System Design · Core

CDN

Edge caches that put static (and some dynamic) content near users: PoPs, Cache-Control, invalidation, and origin shield.

1. What a CDN is

A Content Delivery Network is a fleet of edge servers (points of presence) distributed worldwide. Users fetch assets from a nearby PoP instead of your origin in one region.

User (Delhi) --> Edge PoP (BOM) --hit--> cached object \ miss --> Origin (or shield) --> fill PoP

2. What to put on a CDN

Images, CSS, JS, fonts, downloadable files.
Video segments (HLS/DASH) with short or medium TTL.
Sometimes cacheable API GETs (public, versioned, low personalization).

Do not blindly cache personalized HTML or authenticated API responses without Vary / cache keys that include the right dimensions.

3. Request flow

DNS for cdn.example.com returns an edge Anycast/VIP near the user.
Edge looks up object by cache key (URL + selected headers).
Hit → serve; Miss → pull from origin (or parent shield), store, serve.
Respect TTL / revalidation (ETag, If-None-Match).

4. Cache-Control and related headers

Header / directive	Role
`Cache-Control: max-age=...`	Freshness lifetime for shared/private caches
`public` / `private`	CDN may store vs browser-only
`no-store` / `no-cache`	Do not store / revalidate before use
`s-maxage`	TTL specifically for shared caches (CDN)
`ETag` / `Last-Modified`	Conditional revalidation
`Vary`	Separate entries by header (e.g. Accept-Encoding)

Fingerprinted asset: /app.9f3a21.js → Cache-Control: public, max-age=31536000, immutable HTML shell: /index.html → Cache-Control: no-cache (always revalidate)

Fingerprint filenames so you can cache forever and deploy new bytes under a new URL. That beats emergency purges for routine releases.

5. Invalidation and purge

Purge by URL: remove one object from edges (eventual across PoPs).
Purge by tag/key: group related objects (product 42’s images).
Soft purge / stale-while-revalidate: serve stale while refreshing.

Purge is not instant worldwide and can be expensive. Prefer versioned URLs; purge for mistakes and urgent takedowns.

6. Origin shield

Many edges miss at once (popular new object) Without shield: thundering herd → Origin With shield: edges → Shield PoP → Origin (one miss folded)

A designated mid-tier cache collapses duplicate origin fetches, lowers origin load, and can improve cache hit ratio. Trade-off: extra hop on some misses.

7. Dynamic and security notes

Signed URLs / cookies for private content (course videos, paid downloads).
WAF and DDoS absorption often sit at the same edge.
TLS terminates at edge; origin pull over HTTPS.

8. Interview placement

In almost every media-heavy design (YouTube, Drive, Instagram), draw CDN in front of object storage. State TTL strategy, versioning, and purge for bad assets. Mention origin shield if origin QPS is a concern.

9. Cache key design

The cache key is usually the full URL plus selected headers from Vary. Query-string noise (utm_*) can fragment the cache — strip tracking params at the edge when safe. Prefer content-hashed paths for static assets so keys naturally change on deploy.

Good key: https://cdn.example.com/static/app.9f3a21.js Bad key: /app.js?utm_source=twitter (unless normalized)

10. Measuring CDN health

Edge hit ratio (bytes and requests).
Origin bandwidth and 5xx from origin pulls.
Time to first byte by geography.
Purge latency and failure rate after incidents.

Quick revision

CDN = edge PoPs caching content near users.
Best for static and segment-based media; careful with personalized data.
Cache-Control, ETag, Vary control freshness and variants.
Fingerprint assets for long TTL; short TTL or revalidate for HTML.
Purge for emergencies; versioning for routine deploys.
Origin shield collapses miss stampedes onto origin.
Signed URLs protect private objects at the edge.