SnowPro Specialty: Gen AI Question 206
Single answerCreate compute poolA data engineering team is preparing to deploy a containerized embedding service in Snowflake using Snowpark Container Services. The service will run continuously, but demand varies during the day. The security team also requires the deployment to use only approved instance families and to avoid overprovisioning costly GPU resources when they are not needed. When creating the compute pool, which action best meets these requirements?
- A
Create the compute pool with a permitted CPU-based instance family and configure MIN_NODES lower than MAX_NODES so the pool can scale within approved limits.
- B
Create the compute pool with a GPU instance family and set MIN_NODES equal to MAX_NODES to guarantee consistent performance at all times.
- C
Create the compute pool without specifying instance family restrictions, because services define hardware selection at runtime.
- D
Create the compute pool with a large fixed number of nodes and rely on warehouse auto-suspend behavior to reduce costs when traffic drops.
Show answer and explanation
Correct answer: A
Explanation
In Snowpark Container Services, a compute pool provides the infrastructure on which containerized services and jobs run. When you create a compute pool, you choose an instance family and define scaling boundaries such as minimum and maximum nodes. For a workload with variable demand, using a lower MIN_NODES and higher MAX_NODES enables scaling while controlling cost. If the workload does not require GPUs, selecting a CPU-based approved instance family avoids unnecessary expense and aligns with governance requirements. This reflects Snowflake best practices for right-sizing compute pools and using instance families intentionally rather than treating them like warehouses or leaving hardware choice to the service definition.
- A. Correct.
Correct. A compute pool is created with an instance family and node limits. Choosing an approved CPU-based instance family satisfies the security requirement, and setting MIN_NODES lower than MAX_NODES allows the pool to scale elastically for variable demand. This is the most cost-conscious and policy-aligned approach when GPUs are not required.
- B. Incorrect.
Incorrect. GPU resources may be appropriate for some inference workloads, but the scenario specifically emphasizes avoiding unnecessary GPU cost. Setting MIN_NODES equal to MAX_NODES also removes elasticity, which conflicts with the requirement for variable daily demand.
- C. Incorrect.
Incorrect. Instance family selection is part of compute pool creation, not something left entirely to the service at runtime. The compute pool defines the infrastructure available to services, so omitting this concept would not satisfy the requirement to restrict deployments to approved hardware families.
- D. Incorrect.
Incorrect. Compute pools are not managed like virtual warehouses, and warehouse auto-suspend is not the mechanism used to control compute pool cost. A large fixed pool would also directly conflict with the requirement to avoid overprovisioning.