Cache Keys and TTL:… — CDN Caching: Global… | OneNoughtOne

0/121

System Design (HLD)CDN Caching

CDN Caching: Global Content Distribution at Scale

LevelIntermediate

Duration90 mins

TopicCDN Caching

3 / 5

Cache Keys and TTL: Controlling What Gets Cached and For How Long

The Cache Identity Problem

Imagine two users requesting what appears to be the same URL: https://shop.example.com/products/shoes — User in US, English locale https://shop.example.com/products/shoes — User in France, French locale Should the CDN serve a cached response from the first request to the second user? Clearly not—they expect different content. But how does the CDN know these are different resources? This is the cache key problem : determining the unique identity of cacheable content. Closely related is the TTL problem : how long should cached content be considered valid before requiring revalidation? These two mechanisms— cache keys and TTL —are the primary controls architects use to configure CDN behavior. Understanding them deeply is essential for achieving high cache hit rates while avoiding stale or incorrect content delivery.

What You Will Master

By the end of this page, you will understand how CDNs construct cache keys from request components, how to configure cache keys for complex content personalization scenarios, the complete hierarchy of TTL configuration sources, and best practices for balancing cache efficiency with content freshness. You'll be equipped to optimize CDN configurations for maximum performance.

Understanding Cache Keys

A cache key is a unique identifier that the CDN uses to store and retrieve cached content. When a request arrives at an edge server, the CDN computes the cache key for that request and checks if a cached response exists with that key. If found (cache hit), the cached content is returned; if not (cache miss), the request is forwarded to the origin. Default Cache Key Components: By default, most CDNs construct cache keys from: Host Header : The domain name in the request (cdn.example.com) Request Path : The URL path (/products/shoes) Query String : URL parameters (?color=red&size=10) This default produces cache keys like: cdn.example.com|/products/shoes|?color=red&size=10 Why Cache Keys Matter: Under-keying (cache key too simple): Different content gets the same cache key Users receive incorrect content (e.g., wrong locale, wrong user data) Security issues: one user seeing another's personalized content Over-keying (cache key too complex): Same content gets different cache keys Lower cache hit rates; more origin load Wasted edge storage on duplicate content The art of cache key configuration is finding the minimal key that correctly distinguishes between distinct content.

Cache Key Configuration Impact
Scenario	Cache Key Issue	Consequence
Same page, different locales	Under-keying (no locale in key)	French user sees English content
Same image with tracking parameters	Over-keying (full query string)	Same image cached 1000x with different keys
A/B test variants	Under-keying (no variant ID)	Users see wrong experiment variant
Unnecessary headers in key	Over-keying (User-Agent included)	Same content cached per browser version

Cache Key Mistakes Are Silent

Cache key configuration errors often don't produce obvious failures. Under-keyed content appears to work—until a user reports seeing wrong content. Over-keyed content also works—but with degraded cache hit rates that may not be immediately noticed. Invest time in correct cache key design upfront.

Cache Key Components: A Deep Dive

Let's examine each potential cache key component and understand when to include or exclude it. Host / Domain:

Host Header Considerations

•Include (default): Multiple domains on same CDN serve different content
•Normalize: www.example.com and example.com should often share cache keys
•Exclude: Single-site CDN configurations where host is redundant

URL Path Considerations

•Always Include: Path identifies the resource; essential for cache key
•Normalize Case: /Products/Shoes vs /products/shoes—many CDNs match insensitively
•Trailing Slash Normalization: /products/ vs /products—should these be same key?

