What best describes a cache-as-denormalized layer?

What best describes a cache-as-denormalized layer?

• A. A cache that stores an exact, always-updated mirror of the source data.
• B. A pattern where the source-of-truth database stays normalized while a cache holds a denormalized pre-computed representation for fast reads.
• C. A backup copy that duplicates the entire database for disaster recovery.
• D. A cache that only stores raw, non-aggregated data.

Answer: (B)

This question centers on the idea of using a cache as a denormalized layer to speed up reads. The cache stores a denormalized, pre-computed representation of data so that read requests can be served quickly without performing multiple joins or on-the-fly aggregations in the primary database. Meanwhile, the source of truth remains normalized, preserving data consistency and avoiding data duplication in the authoritative store. The benefit is clear: reads are fast because the heavy lifting—the denormalization and aggregation—has already been done and cached. The cache acts as a performance optimization, not as the primary data source. Because the cached data can become stale, updates to the underlying normalized data must propagate to the cache through invalidation, updates, or expiration policies, leading to eventual consistency. Why this differs from the other ideas: storing an exact, always-updated mirror would imply the cache is the source of truth and would largely defeat the purpose of denormalization for fast reads. A backup copy for disaster recovery is about data durability, not read-time performance. Storing only raw, non-aggregated data doesn’t provide the fast, joined/derived view that a denormalized cache is designed to deliver.

This question centers on the idea of using a cache as a denormalized layer to speed up reads. The cache stores a denormalized, pre-computed representation of data so that read requests can be served quickly without performing multiple joins or on-the-fly aggregations in the primary database. Meanwhile, the source of truth remains normalized, preserving data consistency and avoiding data duplication in the authoritative store.

The benefit is clear: reads are fast because the heavy lifting—the denormalization and aggregation—has already been done and cached. The cache acts as a performance optimization, not as the primary data source. Because the cached data can become stale, updates to the underlying normalized data must propagate to the cache through invalidation, updates, or expiration policies, leading to eventual consistency.

Why this differs from the other ideas: storing an exact, always-updated mirror would imply the cache is the source of truth and would largely defeat the purpose of denormalization for fast reads. A backup copy for disaster recovery is about data durability, not read-time performance. Storing only raw, non-aggregated data doesn’t provide the fast, joined/derived view that a denormalized cache is designed to deliver.