Understanding clean room table policies¶

Clean rooms can implement data policies to control how data can be used by collaborators. These are in addition to any Snowflake table policies set on the underlying tables linked into the clean room.

Each collaborator in a clean room can set policies on their own data. Your policies are enforced only in requests from other users; your policies are not enforced against your own requests. For example, if your join policy allows joins against only column A, other users are restricted to joining on column A, but you can run joins against any of your columns.

Clean room policies can be set using either the clean room API or UI.

To implement policy checks, the following must be true:

The data owner must set a policy in their clean room. You can set policies using either the API or the UI. Each policy type is set separately. Clean room natively implements column policies, row policies, and activation policies. Clean room policies are not additive: When you set a clean room policy, all previous values are deleted.

-- Sets a join policy on column HASHED_EMAIL.
CALL samooha_by_snowflake_local_db.provider.set_join_policy(
  'my_provider_cleanroom',
  ['my_db.my_sch.T1:HASHED_EMAIL']);

-- Replaces the previous join policy. Now the only column in the join policy is AGE_BND.
CALL samooha_by_snowflake_local_db.provider.set_join_policy(
  'my_provider_cleanroom',
  ['my_db.my_sch.T1:AGE_BAND']);

Copy

The template must check the policy in the appropriate place in the template. A clean room policy is checked only if it has the appropriate policy filter applied to the column in the template. If you set a clean room policy to protect your data, you should examine the template to confirm that the template is enforcing your policies as you expect. The following template checks whether col1 is allowed by the data owner’s column policy:
```
SELECT
  IDENTIFIER( {{ col1 | column_policy }} )
FROM {{ source_table[0] }} AS c;
```
Copy
The following template does not check whether col1 has a clean room policy:
```
SELECT
  IDENTIFIER( {{ col1 }})
FROM {{ source_table[0] }} AS c;
```
Copy
Clean rooms supports a different template filter for each policy type. However, the semantics of the filter are not checked, only whether the column is in the policy for that filter type. For example, in the following snippet, the join policy is checked for col1, even though the column is not being joined against. If col1 is in the data owner’s join policy, the query can succeed; if col1 is not in the data owner’s join policy, the query will be blocked.
```
SELECT
  IDENTIFIER( {{ col1 | join_policy }})
FROM {{ source_table[0] }} AS c;
```
Copy

Note

Column policy checks are carried out when the template JinjaSQL is parsed. Queries with wildcards might not be caught using these checks, and discretion should be used when designing an analysis template. If some columns should really never be queried, consider creating a view of your source table that eliminates these sensitive columns, and link in that view instead.

Snowflake policies in clean rooms¶

When you link tables into a clean room, any Snowflake table policies on the source tables are enforced in the linked tables in the clean room, but these policies aren’t necessarily reported by the clean room API or UI. For instance, a Snowflake join policy continues to be enforced in the clean room, but that join policy is not visible by calling consumer.view_provider_join_policy or consumer.view_join_policy. Therefore, you should either remove policies from the underlying linked tables, create equivalent clean room policies (when they exist), or communicate the existence of these policies clearly to your collaborators so that their queries don’t fail or behave unexpectedly (“why can’t I join on this column?”).

Any changes to Snowflake policies in the source tables are automatically propagated to the linked views in the clean room.

Snowflake privacy policies prevent creation of a view from a protected table, so you cannot link in tables that have privacy policies.

Join policies¶

Set a join policy to indicate which columns in your data can be joined on by any template in the clean room. (Snowflake join policies, in contrast, specify which columns must be joined on.) Join policies apply to all templates in the clean room.

A column cannot be in both a join policy and a column policy, but a column can be in both a join policy and an activation policy.

Implementing a join policy¶

Clean room join policies are enforced against a column if the template applies the join_policy or join_and_column_policy filter to the column.

If a template checks a join policy for a column, and the clean room has no join policies set, or the column is not in the join policy, the query will be blocked.

The following code shows how to set join policies as a provider or a consumer. Remember that policies are only enforced against queries run by another account.

-- Set join policies on two columns in a clean room where you are a provider.
CALL samooha_by_snowflake_local_db.provider.set_join_policy(
  'my_provider_cleanroom',
  ['SAMOOHA_SAMPLE_DATABASE.DEMO.CUSTOMERS:HASHED_EMAIL', 'MYDB.MYSCH.EXPOSURES:HASHED_EMAIL']);

-- Set join policies on two columns in a clean room where you are a consumer.
CALL samooha_by_snowflake_local_db.consumer.set_join_policy(
  'my_consumer_cleanroom',
  ['SAMOOHA_SAMPLE_DATABASE.DEMO.CUSTOMERS:HASHED_EMAIL', 'MYDB.MYSCH.EXPOSURES:HASHED_EMAIL']);

Copy

The following procedures are used to view or manage join policies in code:

consumer.set_join_policy
consumer.view_provider_join_policy
consumer.view_join_policy
provider.view_join_policy
provider.set_join_policy

Column policies¶

Set a column policy to indicate which of your columns can be projected in analysis results from a specific template. Column policies are applied to specific templates in a specific clean room.

