SnowPro Specialty: Gen AI Question 267
Single answerCORTEX_FUNCTIONS_USAGE_HISTORY viewA GenAI platform team uses Snowflake Cortex functions in several production pipelines and needs to produce a weekly governance report showing which Cortex functions were called, when they were called, and how much usage was generated by each workload. They want to avoid parsing ad hoc query text from QUERY_HISTORY and instead use the system-provided usage view designed for Cortex activity. Which approach best meets this requirement?
- A
Query the CORTEX_FUNCTIONS_USAGE_HISTORY view and aggregate usage by function and time period for the reporting window.
- B
Query WAREHOUSE_METERING_HISTORY because Cortex function usage is fully represented there at the individual function-call level.
- C
Query QUERY_HISTORY only, because Cortex function invocations are not exposed through account usage views.
- D
Create an event table and rely on task history, because Snowflake does not provide a dedicated history view for Cortex function usage.
Show answer and explanation
Correct answer: A
Explanation
The key requirement is to use the Snowflake-provided metadata source intended for historical Cortex usage reporting. CORTEX_FUNCTIONS_USAGE_HISTORY is designed for this purpose and is the best fit for governance, chargeback, and trend analysis of Cortex function activity. In contrast, QUERY_HISTORY is useful for SQL troubleshooting but requires parsing statements and is not the most direct or reliable source for Cortex usage reporting. Likewise, WAREHOUSE_METERING_HISTORY focuses on warehouse compute rather than Cortex function usage details, and operational objects like event tables or task history do not substitute for the dedicated Cortex usage history view. This follows Snowflake best practice: use the most specific system view available for observability and governance rather than inferring usage indirectly.
- A. Correct.
Correct. The CORTEX_FUNCTIONS_USAGE_HISTORY view is the purpose-built system view for analyzing historical Cortex function activity. It is the most appropriate source when a team needs to report on Cortex function usage over time without reverse-engineering function calls from SQL text. This aligns with governance and monitoring use cases where you want structured usage records rather than inference from query logs.
- B. Incorrect.
Incorrect. WAREHOUSE_METERING_HISTORY is for warehouse compute consumption and cost/credit analysis, not a dedicated per-Cortex-function usage log. A candidate might choose this because both relate to monitoring usage, but warehouse metering does not replace the Cortex-specific history view when the requirement is to identify Cortex function activity by function and time period.
- C. Incorrect.
Incorrect. QUERY_HISTORY can sometimes help investigate SQL statements that invoked Cortex functions, but it is not the best answer when the requirement explicitly calls for the system-provided usage view designed for Cortex activity. Relying only on query text parsing is more brittle and less governance-friendly than using the dedicated Cortex usage history view.
- D. Incorrect.
Incorrect. Event tables and task history can support operational observability for certain workloads, but they do not replace the dedicated Snowflake account usage view for Cortex function usage. This option reflects a common misconception that teams must build their own telemetry pipeline even when Snowflake already exposes purpose-built monitoring metadata.