Glossary of Snowflake Data Clean Room terms¶

Get to know these terms as they are used in Snowflake Data Clean Rooms. Some terms are used differently here than they are used in the rest of Snowflake.

Activate / activation: Exporting the results table of a query out of the clean room, typically to a third party. If allowed by the other party and the clean rooms settings, you can export query results to your own account or to an approved third-party partner, such as Google Ads or Meta Ads Manager.
Collaborator: Typically used as a synonym for consumer, but can also be used to mean anyone who can use a clean room, either as a provider or as a consumer.
Column policy: The columns in a clean room table that are allowed to be projected in query results for a specific template in a specific clean room. The policy is specified by the person who linked the data into the clean room. A clean room column policy is determined entirely within a clean room, and is not derived from any policies that might be applied to the source table outside of the clean room.
Connector: A plug-in that provides extra functionality to a clean room. Snowflake clean rooms can access a number of connectors to provide additional functionality, including clean room connectors (to allow usage of clean rooms created on other platforms), identity connectors (which can help identify similar entities in a dataset so they can be used in joins – for example, multiple email addresses that identify the same person), and external data connectors (which enable importing data from non-Snowflake sources). The clean room administrator determines which connectors are available to clean rooms in that Snowflake account.
Consumer: A person or account invited to use a clean room by the clean room provider. Consumers typically import their own data and run one or more queries supported by that clean room. However, a clean room can be configured to allow consumers to propose their own query, subject to approval by the provider.
Dataset: A table or view that has been linked (imported) into a clean room.
Differential privacy: An algorithmic and mathematical system that adds protection to individual rows or entities in a dataset by adding noise to numerical results and requiring grouping in queries to prevent exact values from being associated with exact rows or entities in the data.
Join policy: The columns in a clean room that are allowed to be used as join columns in a template in the designated clean room. The join policy is specified by the person linking the table into a clean room. A clean room join policy is entirely independent of Snowflake join policies, which are stripped when data is imported into a clean room.
Linking: Importing a protected view of data into a clean room. The provider and consumer can both link their own data into a clean room to make it available to any queries supported by that clean room. Linking a table or view means creating a copy (a view) of the source data within the clean room, dynamically linked to the source table or view outside of the clean room. Note that all Snowflake join, aggregation, privacy, and other policies are stripped from the linked table. If you want to enforce join or other policies, you need to add them using supported clean room functionality.
Provider: A clean room creator. The provider typically provides some data and the list of permitted queries that can be run in that clean room, and sets high-level clean room configurations.
Template: Each clean room has one or more templates, which are SQL queries written in JinjaSQL to allow them to be configurable by the consumer at run time. Each template resolves to a single SQL query. Snowflake provides a set of templates in the web app for common actions, but you can create your own custom templates using the API.
Secure view: When you link a table or view into the clean room, a secure view is created. This is an encrypted view based on the source table or view outside the clean room. The secure view is generally invisible to you, but might sometimes appear in an error message or when you are browsing the database objects using various tools, where you will see some name mangling from the original linked dataset. Unless directed otherwise, always refer to your data using the dataset name, which is identical to the linked source table or view.
Versioning: Each clean room has a version that is incremented whenever Python code is imported into the clean room by the provider. Consumers do not need to think about versioning. Providers need to remember to update the default clean room version every time they change any Python code within a clean room.
Web app: The clean room environment supports a browser-based web application you can use to manage the Snowflake clean rooms environment, create new clean rooms, or use clean rooms to which you have been invited.