Snowflake SPCS now supports backup instance types for compute pools

Imagine you need to run services in Snowpark Container Services, but the primary machine size is capacity-constrained in your region. With this feature, you can define an ordered list of fallback instance families so Snowflake automatically retries provisioning without failing.

Use cases

• Prevent insufficient capacity errors (ICE) when provisioning nodes.
• Configure a transparent fallback order for compute pools.
• Bill backup nodes according to the rate of the specific family in use.

How to use

Configure the fallback list using the BACKUP_INSTANCE_FAMILIES parameter when creating or altering a compute pool.

CREATE COMPUTE POOL my_compute_pool
  MIN_NODES = 1
  MAX_NODES = 10
  INSTANCE_FAMILY = GEN_X64_G2_4
  BACKUP_INSTANCE_FAMILIES = ('GEN_X64_G2_8', 'GEN_X64_G2_16');

What happens in this code:

1. Snowflake attempts to create nodes using the primary GEN_X64_G2_4 family.
2. If capacity is constrained on the primary family, the system tries the first fallback (GEN_X64_G2_8).
3. If needed, Snowflake proceeds to the next alternative in the list (GEN_X64_G2_16).

To alter an existing compute pool:

ALTER COMPUTE POOL my_compute_pool
  SET BACKUP_INSTANCE_FAMILIES = ('GEN_X64_G2_X');

What happens in this code:

1. The command updates the existing compute pool with the new fallback configuration.
2. Snowflake will use the specified family if capacity constraints occur on the primary type.

Limitations and requirements

• Generally available (GA).
• The backup list must contain at least one entry and must not include the primary family or duplicates.
• Users must ensure hardware, CPU architecture, accelerator (GPU), and service density compatibility between the primary and backup families.