COF-C03 Question 274
Single answerGrouping similar workloadsA Snowflake administrator is reviewing warehouse usage for a shared analytics environment. BI dashboards run every few minutes with short, repeatable queries during business hours. Data scientists run ad hoc exploratory queries that can be long-running and resource-intensive. Overnight ELT jobs perform large batch transformations on a fixed schedule. Users are complaining that dashboard performance is inconsistent, especially when other activity is high. Which approach BEST aligns with Snowflake best practices for grouping similar workloads?
- A
Keep all workloads on one larger virtual warehouse and rely on auto-suspend and auto-resume to separate the resource usage patterns
- B
Place dashboards, ad hoc data science queries, and overnight ELT jobs on separate virtual warehouses based on workload characteristics and concurrency needs
- C
Move all workloads to a multi-cluster warehouse so Snowflake can automatically optimize every query type without any workload separation
- D
Create one warehouse for dashboards and one warehouse shared by ad hoc data science queries and overnight ELT jobs because both use more compute than dashboards
Show answer and explanation
Correct answer: B
Explanation
The best answer is to separate workloads by similar behavior and business requirements. In Snowflake, virtual warehouses provide compute isolation, so assigning different workload types to different warehouses is a core design pattern for performance consistency and workload management. BI/dashboard workloads often benefit from their own warehouse because they require steady, predictable response times and may need concurrency scaling via multi-cluster warehouses. ELT batch jobs usually run on a defined schedule and can be sized independently. Ad hoc data science workloads are often irregular and resource-intensive, so isolating them prevents unpredictable impact on user-facing analytics. This aligns with Snowflake guidance to use separate warehouses for different workload patterns and use warehouse sizing, auto-suspend/resume, and multi-cluster settings as complementary controls rather than substitutes for workload isolation.
- A. Incorrect.
Incorrect. Auto-suspend and auto-resume help control cost by stopping and starting compute when needed, but they do not isolate competing workloads that run at the same time. If BI, ad hoc analytics, and ELT all share one warehouse, they can still contend for resources and cause inconsistent dashboard performance.
- B. Correct.
Correct. Snowflake best practice is to group similar workloads on separate virtual warehouses so workloads with different performance and concurrency characteristics do not interfere with one another. Dashboards typically need predictable low-latency response, data science queries are often variable and long-running, and scheduled ELT jobs can be isolated to their own warehouse to improve performance management and cost visibility.
- C. Incorrect.
Incorrect. Multi-cluster warehouses primarily help with handling concurrency for many similar simultaneous queries, such as BI workloads. They do not automatically make all mixed workload patterns optimal. Long-running ELT and exploratory data science activity can still be better managed by isolating them on separate warehouses rather than combining everything into one multi-cluster warehouse.
- D. Incorrect.
Incorrect. This is better than putting everything on one warehouse, but it still groups two dissimilar workloads together. Ad hoc data science activity is unpredictable and interactive, while overnight ELT is batch-oriented and scheduled. Keeping them separate usually provides better isolation, sizing flexibility, scheduling control, and clearer chargeback or monitoring.