SnowPro Specialty: Gen AI Question 177
Single answerUpdate parametersA data engineering team is using Snowflake Cortex Search to power a support assistant. After several weeks, users report that newly added knowledge-base articles are not appearing in search results quickly enough. The search service was created with an infrequent refresh cadence to reduce compute usage. The team wants to improve freshness without recreating the service. Which action should they take?
- A
Use ALTER CORTEX SEARCH SERVICE ... SET TARGET_LAG = '
' to reduce how far behind the source data the service is allowed to be - B
Use ALTER CORTEX SEARCH SERVICE ... SET WAREHOUSE_SIZE = 'LARGE' to force immediate reindexing whenever source tables change
- C
Use ALTER CORTEX SEARCH SERVICE ... REFRESH NOW to permanently change the update behavior for future source data changes
- D
Drop and recreate the Cortex Search service with a new embedding model, because update cadence cannot be changed after creation
Show answer and explanation
Correct answer: A
Explanation
This scenario tests whether the candidate understands how to modify update-related parameters on an existing Gen AI service rather than treating refresh behavior as fixed at creation time. In Snowflake Cortex Search, parameters such as TARGET_LAG are used to express freshness expectations for keeping the search index synchronized with underlying source data. When users need newer content to appear in results more quickly, the correct action is to update the service parameter to a shorter lag using ALTER CORTEX SEARCH SERVICE. Best practice is to tune freshness requirements thoughtfully, because lower lag targets can increase resource consumption. This aligns with Snowflake guidance to manage service behavior through supported ALTER operations and parameter updates instead of unnecessary service recreation.
- A. Correct.
Correct. For Cortex Search, update freshness is controlled through service parameters such as TARGET_LAG. Reducing TARGET_LAG tells Snowflake to keep the search index closer to the source data, improving how quickly newly ingested or modified records become searchable. This is the appropriate way to update the service's refresh behavior without recreating it.
- B. Incorrect.
Incorrect. Increasing warehouse size may provide more resources for maintenance work, but it does not by itself define the service's freshness objective or guarantee immediate reindexing on every data change. The misconception is confusing compute capacity with the update parameter that governs acceptable data staleness.
- C. Incorrect.
Incorrect. A manual refresh, where supported operationally, addresses a one-time synchronization need rather than permanently changing the ongoing update policy. The scenario requires changing future behavior, so the team should update the relevant service parameter instead of relying on ad hoc refreshes.
- D. Incorrect.
Incorrect. Recreating the service is unnecessary for this requirement. Update-related parameters can be modified after creation using ALTER statements. Changing the embedding model is unrelated to the problem of stale search results and would introduce unnecessary operational risk and cost.