Reverse Proxy vs CDN

What it is

• Reverse proxy (e.g. nginx, Envoy, HAProxy in front of origins): a server that accepts client connections on behalf of backend apps, often terminating TLS, enforcing limits, and caching responses at the edge of your stack.
• CDN (Content Delivery Network): a globally distributed set of edge PoPs that cache static (and sometimes dynamic) content close to users. Origin is your app or object storage; edges answer repeated requests without hitting origin every time.

When to use

Cache keys and TTL

• Cache key typically includes: URL path, query string normalization rules, Vary headers (e.g. Accept-Encoding, Accept-Language), and sometimes custom headers for cache segmentation.
• TTL (Cache-Control, s-maxage, CDN-Cache-Control): shorter for HTML that changes often; longer for versioned static assets (/assets/app-v123.js).
• Stale-while-revalidate: serve stale content while refreshing in background—better perceived latency.

  Client --GET--> CDN edge --miss--> Origin / reverse proxy --> App
                     |
                     +-- hit --> return cached body + Age header

Alternatives

• Direct to origin with no CDN: simpler, higher latency and load globally.
• Client-side caching (service workers, browser cache): complements CDN; does not replace edge for first visits.
• Application-level caching (Redis): for personalized or dynamic data; different layer than HTTP CDN.

Failure modes

• Stale content: long TTL after deploy; mitigate with cache busting (hashed filenames), purge APIs, or short max-age for HTML.
• Wrong Vary: serving gzip HTML to a client that cannot decode, or wrong language slice.
• Cache poisoning: malicious query strings create unique keys; normalize keys and strip unsafe inputs.
• Origin overload on purge storms or thundering herd after expiry—coordinate with caching strategies.

Interview talking points

• Separate static (CDN-friendly) from dynamic (API, auth); cookies often prevent CDN cache for personalized pages.
• Mention shield/origin capacity: CDN absorbs spikes; still estimate origin with back-of-envelope for cache miss ratio.
• Tie edge latency to latency-throughput: fewer round trips to origin improves user-visible latency.

Related fundamentals

• Back-of-the-envelope estimation
• Latency and throughput