Query String Handling: Query strings are where cache key configuration gets complex. Consider these URLs: /products?category=shoes&sort=price&utm_source=google /products?sort=price&category=shoes&fbclid=abc123 /products?category=shoes Should these share a cache key? category and sort affect content utm_source and fbclid are tracking parameters—they don't change content Parameter order technically makes URLs different, but content is the same Query String Strategies:

Query String Cache Key Strategies
Strategy	Description	Use Case
Include All	Full query string in cache key	Every parameter affects content (rare)
Exclude All	Ignore query string entirely	Static assets where params are irrelevant
Include Specific	Whitelist: only specified params in key	Known content-affecting params (category, locale)
Exclude Specific	Blacklist: remove specified params	Known tracking params (utm_*, fbclid)
Sort Parameters	Normalize param order before keying	Prevent order-based cache fragmentation

Headers in Cache Keys: SOme content varies based on HTTP headers:

Header	When to Include in Key
`Accept-Language`	Multi-language sites serving different content per locale
`Accept-Encoding`	Usually handled separately (content-encoding negotiation)
`Cookie`	Personalized/authenticated content (use carefully!)
`Accept`	Content negotiation (JSON vs HTML for same URL)
`User-Agent`	Rarely—mobile vs desktop sites if using same URL
`X-Forwarded-Proto`	HTTP vs HTTPS responses differ (usually normalized)

Cookie-Based Cache Keys: Including cookies in cache keys enables serving personalized content from cache: Cache-Key: host + path + cookie:locale + cookie:ab_variant But this creates per-user cache fragmentation. Best practice: Cache anonymous content aggressively Use edge compute for personalization (fetch from cache, transform at edge) Include only specific, structured cookies (locale, variant ID) not full cookie header

Start Simple, Add Complexity

Begin with the simplest cache key that produces correct behavior (host + path). Add query params only if needed for content differentiation. Add headers only with clear justification. Every addition to the cache key reduces cache efficiency—make each component earn its place.

Cache Key Configuration Patterns

Real-world applications require sophisticated cache key strategies. Here are common patterns and their implementations. Pattern 1: Static Asset Optimization For images, CSS, JS files: Path: /static/* Cache Key: Host + Path only Query String: Ignored entirely Headers: None Rationale: Static assets are immutable at a given path. Query strings are typically cache-busting versions or tracking parameters that shouldn't fragment the cache. Pattern 2: API Response Caching For cacheable API endpoints: Path: /api/products/* Cache Key: Host + Path + Selected Query Params (category, page, limit) Excluded Params: timestamp, requestId, tracking params Headers: Accept (if content negotiation used) Rationale: API responses vary by business-logic parameters but not by housekeeping parameters. Pattern 3: Localized Content For multi-language websites: Path: /* Cache Key: Host + Path + Query String + Accept-Language (normalized) Alternative: Host + Path + Query String + locale cookie Rationale: Same URL serves different content per locale. Normalization (en-US \to en) prevents over-fragmentation. Pattern 4: A/B Testing Variants For experiment-aware caching: Path: /* Cache Key: Host + Path + Query String + experiment_variant (from cookie or header) Rationale: Different experiment variants must be cached separately. Using a dedicated experiment identifier (not user ID) keeps fragmentation bounded.

Cloudflare Cache Key Configuration (Workers)
JavaScript
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
// Cloudflare Worker - Custom Cache Key
async function handleRequest(request) {
  const url = new URL(request.url);
  
  // Build custom cache key
  const cacheKeyParts = [
    url.hostname,
    url.pathname,
  ];
  
  // Include only content-affecting query params
  const relevantParams = ['category', 'sort', 'page', 'locale'];
  const searchParams = new URLSearchParams();
  
  relevantParams.forEach(param => {
    if (url.searchParams.has(param)) {
      searchParams.set(param, url.searchParams.get(param));
    }
  });
  
  // Sort params for consistent key generation
  searchParams.sort();
  if (searchParams.toString()) {
    cacheKeyParts.push(searchParams.toString());
  }
  
  // Include locale from Accept-Language if present
  const acceptLang = request.headers.get('Accept-Language');
  if (acceptLang) {
    const primaryLang = acceptLang.split(',')[0].split('-')[0]; // Normalize
    cacheKeyParts.push(`lang:${primaryLang}`);
  }
  
  const cacheKey = cacheKeyParts.join('|');
  
