📚 Chapter 9: Advanced Caching Strategies

Minimize latency and load via multi-layer caches, robust population/update patterns, and sound invalidation tactics.

1. Multi-Layer Caching Architecture

Effective scaling uses a hierarchy of caches, each at different proximity to users and services.

Layer Purpose Key Strategy
CDN (Edge) Static assets; public read-only content close to end users. TTL-based expiry; provider cache purge/invalidation.
Reverse Proxy / Gateway Cache API responses before app logic (Nginx, Varnish, API GW). Honor `Cache-Control`, `ETag`, and conditional requests.
Distributed In-Memory Shared cache cluster (Redis/Memcached) for DB reads, sessions, profiles, computed objects. Scale reads; manage network latency and memory policies.

2. Cache Population and Update Patterns

A. Cache-Aside (Lazy Loading)

  1. Read: Check cache first.
  2. Hit: Return cached value.
  3. Miss: Fetch DB → write to cache → return.
  • Pros: Cache only what is needed; simple.
  • Cons: Miss latency; risk of stampede on popular keys; temporary inconsistency.

B. Write-Through

Synchronously write to cache and DB within the same operation.

  • Pros: Cache consistent at write time; simplifies reads.
  • Cons: Higher write latency; not ideal for heavy-write workloads.

C. Write-Back (Write-Behind)

Write to cache and acknowledge immediately; cache asynchronously persists to DB.

  • Pros: Very fast writes.
  • Cons: Risk of data loss on cache crash; requires journaling and recovery.

Strategy Trade-offs

Relative scores (1–5) for Read Latency, Write Latency, Consistency-at-Write, and Operational Complexity.

3. Cache Invalidation Tactics

A. Time-To-Live (TTL)

  • Mechanism: Automatic eviction after duration.
  • Trade-off: Data can be stale until TTL elapses.

B. Write-Invalidate

  • Mechanism: On DB write, explicitly delete affected cache keys.
  • Trade-off: Tight coupling; failure to invalidate leads to stale cache until TTL.

C. Change Data Capture (CDC)

  • Mechanism: Read transaction logs and publish events to update/invalidate cache.
  • Trade-off: Atomic and reliable, but adds tooling and ops complexity.

D. Stale-While-Revalidate

  • Mechanism: Serve stale data immediately while refreshing asynchronously.
  • Trade-off: Great UX; brief staleness window for some users.