Add a compute pool to an app with containers¶

About compute pools in apps with containers¶

A compute pool is a collection of one or more virtual machine (VM) nodes on which Snowflake runs Snowpark Container Services. Apps with containers uses a compute pool in the consumer account to manage the container images required by the app.

An app can create multiple compute pools and each compute pool is exclusive to the app. Compute pools used by the app cannot be used for other purposes.

Containers within an app can directly access each other, even if they are in different compute pools.

However, using different compute pools allows providers to separate types of services. For example, a provider can separate their frontend services from backend services.

Compute pools are account-level objects, meaning the name of each compute pool must be unique within the consumer account.

Best practices for using compute pools in an app with containers¶

Providers should consider the following best practices when creating compute pools a consumer account:

• Compute pools have cost implications. It is important to set values for the min_nodes, max_nodes, and instance_family properties to consume the correct amount of resources. Providers should also set the AUTO_SUSPEND_SECS property to automatically suspend inactive compute pools.

  See CREATE COMPUTE POOL for more information.

• Compute pools are account-level objects, as such their pool names must be unique within a consumer account. Consider using the application name as a prefix of the compute pool name to ensure uniqueness.

• When adding a computer pool to an app with containers that is installed on different cloud service providers, the code used to create the compute pool must account for differences in the instance families across different cloud service providers. For example, the HIGHMEM_X64_L instance family has a different configuration for each cloud service provider.

  See CREATE COMPUTE POOL for more information on available instance families. See Choose different instance families for each provider for an example of how to set the instance family for different cloud service providers.

• Set the uses_gpu property to TRUE only if the app with containers uses a GPU as the instance family of the compute pool. See Set the uses_gpu property in the manifest file for more information.

Create a compute pool for an app¶

There are two ways to create a compute pool for an app with containers:

• The app creates the compute pools required during installation. This requires that the consumer grants the CREATE COMPUTE POOL privilege on the compute pool to the app. A provider can configure the app to request these privileges using Snowsight.

  See Configure an app to request the CREATE COMPUTE POOL privilege for more information.

• The consumer manually creates the compute pools required by the app. The consumer must run the CREATE COMPUTE POOL to create the compute pool, then manually grant the CREATE COMPUTE POOL privilege on the compute pool to the app.

Set the uses_gpu property in the manifest file¶

If an app with containers specifies a GPU as the instance family for the Compute Pool, providers must set the uses_gpu flag to true in the manifest. The following example shows how to set this flag in the artifacts block:

artifacts:
  readme: readme.md
  setup_script: scripts/setup.sql
  container_services:
    uses_gpu: true|false
    images:
    - /provider_db/provider_schema/provider_repo/server:prod
    - /provider_db/provider_schema/provider_repo/web:1.0

The automated security scan uses this flag security scanning framework to validate behavior during the app version scanning process.

Configure an app to request the CREATE COMPUTE POOL privilege¶

Providers can configure an app to request the CREATE COMPUTE POOL privilege. They can also create the compute pool from the setup script when the app is installed or upgraded.

...
privileges:
 - CREATE COMPUTE POOL
   description: "Enable application to create one to five compute pools"
 ...

CREATE COMPUTE POOL IF NOT EXISTS app_compute_pool
  MIN_NODES = 1
  MAX_NODES = 1
  INSTANCE_FAMILY = standard_1
  AUTO_RESUME = true;

LET POOL_NAME := (select current_database()) || '_app_pool';
CREATE COMPUTE POOL IF NOT EXISTS identifier(:pool_name)
  MIN_NODES = 1
  MAX_NODES = 1
  INSTANCE_FAMILY = STANDARD_2;

Choose different instance families for each provider¶

When creating a compute pool for an app that is published across multiple cloud service providers, the code that creates the setup script must be written to account for differences in how instance families are configured.

The following example shows how to write a stored procedure to create a compute pool based on the cloud service provider where the app is being installed:

 CREATE OR REPLACE PROCEDURE public.create_cp()
 RETURNS VARCHAR
 LANGUAGE SQL
 EXECUTE AS OWNER
 AS $$
  BEGIN
      LET POOL_NAME := (select current_database()) || '_app_pool';
      LET INSTANCE_FAMILY := IFF( CONTAINS(current_region(), 'AZURE') , 'GPU_NV_XS' , 'GPU_NV_S' );
      CREATE COMPUTE POOL IF NOT EXISTS identifier(:pool_name)
          MIN_NODES = 1
          MAX_NODES = 1
          INSTANCE_FAMILY = :instance_family;
      RETURN 'Compute Pool Created';
  END;
$$;