Caching layers: from CPU to CDN
How caches stack from L1/L2/L3 through RAM, application caches, and CDNs — and why each layer exists.
topic 10
Caching
Layers, invalidation, stampede protection, ETags.
Cache invalidation: the hardest problem in CS
Strategies for evicting stale entries — TTL, event-driven purge, write-through, write-behind — and why getting it wrong corrupts user state.
Cache stampede: when one TTL expires and thousands of requests hit the DB
Why a single TTL expiry turns a hot cache key into a flash-DDoS against the origin, and the four mechanisms — locks, single-flight, XFetch, and stale-while-revalidate — that keep the database alive.
- 1 What is a cache stampede and why it makes things worse
- 2 Lock and single-flight: bounding concurrent rebuilds
- 3 XFetch: coordination-free probabilistic early expiration
- 4 Stale-while-revalidate and CDN request coalescing
- 5 Detecting stampedes and designing TTL for production
- 6 Metastable failure, fencing tokens, and production postmortems
- 7 Cache stampede: multiple-choice review
- 8 Cache stampede: free-recall review
- 9 Cache stampede: code reading
- 10 Cache stampede: build and tame the herd
ETags: conditional requests and zero-byte responses
How entity tags enable conditional GET — the server returns 304 Not Modified when content unchanged, saving bandwidth and reducing latency.
Cache-Control: directing browsers and CDNs
The directives — max-age, s-maxage, no-store, stale-while-revalidate — that tell every cache in the chain exactly how long to keep a response.
Stale-while-revalidate: serve stale, refresh in background
How SWR decouples freshness from latency — serve the cached version immediately, then update it silently, eliminating the tail-latency spike of synchronous revalidation.
Dogpile effect: concurrent cache misses that crush the origin
When a popular key expires, all concurrent requests miss simultaneously and pile onto the database — the pattern, why it differs from stampede, and how mutex locks and probabilistic early expiry prevent it.
Caching system design: combining all layers
How to compose CDN, reverse-proxy, application, and database caches into a coherent strategy — choosing TTLs, invalidation triggers, and fallback paths that hold under real traffic.
You can see the whole topic now.
Every unit, from the underground to the orbit. Caching is one constellation in the atlas — climb another topic and watch the next one light up.