Cache Invalidation

Three patterns, each with different consistency guarantees and complexity. Pick the right one before writing any code.

Phase 1: Discovery

Before recommending a pattern, establish:

What is being cached?

• Per-user vs shared data — shared data is harder; one user's write invalidates everyone's cache
• Aggregate/computed data (counts, totals, ranked lists) — these have fan-out invalidation problems
• Relational data — a single entity change may invalidate N cache keys that join against it