Compute Group Management: Creation, Authorization, and Scaling

Compute Group (called "Compute Cluster" in versions before 3.0.2) is the mechanism for physically isolating different workloads in the storage-compute separation architecture. One or more BE nodes form a Compute Group, and multiple Compute Groups access the same data through a shared storage layer.

Key characteristics:

• BE nodes are locally stateless, and data is stored on shared storage.
• Multiple Compute Groups share the same data without additional replicas.
• Adding or removing a Compute Group requires no data migration, only cache warm-up at query time.

Compared with Resource Group, Compute Group has the following advantages:

Dimension	Compute Group	Resource Group
Storage cost	Data resides in shared storage. The number of Compute Groups is not limited by replica count, and cost does not increase with the number of groups.	Adding replicas means storage cost grows linearly.
Scaling flexibility	Adding a new Compute Group requires only cache warm-up, no data migration.	Adding a replica requires migrating a large amount of data.
Isolation completeness	The shared storage layer guarantees multiple replicas. A BE outage within a single Compute Group does not affect data loading.	A BE outage may cause data loading to fail.

Note

In versions before 3.0.2, Compute Group is called Compute Cluster.

View Compute Groups

View all Compute Groups under the current warehouse:

SHOW COMPUTE GROUPS;

Add a Compute Group

Use the ADD BACKEND command to add a BE node and specify the Compute Group it belongs to.

Specify a Compute Group:

ALTER SYSTEM ADD BACKEND 'host:9050' PROPERTIES ("tag.compute_group_name" = "new_group");

Without specifying a Compute Group (joins default_compute_group by default):

ALTER SYSTEM ADD BACKEND 'host:9050';

Manage Compute Group Access Permissions

Grant access permission

Grant the usage permission of a specified Compute Group to a user:

GRANT USAGE_PRIV ON COMPUTE GROUP {compute_group_name} TO {user};

Revoke access permission

Revoke a user's usage permission on a specified Compute Group:

REVOKE USAGE_PRIV ON COMPUTE GROUP {compute_group_name} FROM {user};

Set the Default Compute Group

Set and view the default Compute Group

Operation	Command	Permission requirement
Set the default Compute Group for the current user	SET PROPERTY 'default_compute_group' = '{clusterName}';	No additional permission required
Set the default Compute Group for another user	SET PROPERTY FOR {user} 'default_compute_group' = '{clusterName}';	Requires Admin permission
View the default Compute Group of the current user	SHOW PROPERTY;	No additional permission required
View the default Compute Group of another user	SHOW PROPERTY FOR {user};	Requires the relevant view permission

The value of the default_compute_group field in the returned result is the current default Compute Group.

Permission description

Admin user (for example, CREATE USER jack IDENTIFIED BY '123456' DEFAULT ROLE "admin"):

• Can set the default Compute Group for themselves and for other users.
• Can view the PROPERTY of themselves and of other users.

Regular user (for example, CREATE USER jack1 IDENTIFIED BY '123456'):

• Can only set the default Compute Group for themselves.
• Can only view their own PROPERTY.
• Cannot execute SHOW COMPUTE GROUPS (the operation requires the GRANT ADMIN permission).

FAQ

Q: An error occurs when performing data reads or writes

The current user has not configured a default Compute Group. Execute use @cluster to specify the Compute Group for the current session, or use SET PROPERTY to set a default value.

Q: A default Compute Group has been set, but reads and writes still report errors

The previously specified Compute Group has been deleted. Execute use @cluster to specify it again, or use SET PROPERTY to update the default setting.

Automatic Selection of the Default Compute Group

When a user has not explicitly set a default Compute Group, the system automatically selects one that meets the following conditions:

• Has Active BE nodes.
• The current user has usage permission on it.

Within the same session, the default Compute Group remains unchanged. Across sessions, the following situations cause the system to re-select automatically:

Trigger condition	Always changes
The user loses usage permission on the previously selected Compute Group	Always changes
A Compute Group is added or removed	Always changes
The previously selected Compute Group no longer has Active BE nodes	May change

Switch Compute Groups

In the storage-compute separation architecture, you can specify the database and the Compute Group in the same statement:

USE { [catalog_name.]database_name[@compute_group_name] | @compute_group_name }

If a database name or Compute Group name contains a reserved keyword, enclose the corresponding name with backticks (`).

Compute Group Scaling

Use the addition or removal of BE nodes to elastically scale a Compute Group:

• Scale out: ALTER SYSTEM ADD BACKEND adds a new BE to the specified Compute Group.
• Scale in: ALTER SYSTEM DECOMMISSION BACKEND decommissions a BE from the Compute Group.

For detailed operations, see Storage-compute separation operations.