  // Use custom cache key for fetching
  const cache = caches.default;
  const cacheKeyRequest = new Request(
    `https://cache-key/${encodeURIComponent(cacheKey)}`,
    request
  );
  
  let response = await cache.match(cacheKeyRequest);
  if (response) {
    return response;
  }
  
  // Cache miss - fetch from origin
  response = await fetch(request);
  
  // Store with custom cache key
  const responseToCache = response.clone();
  event.waitUntil(cache.put(cacheKeyRequest, responseToCache));
  
  return response;
}

CDN-Specific Configuration

Cache key configuration syntax varies by CDN provider. Cloudflare uses Page Rules, Cache Rules, and Workers. Fastly uses VCL. Akamai uses Property Manager behaviors. AWS CloudFront uses Cache Policies. The concepts are universal; the implementation details differ.

Understanding Time-To-Live (TTL)

Time-To-Live (TTL) defines how long cached content remains valid before the CDN must revalidate or re-fetch it from the origin. TTL is the primary mechanism for balancing cache efficiency (longer TTL = more hits) with content freshness (shorter TTL = more timely updates). TTL Sources (Priority Order): CDNs determine TTL from multiple sources, typically evaluated in this order: CDN Override Rules : CDN configuration that overrides origin headers (highest priority) Cache-Control: s-maxage : Shared cache directive from origin response Cache-Control: max-age : Cache directive from origin response Expires Header : Specific expiration timestamp (legacy) CDN Default TTL : Fallback when origin provides no caching directives

HTTP Cache Control Directives for CDNs
Directive	Purpose	CDN Behavior
max-age=N	Content valid for N seconds	Cache for N seconds (unless s-maxage present)
s-maxage=N	Shared cache validity (N seconds)	Cache for N seconds (overrides max-age for CDN)
public	Response can be cached by shared caches	Explicitly cacheable
private	Response is user-specific	Do NOT cache at CDN
no-cache	Must revalidate before use	Cache but always revalidate
no-store	Never cache	Must not cache (bypass CDN)
stale-while-revalidate=N	Serve stale for N seconds while revalidating	Serve stale, async revalidate in background
stale-if-error=N	Serve stale if origin errors	Fallback to stale if origin returns 5xx

The TTL Trade-off:

Concern	Short TTL (e.g., 60s)	Long TTL (e.g., 1 day)
Cache Hit Rate	Lower (more revalidation)	Higher (more hits)
Origin Load	Higher (more requests)	Lower (fewer requests)
Content Freshness	More current	More stale
Update Propagation	~60 seconds	~24 hours (without purge)
Bandwidth Costs	Higher (duplicate transfers)	Lower

Understanding Stale Content: When TTL expires, content becomes stale . The CDN's behavior with stale content depends on configuration: Strict Freshness : Stale content is never served; CDN blocks until revalidation completes Stale-While-Revalidate : Stale content served immediately; revalidation happens async Stale-If-Error : Stale content served if origin returns error (failover) Modern CDNs strongly favor stale-while-revalidate for performance—serving slightly stale content is almost always preferable to adding latency.

s-maxage vs max-age

Use s-maxage to set CDN cache duration independent of browser cache duration. Example: Cache-Control: max-age=60, s-maxage=3600 tells browsers to cache for 1 minute but CDN can cache for 1 hour. This allows long CDN caching with short browser caching for faster updates (browser fetches from CDN after 60s, CDN still has content).

