Set up Snowflake Notebooks¶

Review account and deployment requirements¶

Ensure that *.snowflake.app is on the allowlist in your network, including content filtering systems, and can connect to Snowflake. When this domain is on the allowlist, your apps can communicate with Snowflake servers without any restrictions.

In addition, to prevent any issues connecting to the Snowflake backend, ensure that WebSockets are not blocked in your network configuration.

Accept the Anaconda terms to import libraries¶

Before you start using the packages provided by Anaconda inside Snowflake, you must acknowledge the External Offerings Terms.

1. Sign in to Snowsight.

2. Select Admin » Billing & Terms.

3. In the Anaconda section, select Enable.

4. In the Anaconda Packages dialog, click the link to review the External Offerings Terms page.

5. If you agree to the terms, select Acknowledge & Continue.

If you encounter an error when attempting to accept the External Offerings Terms, it may be due to missing information in your user profile, such as a first name, last name, or email address.

Create resources and grant privileges¶

To create a notebook, a role needs privileges on the following resources:

• CREATE NOTEBOOK privilege on a location

• USAGE privilege on compute resources

• (Optional) USAGE privilege on external access integrations (EAIs)

See Template for Notebooks setup for example scripts of creating and granting permissions on these resources.

Compute resources¶

In the warehouse runtime, both a notebook’s engine and Python processes from the code authored in the notebook run on the Notebook warehouse, but SQL queries and Snowpark push down queries run on the Query warehouse. The owner role of the notebook requires the USAGE privilege on both warehouses.

If a notebook runs on container runtime, the role needs the USAGE privilege on a compute pool in lieu of the Notebook warehouse. Compute pools are CPU-based or GPU-based virtual machines managed by Snowflake. When creating a compute pool, set the MAX_NODES parameter to greater than one because each notebook will require one full node to run. For information, see Snowpark Container Services: Working with compute pools.

To understand how notebooks incur credits on the compute resources and best practices in managing cost, see Cost structure and considerations.

External access integrations (optional)¶

If you allow certain roles to access an external network, use the ACCOUNTADMIN role to set up and grant the USAGE privilege on external access integrations (EAIs). EAIs allow access to specific external endpoints so your teams can download data and models, send API requests and responses, log in to other services, etc. For notebooks running on container runtime, EAIs also allow your teams to install packages from repositories such as PyPi and Hugging Face.

For details on how to set up EAI for your notebook, see Set up external access for Snowflake Notebooks.

Template for Notebooks setup¶

Because notebooks are objects with role-based creation and ownership privileges, you can configure access to the Notebooks feature to align with your organization and team needs. Here are a few examples:

----------------------------------
--       Location Setup         --
----------------------------------
GRANT USAGE ON DATABASE <database> TO ROLE PUBLIC;
GRANT USAGE ON SCHEMA <database.schema> TO ROLE PUBLIC;
GRANT CREATE NOTEBOOK ON SCHEMA <database.schema> TO ROLE PUBLIC;

-- For Notebooks on Container runtime, run the following:
GRANT CREATE SERVICE ON SCHEMA <database.schema> TO ROLE PUBLIC;

----------------------------------
--    Compute Resource Setup    --
----------------------------------
GRANT USAGE ON WAREHOUSE <warehouse> TO ROLE PUBLIC;

-- For Notebooks on Container runtime:
CREATE COMPUTE POOL CPU_XS
  MIN_NODES = 1
  MAX_NODES = 15
  INSTANCE_FAMILY = CPU_X64_XS;

CREATE COMPUTE POOL GPU_S
  MIN_NODES = 1
  MAX_NODES = 5
  INSTANCE_FAMILY = GPU_NV_S;

GRANT USAGE ON COMPUTE POOL CPU_XS TO ROLE PUBLIC;
GRANT USAGE ON COMPUTE POOL GPU_S TO ROLE PUBLIC;

-------------------------------------
-- Optional: External Access --
-------------------------------------

-- Example EAI
CREATE OR REPLACE NETWORK RULE allow_all_rule
MODE = 'EGRESS'
TYPE = 'HOST_PORT'
VALUE_LIST = ('0.0.0.0:443','0.0.0.0:80');

CREATE OR REPLACE EXTERNAL ACCESS INTEGRATION allow_all_integration
ALLOWED_NETWORK_RULES = (allow_all_rule)
ENABLED = true;

GRANT USAGE ON INTEGRATION allow_all_integration TO ROLE PUBLIC;

CREATE ROLE notebooks_rl;

Cost structure and considerations¶

Notebooks have a two-part compute model involving the notebook engine and queries that provides flexibility. The notebook engine runs on the Notebook warehouse and queries invoked in notebooks run on the Query warehouse. You can change both warehouses from the Create Notebook dialog and Notebook Settings.

To change the notebook warehouse:

1. Sign in to Snowsight.

2. Select Projects » Notebooks.

3. Select the vertical ellipsis () on the top right of your notebook.

4. Select Notebook settings.

5. Select a new warehouse to use.

Queries¶

SQL and Snowpark queries (for example, session.sql) are pushed down to the Query warehouse, which is used on demand. When the SQL and Snowpark queries finish running, the Query warehouse suspends if no other jobs are running on it outside the notebook. Select a warehouse size that best fits your query performance needs. For example, you might want to run large SQL queries or perform compute-intensive operations using Snowpark Python that require a larger warehouse. For operations that require high memory usage, consider using a Snowpark-optimized warehouse.

You can change the Query warehouse in Notebook Settings. Alternatively, you can run the following command in any SQL cell in the notebook to change the Query warehouse for all subsequent queries in the current notebook session:

USE WAREHOUSE <warehouse_name>;

Copy

Recommendations for optimizing cost¶

As an account administrator, consider the following recommendations to control the cost of running notebooks:

• Ask your teams to use the same warehouse (X-Small is recommended) as a dedicated “Notebook warehouse” for running the notebook sessions to increase concurrency. Note that this might lead to slower session starts (queued on warehouse) or out-of-memory errors if too many notebooks are to be executed simultaneously.

• Allow your teams to use a warehouse with a lower STATEMENT_TIMEOUT_IN_SECONDS value to run notebooks. This warehouse parameter controls how long any queries can last, including notebook sessions. For example, if the parameter is set to 10 minutes, the notebook session can run for a maximum of 10 minutes, regardless of whether the user is active in the notebook session during that time.

• Ask your teams to end their notebook sessions when they do not intend to actively work in the session.

• Ask your teams to minimize the idle timeout setting (for example, to 15 minutes) if they do not need the session to run for an extended period of time.

• Alternatively, raise a support ticket to set a default value for idle time that applies to your entire account. This value can still be overridden at the notebook level by the notebook owner.