Home/Labs/Edge Cache Keys & Purge
All 280 Labs
INTERACTIVE LAB🗝️

Edge Cache Key & Purge Lab (Interactive)

Compose cache keys from Vary headers, watch fragmentation crater the hit ratio, then purge by Surrogate Key. Toggle cache-key components against a real LRU capacity model, then hard- or soft-purge a product tag and measure latency spikes and origin fetch storms.

Edge Cache Key & Purge Lab

Compose the cache key, watch Vary fragmentation destroy the hit ratio, then purge with Surrogate Keys.

Cache key = Hash(components)

Global invalidation (Surrogate-Key: product-102)

One tag purge invalidates product page + category tiles + search fragments in <150ms (Shopify/Fastly pattern).

Distinct cached variants3working set 6 MB vs 1,000 MB capacity → fits
Edge hit ratio100.0%

Healthy CDNs absorb 85–98% of traffic; below 50% the edge has become a pass-through.

Avg latency5.0ms
Origin fetches/s0

Cache-Control recipe

public, max-age=3600,

s-maxage=2592000, stale-while-revalidate=86400

max-age = browser; s-maxage = shared CDN edge.

How It Works Under the Hood

Edge nodes look up requests by a composite key — Hash(scheme, host, path, query, Vary headers) — not the URL alone. Every Vary dimension multiplies stored variants: Accept-Encoding triples them, while Vary: User-Agent with thousands of distinct strings shatters a 95% hit ratio into cache-busting fragments that thrash the PoP's finite NVMe capacity. When a price changes, Surrogate-Key tags let the backend purge every related page across 300 PoPs with one API call; a hard purge makes the next visitors eat full miss latency, whereas a soft purge marks entries stale so stale-while-revalidate keeps serving 5ms responses while the origin refreshes asynchronously.

Core Architectural Principles

  • Cache key = Hash(URI + Host + Query + Vary); each varied dimension multiplies distinct stored variants.
  • Vary: User-Agent fragments the cache into thousands of buckets — normalize to X-Device at the edge worker.
  • Surrogate-Key tag purge invalidates thousands of related URLs in <150ms; soft purge + SWR hides the miss spike.
Interview Round Script

Talk cache invalidation with precision: tag-based purge via Surrogate-Keys for surgical global eviction, soft purge with stale-while-revalidate to eliminate latency spikes, and header normalization so Vary never explodes cardinality. Quantifying hit-ratio loss from Vary: User-Agent instantly signals production experience.

Key Trade-Offs

Rich per-variant personalization gives razor-precise caching, but unnormalized keys and hard purges convert the edge into an expensive origin passthrough.

Related Curriculum Chapter

Edge Caching & Points of Presence (PoPs)

Read Full Chapter Blueprint

Explore More Interactive Labs

View All 280 Labs