TTL Strategy Patterns

Different content types warrant different TTL strategies. Here's a comprehensive framework for TTL configuration. Immutable Content Strategy: For versioned/fingerprinted assets (content that never changes at a given URL): URL Pattern: /static/app.a1b2c3d4.js, /images/product-v2.png TTL: 1 year (31536000 seconds) Headers: Cache-Control: public, max-age=31536000, immutable Rationale: Content is immutable—the filename changes when content changes. Maximum caching is safe and optimal. Frequently Updated Content Strategy: For content that updates regularly (news, stock prices, dashboards): URL Pattern: /api/prices, /news/latest TTL: 60-300 seconds Headers: Cache-Control: public, s-maxage=60, stale-while-revalidate=300 Rationale: Balance freshness with caching. stale-while-revalidate ensures fast responses even during revalidation. Semi-Static Content Strategy: For content that changes occasionally (product pages, articles): URL Pattern: /products/*, /articles/* TTL: 3600-86400 seconds (1 hour to 1 day) Headers: Cache-Control: public, s-maxage=3600, stale-while-revalidate=86400 Invalidation: Event-driven purge when content updates Rationale: Long cache duration with on-demand invalidation for updates. stale-while-revalidate provides buffer for invalidation propagation. Never Cache Strategy: For personalized or sensitive content: URL Pattern: /api/me, /dashboard, /checkout TTL: 0 (no caching) Headers: Cache-Control: private, no-store Rationale: User-specific content must never be cached at CDN layer.

TTL Best Practices

•Use Immutable for Versioned Assets — If filenames include hashes, set max-age to 1 year with immutable directive. This is the most efficient caching pattern.
•Leverage stale-while-revalidate — Almost always preferable to blocking on revalidation. Users get fast responses; freshness is maintained async.
•Separate CDN and Browser TTL — Use s-maxage for CDN and max-age for browsers. Long CDN TTL + short browser TTL = fast CDN updates with revalidation.
•Default to Shorter TTL, Extend with Monitoring — Start conservative, increase TTL as you gain confidence in invalidation and monitor cache hit rates.
•Plan for Invalidation — Long TTL only works if you have reliable invalidation for updates. Without invalidation, TTL is your only freshness mechanism.

The Uncacheable Mistake

A common mistake is making content uncacheable when it should be cached with short TTL. Even TTL=10 seconds provides 90%+ cache hit rate under sustained traffic. Don't default to no-cache when s-maxage=60 with stale-while-revalidate would provide good freshness and dramatically better performance.

TTL Configuration Examples

Here are concrete examples of TTL configuration for common CDN platforms. Origin Server Headers (Node.js/Express):

Express.js Cache Control Middleware
JavaScript
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
const express = require('express');
const app = express();
 
// Middleware to set cache control headers based on path
function cacheControl(req, res, next) {
  const path = req.path;
  
  // Immutable static assets (fingerprinted filenames)
  if (path.match(/\.(js|css)$/) && path.includes('.')) {
    res.set('Cache-Control', 'public, max-age=31536000, immutable');
    return next();
  }
  
  // Images and fonts - long cache with revalidation
  if (path.match(/\.(png|jpg|jpeg|gif|svg|woff2|woff)$/)) {
    res.set('Cache-Control', 'public, max-age=86400, s-maxage=604800, stale-while-revalidate=86400');
    return next();
  }
  
  // API responses - short cache, async revalidation
  if (path.startsWith('/api/')) {
    res.set('Cache-Control', 'public, s-maxage=60, stale-while-revalidate=300, stale-if-error=86400');
    return next();
  }
  
  // HTML pages - moderate cache
  if (path.match(/\.html$/) || path === '/') {
    res.set('Cache-Control', 'public, s-maxage=300, stale-while-revalidate=3600');
    return next();
  }
  
  // User-specific pages - no caching
  if (path.startsWith('/dashboard') || path.startsWith('/account')) {
    res.set('Cache-Control', 'private, no-store');
    return next();
  }
  
  // Default
  res.set('Cache-Control', 'public, s-maxage=60');
  next();
}
 
app.use(cacheControl);

CDN Configuration Examples:

Cloudflare Cache Rules (Terraform)

HCL

# Cloudflare Ruleset for Cache Rules
resource "cloudflare_ruleset" "cache_rules" {
  zone_id     = var.zone_id
  name        = "Cache Rules"
  description = "Caching configuration"
  kind        = "zone"
  phase       = "http_request_cache_settings"
 