A column cannot be in both a join and a column policy. A column can be in both an activation and a column policy.

Implementing a column policy¶

Clean room column policies are enforced against a column only if the template uses the column_policy or join_and_column_policy filter.

If a clean room checks a column policy for a column, and the column is not in the column policy, or the clean room has no column policies, the query will be blocked.

The following code shows how to set column policies for three columns when accessed by the prod_overlap_analysis template. The example shows how to set the policy both as a provider and a consumer. Remember that policies are only enforced against queries run by another account.

-- Set column policy on prod_overlap_analysis template in a clean room where
-- you are a provider.
call samooha_by_snowflake_local_db.provider.set_column_policy(
  'my_provider_cleanroom',
  ['prod_overlap_analysis:SAMOOHA_SAMPLE_DATABASE.DEMO.CUSTOMERS:STATUS',
   'prod_overlap_analysis:SAMOOHA_SAMPLE_DATABASE.DEMO.CUSTOMERS:AGE_BAND',
   'prod_overlap_analysis:SAMOOHA_SAMPLE_DATABASE.DEMO.CUSTOMERS:DAYS_ACTIVE']);

-- Set column policy on prod_overlap_analysis template in a clean room where
-- you are a consumer.
call samooha_by_snowflake_local_db.consumer.set_column_policy(
  'my_consumer_cleanroom',
  ['prod_overlap_analysis:SAMOOHA_SAMPLE_DATABASE.DEMO.CUSTOMERS:STATUS',
   'prod_overlap_analysis:SAMOOHA_SAMPLE_DATABASE.DEMO.CUSTOMERS:AGE_BAND',
   'prod_overlap_analysis:SAMOOHA_SAMPLE_DATABASE.DEMO.CUSTOMERS:DAYS_ACTIVE']);

Copy

The following procedures are used to view or manage column policies in code:

consumer.set_column_policy
consumer.view_column_policy
consumer.view_provider_column_policy
provider.set_column_policy
provider.view_column_policy

Activation policies¶

Set an activation policy to indicate which of your columns can be activated by an activation template. Activation saves query results to a table in the Snowflake account of the provider or consumer, or to a third-party activation connector.

A column can be part of an activation policy as well as any other policy.

Implementing an activation policy¶

Activation policies can be set in the clean rooms UI if the template allows activation.

Activation policies are set for a specific column in a specific template.

Activation policies are enforced against a column only if the template applies the activation_policy filter to the column.

The following code demonstrates setting an activation policy to allow the HASHED_EMAIL and REGION_CODE columns to be activated in a clean room. This policy affects all users and all activation templates in the clean room. There are equivalent procedures for providers and consumers in a clean room. Call the procedure that reflects your role in the clean room.

-- Set activation policy on prod_overlap_analysis template in a clean room where you are a provider
call samooha_by_snowflake_local_db.provider.set_activation_policy('my_cleanroom', [
    'prod_overlap_analysis:SAMOOHA_SAMPLE_DATABASE.DEMO.CUSTOMERS:HASHED_EMAIL',
    'prod_overlap_analysis:SAMOOHA_SAMPLE_DATABASE.DEMO.CUSTOMERS:REGION_CODE' ]);

-- Set activation policy on prod_overlap_analysis template in a clean room where you are a consumer
call samooha_by_snowflake_local_db.consumer.set_activation_policy('my_cleanroom', [
    'prod_overlap_analysis:SAMOOHA_SAMPLE_DATABASE_NAME.DEMO.CUSTOMERS:HASHED_EMAIL',
    'prod_overlap_analysis:SAMOOHA_SAMPLE_DATABASE_NAME.DEMO.CUSTOMERS:REGION_CODE' ]);

Copy

The following procedures are used to manage activation policies in code:

consumer.set_activation_policy
provider.set_activation_policy

Aggregation policies¶

Aggregation policies require that all queries against a table contain aggregations (GROUP BY, COUNT, and other functions), and also specify a minimum number of rows per result group, or the group will be omitted from the results.

Clean rooms do not have their own implementation of aggregation policies; to apply aggregation constraints on your linked data, either apply an aggregation policy on the source table, or implement aggregation constraints in your template.

Some Snowflake-provided templates use the threshold and threshold_value parameters set for a user or template. These values can be modified in the clean rooms UI, or by calling provider.add_consumers or provider/consumer.set_privacy. If set for a consumer, you can access these values in your template.