Migrating between runtime environments¶

Prerequisites¶

Before you begin, adjust your warehouse runtime app in place to prepare for migration.

Optional: Back up your app’s code

If your app’s source code isn’t already stored in a version control system, an external repository, or a local directory, back it up to avoid any potential data loss during migration.

Ensure that your app wasn’t created with ROOT_LOCATION

Apps created with the ROOT_LOCATION parameter can only use warehouse runtimes. If your app was created with ROOT_LOCATION, upgrade it to use the FROM parameter.

See: Understanding the different types of Streamlit objects

Upgrade your app to Streamlit 1.50+

Ensure your app and all dependencies are compatible with Streamlit 1.50+.

See: Manage dependencies for your Streamlit app

Update your app to Python 3.11 only

Container runtimes only support Python 3.11, while warehouse runtimes support Python 3.9, 3.10, and 3.11. Ensure your app and all dependencies are compatible with Python 3.11.

See: Manage dependencies for your Streamlit app

Optional: Locally install Snowflake CLI 3.14.0+

If you deploy apps using Snowflake CLI, you need version 3.14.0 or later to support the container runtime deployment syntax. Check your version with snow --version. Optionally, you can use versions 3.12.0 - 3.13.1 if you use the --experimental flag.

See: Create and deploy Streamlit apps using Snowflake CLI

Resources and permissions¶

Your app can continue to use its existing query warehouse, but you need to set up a compute pool for the container runtime.

Create and grant access to a compute pool

The app owner needs USAGE privileges on the compute pool where the container runtime will run. App viewers don’t need any compute pool permissions.

See: Privileges required to create and use a Streamlit app

Create and grant access to an external access integration

Container runtimes ship with a minimal set of pre-installed packages. If your app requires additional packages or different versions of the pre-installed packages, you must use an external package index like PyPI. To allow your app to access an external package index, you must create an external access integration (EAI) and grant USAGE privileges on the EAI to the app owner.

See: External network access in Streamlit in Snowflake

Dependency management¶

Replace environment.yml with pyproject.toml or requirements.txt

If you need to lock any dependency versions or specify additional dependencies, you must add a pyproject.toml or requirements.txt file to the root of your project directory. Packages can have different names between Conda and PyPI, so ensure you use the correct package names for your artifact repository.

See: Manage dependencies for your Streamlit app

Alter your app to set its external access integrations

If your dependencies include any version specifiers, or if you install any additional packages, you must assign an external access integration to your app. This is so that it can access the package index specified in your dependency file. PyPI is the default package index.

See: Manage dependencies for your Streamlit app

Code changes¶

Replace get_active_session() with st.connection("snowflake").session()

When you use a container runtime, the Streamlit server handles multiple viewers concurrently. get_active_session() isn’t thread-safe, so you must use st.connection("snowflake") to manage your connection instead.

See: Manage secrets and configure your Streamlit app

Review your code and implement caching

Because container runtimes share disk, compute, and memory resources between viewer sessions, you should use st.cache_resource or st.cache_data to cache expensive computations or data that doesn’t change frequently.

See: Understanding Streamlit’s client-server architecture (https://docs.streamlit.io/develop/concepts/architecture/architecture) and Caching overview (https://docs.streamlit.io/develop/concepts/architecture/caching) in the Streamlit documentation.

Ensure thread-safety

When using a container runtime, your app code must be thread-safe to handle multiple viewers concurrently. While each viewer gets a unique instance of the app script, you should review any imported code for shared state or global variables that could lead to race conditions or inconsistent behavior. If you introduce new threads into a Streamlit app, review Streamlit’s architecture and don’t use Streamlit commands in your custom threads.

See: Multithreading in Streamlit (https://docs.streamlit.io/develop/concepts/design/multithreading) in the Streamlit documentation.

Replace _snowflake usage with native Python equivalents

_snowflake is a private module that is only available in user-defined functions (UDFs) and stored procedures. Warehouse runtimes inherit access to _snowflake, but container runtimes don’t. If your app uses _snowflake, replace it with native Python equivalents, such as the Snowflake Python Connector. If needed, use stored procedures to access secrets.

See: Manage secrets and configure your Streamlit app

Update file paths and organization

The root of your source location is the working directory for your app. For most Python libraries, your app will need to use relative paths from the root of your source location. However, some Streamlit commands require paths relative to the entrypoint file. If your entrypoint file is in a subdirectory, check the paths in your code accordingly.

Verify secrets.toml and config.toml locations.

See: Organize your Streamlit app files

App configuration changes¶

Alter your app to set its compute pool, query warehouse, and runtime

When you are ready to switch the runtime type of your app, you can use Snowsight or SQL.

SnowsightSQL

1. Sign in to Snowsight.

2. In the navigation menu, select Projects » Streamlit, and then select your app.

3. In the upper-right corner, select the vertical ellipsis menu, and then select App settings.

4. For the Python environment, select Run on container.

5. In the Compute pool dropdown, select your compute pool.

6. In the Query warehouse dropdown, select your query warehouse.

7. To save your changes and close the dialog, select Save.

Your app will take a couple minutes to reboot and build its new container.

See: Runtime environments for Streamlit apps