  # Immutable static assets
  rules {
    action = "set_cache_settings"
    action_parameters {
      edge_ttl {
        mode    = "override_origin"
        default = 31536000  # 1 year
      }
      browser_ttl {
        mode    = "override_origin"
        default = 31536000
      }
    }
    expression  = "(http.request.uri.path.extension in {\"js\" \"css\"} and http.request.uri.path contains \"-\")"
    description = "Fingerprinted assets - immutable caching"
    enabled     = true
  }
 
  # API responses
  rules {
    action = "set_cache_settings"
    action_parameters {
      edge_ttl {
        mode    = "override_origin"
        default = 60
      }
      serve_stale {
        disable_stale_while_updating = false
      }
    }
    expression  = "starts_with(http.request.uri.path, \"/api/\")"
    description = "API responses - short TTL with stale-while-revalidate"
    enabled     = true
  }
 
  # Bypass cache for authenticated routes
  rules {
    action      = "set_cache_settings"
    action_parameters {
      cache = false
    }
    expression  = "starts_with(http.request.uri.path, \"/dashboard\") or starts_with(http.request.uri.path, \"/account\")"
    description = "User-specific pages - bypass cache"
    enabled     = true
  }
}

Origin Headers vs CDN Override

Best practice: Set correct Cache-Control headers at the origin and let the CDN respect them. Use CDN overrides only when you can't modify origin behavior or need CDN-specific caching policies. Origin-controlled caching is more maintainable and works consistently across CDN providers.

Cache Variance and the Vary Header

The Vary header is a critical mechanism for indicating that responses differ based on request headers. It directly impacts how CDNs construct cache keys. How Vary Works: When an origin responds with Vary: Accept-Language, it's telling caches: "Responses for this URL differ based on the Accept-Language header. Include Accept-Language in your cache key." Common Vary Header Values:

Vary Header Usage Patterns
Vary Value	Use Case	CDN Impact
Accept-Encoding	Compressed vs uncompressed responses	CDN maintains gzip, br, identity variants
Accept-Language	Localized content	Cache per language preference
Accept	Content negotiation (HTML vs JSON)	Cache per content type
Cookie	Personalized content	DANGER: Per-user cache fragmentation
User-Agent	Device-specific responses	DANGER: Massive cache fragmentation
Origin	CORS preflight caching	Cache per requesting origin

The Vary Pitfalls: Vary: Cookie — Technically correct for personalized content, but creates per-user cache entries. Every unique cookie value creates a separate cache entry. For anonymous users with session cookies, this effectively disables caching. Vary: User-Agent — User-Agent strings are nearly unique per browser version. Vary: User-Agent creates separate cache entries for Firefox 98.0, Firefox 98.0.1, Chrome 99, Chrome 100, etc. This fragments the cache into unusability. The Solution: Explicit Header Normalization Instead of varying on high-cardinality headers, normalize to low-cardinality values: # Instead of: Vary: Cookie # Normalize cookie to specific values: X-Cache-Vary: locale=en|ab-variant=control # Instead of: Vary: User-Agent # Normalize to device class: X-Device-Type: mobile | tablet | desktop Vary: X-Device-Type Many CDNs can perform this normalization at the edge or through configuration.

Vary Header Normalization (Cloudflare Worker)
JavaScript
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
// Normalize high-cardinality headers for caching
async function handleRequest(request) {
  const modifiedHeaders = new Headers(request.headers);
  
  // Normalize Accept-Language to primary language only
  const acceptLang = request.headers.get('Accept-Language') || 'en';
  const primaryLang = acceptLang.split(',')[0].split(';')[0].split('-')[0].toLowerCase();
  modifiedHeaders.set('X-Normalized-Lang', primaryLang);
  
  // Normalize User-Agent to device class
  const userAgent = request.headers.get('User-Agent') || '';
  let deviceType = 'desktop';
  if (/Mobile|Android|iPhone/i.test(userAgent)) {
    deviceType = 'mobile';
  } else if (/iPad|Tablet/i.test(userAgent)) {
    deviceType = 'tablet';
  }
  modifiedHeaders.set('X-Device-Type', deviceType);
  
