Packaged clean rooms

Preview

This feature is in Public Preview.

This page describes packaged clean rooms, a provider-consumer collaboration model for Databricks clean rooms.

In a packaged clean room, one collaborator is the provider and one is the consumer:

• The provider authors the notebooks and JARs and contributes data assets. The provider's code and data assets are hidden from the consumer.
• The consumer contributes their own data assets, triggers runs, and can view the run output. The consumer cannot view the provider's notebook or JAR code or the provider's data assets.

Neither collaborator can see the other's data assets. Either the provider or the consumer can create the clean room; whoever creates it designates which collaborator is the package provider. See Create clean rooms and Work with Databricks clean rooms as an invited collaborator.

This differs from an approval-based clean room, where all collaborators have equal privileges and can review every notebook before it runs. See How does Clean Rooms ensure a no-trust environment?. A packaged clean room is useful when a provider wants to offer a packaged analysis—similar to a private library—that a consumer can run against their own data without inspecting the provider's code.

How packaged clean rooms differ from approval-based clean rooms

The following table compares the capabilities of each collaborator in a packaged clean room with those in an approval-based clean room:

Capability	Approval-based	Packaged (consumer)	Packaged (provider)
View notebooks and JARs	All code visible	Names only, no code	Own notebooks and JARs
Add notebooks and JARs	Yes	No	Yes
View data assets	All visible	Own assets only	Own assets only
Add data assets	Yes	Yes	Yes
View run output	Yes	Yes	No
Access output tables	Runner only	Consumer (runner) only	No
Access shared output tables	Yes	Yes	Yes
Trigger runs	Yes	Yes	No
Auto-approval rules	Yes	No	No

Security and trust model

Because the consumer cannot inspect the provider's notebook or JAR code, the consumer must trust that the provider's code behaves as described, similar to using a private library.

The clean room's egress network policy controls whether the provider's code can reach the internet or specific external endpoints. If egress is open, or if it allows endpoints that the consumer is not comfortable with, the provider's code could potentially send the consumer's data outside the clean room.

Before you contribute sensitive data to a packaged clean room, review its egress network policy. See What is serverless egress control?.

Before you begin

• Packaged mode is set when the clean room is created and cannot be changed afterward.

Create a packaged clean room

Either the provider or the consumer can create a packaged clean room. The steps are the same as for an approval-based clean room, except that you select the packaged clean room type and designate the package provider.

1. In your Databricks workspace, click Catalog.
2. In the upper-right side, click Share > Clean Rooms.
3. Click Create Clean Room.
4. Under Clean room type, select Packaged Clean Room.
5. Enter a Clean Room name and select the Region for the central clean room.
6. Under Collaborator details, enter the Clean Room sharing identifier for the other collaborator, then click Add collaborator.
7. For Designated Clean Room package provider, select You or Invited collaborator to choose which collaborator is the package provider. The package provider's assets are hidden from the other collaborator.
8. Click Create clean room.

Packaged mode is set when the clean room is created and cannot be changed afterward.

Provider tasks

As the provider, you supply the notebooks, JARs, and data, and you can monitor the runs that the consumer triggers.

• Add notebooks, JARs, and data assets: Add your assets the same way you would in an approval-based clean room. See Step 3. Add data assets and notebooks to the clean room. Your assets are visible only to you, and you cannot see the consumer's data assets.

• Monitor run history: The consumer triggers runs that use your notebooks and JARs. You can see the consumer's runs in the run history, but you cannot view the run output. See Monitor clean room notebook runs.

Consumer tasks

As the consumer, you contribute your own data, trigger runs against the provider's code, and view the results.

• Open the packaged clean room: You see the names of the provider's notebooks and JARs, but not their source code.

• Add your data assets: Click Add Input Data to add your tables, volumes, or models to the clean room. Your data is visible only to you, and the provider cannot see it.

• Trigger a run: The provider's code runs in the clean room's secure environment with access to the data assets that both collaborators contributed.

• View the run output: You can view the output of the run. To make results available to others, the provider's code can also write to a shared output schema that all collaborators can read. See Create and work with output tables in Databricks Clean Rooms.

Limitations

• Packaged mode is set at clean room creation and cannot be disabled afterward.
• Auto-approval rules are disabled for all collaborators, including the provider.