COF-C03 Question 85
Single answerDifferent teamsA company is onboarding three teams to Snowflake: the Data Engineering team loads raw data and creates curated tables, the Business Intelligence team runs dashboards during business hours, and the Data Science team performs ad hoc model training that can consume significant compute at unpredictable times. The company wants to minimize contention between teams, keep costs attributable to each team, and allow each team to manage its own compute usage without affecting others. Which Snowflake design is the BEST fit?
- A
Use one shared virtual warehouse for all teams, and separate access with roles and schemas
- B
Create a separate virtual warehouse for each team, and grant each team usage on only its own warehouse
- C
Place each team in a separate database, but keep all teams on the same virtual warehouse to simplify administration
- D
Create separate resource monitors for each schema so compute usage can be controlled independently by team
Show answer and explanation
Correct answer: B
Explanation
The best design is to create separate virtual warehouses for different teams when the goal is compute isolation, independent performance management, and clearer cost attribution. In Snowflake, storage and compute are decoupled. Databases and schemas organize and secure data objects, while virtual warehouses provide compute for SQL workloads. Because warehouses are independent, one team's ETL, dashboard, or data science jobs can run without directly consuming the compute resources of another team's warehouse. This is a common best practice for organizations with distinct workload patterns across teams. Snowflake documentation on virtual warehouses and resource monitors supports this approach: warehouses isolate compute, and resource monitors can then be used per warehouse to help control or track spend.
- A. Incorrect.
Incorrect. Roles and schemas help with access control and object organization, but they do not isolate compute resources. If all teams share one virtual warehouse, heavy workloads from one team can compete with and slow down workloads from another team. This also makes cost attribution by team more difficult because all compute is consumed from the same warehouse.
- B. Correct.
Correct. In Snowflake, virtual warehouses are independent compute clusters. Assigning a separate warehouse to each team isolates compute so one team's workload does not directly contend with another team's workload. This also supports chargeback or showback because warehouse consumption can be tracked per team. Each team can size, suspend, resume, or scale its own warehouse according to its workload characteristics.
- C. Incorrect.
Incorrect. Databases separate data objects, not compute. Even if each team has its own database, using the same virtual warehouse means queries still compete for the same compute resources. This does not meet the requirement to let teams manage compute independently or to reduce cross-team contention.
- D. Incorrect.
Incorrect. Resource monitors are applied to warehouses, not schemas. Schemas are logical containers for database objects and do not consume compute directly. This option reflects a common misconception that storage object boundaries can control compute consumption. To control and monitor compute by team, the company should separate warehouses and optionally apply resource monitors to those warehouses.