An AI API repeatedly retrieves unchanged reference data and can tolerate a defined freshness window. Which pattern can reduce database load?

1. Treat Redis eviction as a guarantee of durable historical retention
2. Remove the authoritative database without evaluating durability
3. Cache-aside using Azure Managed Redis with appropriate expiry and invalidation
4. Cache every value forever with no invalidation