SnowPro Specialty: Gen AI Question 268
Single answerCORTEX_FUNCTIONS_USAGE_HISTORY viewA GenAI platform team wants to monitor adoption and spending for Snowflake Cortex functions used by several applications. They need a queryable source that shows historical Cortex function invocations so they can build a dashboard of usage trends by function and time period. Which approach best meets this requirement?
- A
Query the SNOWFLAKE.ACCOUNT_USAGE.CORTEX_FUNCTIONS_USAGE_HISTORY view and aggregate rows by function name and time window.
- B
Use QUERY_HISTORY only, because it contains all Cortex cost and token details and is the recommended source for function-level usage reporting.
- C
Query WAREHOUSE_METERING_HISTORY, because Cortex function invocations are billed as warehouse compute and are best analyzed there.
- D
Use TASK_HISTORY to identify Cortex function calls, because Cortex requests are executed as background tasks and are exposed only through task metadata.
Show answer and explanation
Correct answer: A
Explanation
For historical monitoring of Snowflake Cortex usage, the best source is the SNOWFLAKE.ACCOUNT_USAGE.CORTEX_FUNCTIONS_USAGE_HISTORY view. In a real-world governance or FinOps scenario, teams use Account Usage views to build trend dashboards, attribute usage across workloads, and monitor adoption over time. QUERY_HISTORY can help correlate which SQL statements invoked Cortex functions, but it is not the dedicated usage-history view for this purpose. Likewise, WAREHOUSE_METERING_HISTORY is for warehouse consumption, and TASK_HISTORY is for tasks rather than Cortex invocation reporting. This aligns with Snowflake best practice: use the most specific ACCOUNT_USAGE view available for the workload you are monitoring, and use QUERY_HISTORY only as a complementary source when statement-level investigation is needed.
- A. Correct.
Correct. The CORTEX_FUNCTIONS_USAGE_HISTORY view in SNOWFLAKE.ACCOUNT_USAGE is the purpose-built Account Usage view for analyzing historical usage of Cortex functions. It is the appropriate source when a team needs to report on adoption, usage patterns, and related consumption over time by Cortex function. Aggregating this view by function and time period is the practical way to build a dashboard for historical monitoring.
- B. Incorrect.
Incorrect. QUERY_HISTORY is useful for analyzing executed SQL statements, troubleshooting, and finding statements that called Cortex functions, but it is not the dedicated source for Cortex function usage reporting. Relying on QUERY_HISTORY alone is a common mistake because it focuses on statements rather than the specialized usage history exposed for Cortex functions.
- C. Incorrect.
Incorrect. WAREHOUSE_METERING_HISTORY tracks warehouse compute consumption. Cortex functions are not analyzed through warehouse metering as if they were simply warehouse CPU usage. This distractor reflects the misconception that all Snowflake processing costs are best measured through warehouse billing views.
- D. Incorrect.
Incorrect. TASK_HISTORY is for Snowflake tasks. Cortex function invocations are not generally represented as background task executions, so TASK_HISTORY is not the right source for monitoring application usage of Cortex functions. This option confuses orchestration metadata with function usage telemetry.