Scaling

Design an in-memory caching layer that sits in front of a primary data store and serves reads at sub-millisecond latency for tens of millions of requests per second. The cache must shard across many nodes, survive individual node loss without taking the database down with it, and offer a defensible answer to "what happens when a stale value is served."

Build it as a general-purpose cache — the same design backs feed metadata, session state, product catalogs, profile lookups, and other cache-backed resources in production systems. Out of scope: HTTP/CDN caching semantics (Cache-Control, ETag), the primary store's design, and disk-backed persistence (touched only in variants).