Question 310
A SQL change-tracking stream should trigger an Azure Function that calls an approved embedding model. Initial testing shows: Replayed notifications do not duplicate active embeddings and updates become searchable. Which limitation should the team still consider?
Explanation
Correct Answer
Model throttling and failures can delay freshness. Azure Functions with SQL trigger binding can react to database changes but may retry deliveries. Create an idempotent function that updates versioned embeddings for changed records.
Incorrect Answers
Binding deliveries can be retried; stable document/version keys and idempotent upsert prevent duplicate active embeddings.
Model calls need source key/version correlation and validation before a vector is attached to a record.
Change Tracking retention has limits; expired tokens require controlled full reconciliation or rebuild.