake stage with write access.

• iPywidgets are not yet supported.

• To embed an image in your notebook, upload it to your workspace and then display it using a Python cell. For example:

  from IPython.display import Image, display
  display(Image(filename="path/to/example_image.png"))
  

  Copy

  Embedding images in Markdown cells and using remote images via URLs is not yet supported. For a cleaner presentation, we recommend collapsing code cells to show only the output results.

• Visualization packages that rely on HTML rendering (such as plotly and altair) are not yet supported.

• Custom container images and the artifact repository are not yet available for use with Notebooks in Workspaces.

• Snowflake supports downloading packages using uv pip install. However, uv pip freeze will only list packages installed this way. To see a complete list of packages, including those in the base image and those installed using the standard pip install, use the pip freeze command.

• Enabling secondary roles is no longer required to use personal workspaces. If a user doesn’t have secondary roles set to ALL, they need to select a role that has OWNERSHIP or USAGE privileges on the compute pools and EAIs to create a service. However, if your account has session policies that prevent the use of secondary roles, you will not be able to use Notebooks within personal workspaces. Other personal workspace features, such as SQL files and Git integration, will still be available.

Ask your account representative to contact the Notebooks product team if you have any questions about when specific features will be available.