  // Normalize cookie to cache-relevant values only
  const cookies = request.headers.get('Cookie') || '';
  const locale = cookies.match(/locale=([^;]+)/)?.[1] || 'en';
  const abVariant = cookies.match(/ab_variant=([^;]+)/)?.[1] || 'control';
  modifiedHeaders.set('X-Cache-Key-Cookie', `locale=${locale}|variant=${abVariant}`);
  
  const modifiedRequest = new Request(request.url, {
    method: request.method,
    headers: modifiedHeaders,
    body: request.body,
  });
  
  return fetch(modifiedRequest);
}

Vary Header Validation

Audit your origin responses for Vary headers. A single Vary: Cookie response on your homepage can destroy your cache hit rate. Use CDN analytics to identify responses with problematic Vary headers and fix them at the origin or normalize at the edge.

System Design (HLD)CDN Caching

CDN Caching: Global Content Distribution at Scale

LevelIntermediate

Duration90 mins

TopicCDN Caching

3 / 5

Cache Keys and TTL: Controlling What Gets Cached and For How Long

The Cache Identity Problem

Imagine two users requesting what appears to be the same URL: https://shop.example.com/products/shoes — User in US, English locale https://shop.example.com/products/shoes — User in France, French locale Should the CDN serve a cached response from the first request to the second user? Clearly not—they expect different content. But how does the CDN know these are different resources? This is the cache key problem : determining the unique identity of cacheable content. Closely related is the TTL problem : how long should cached content be considered valid before requiring revalidation? These two mechanisms— cache keys and TTL —are the primary controls architects use to configure CDN behavior. Understanding them deeply is essential for achieving high cache hit rates while avoiding stale or incorrect content delivery.

What You Will Master

By the end of this page, you will understand how CDNs construct cache keys from request components, how to configure cache keys for complex content personalization scenarios, the complete hierarchy of TTL configuration sources, and best practices for balancing cache efficiency with content freshness. You'll be equipped to optimize CDN configurations for maximum performance.

Understanding Cache Keys

A cache key is a unique identifier that the CDN uses to store and retrieve cached content. When a request arrives at an edge server, the CDN computes the cache key for that request and checks if a cached response exists with that key. If found (cache hit), the cached content is returned; if not (cache miss), the request is forwarded to the origin. Default Cache Key Components: By default, most CDNs construct cache keys from: Host Header : The domain name in the request (cdn.example.com) Request Path : The URL path (/products/shoes) Query String : URL parameters (?color=red&size=10) This default produces cache keys like: cdn.example.com|/products/shoes|?color=red&size=10 Why Cache Keys Matter: Under-keying (cache key too simple): Different content gets the same cache key Users receive incorrect content (e.g., wrong locale, wrong user data) Security issues: one user seeing another's personalized content Over-keying (cache key too complex): Same content gets different cache keys Lower cache hit rates; more origin load Wasted edge storage on duplicate content The art of cache key configuration is finding the minimal key that correctly distinguishes between distinct content.

Cache Key Configuration Impact
Scenario	Cache Key Issue	Consequence
Same page, different locales	Under-keying (no locale in key)	French user sees English content
Same image with tracking parameters	Over-keying (full query string)	Same image cached 1000x with different keys
A/B test variants	Under-keying (no variant ID)	Users see wrong experiment variant
Unnecessary headers in key	Over-keying (User-Agent included)	Same content cached per browser version

Cache Key Mistakes Are Silent

Cache key configuration errors often don't produce obvious failures. Under-keyed content appears to work—until a user reports seeing wrong content. Over-keyed content also works—but with degraded cache hit rates that may not be immediately noticed. Invest time in correct cache key design upfront.

Cache Key Components: A Deep Dive

Let's examine each potential cache key component and understand when to include or exclude it. Host / Domain:

Host Header Considerations

•Include (default): Multiple domains on same CDN serve different content
•Normalize: www.example.com and example.com should often share cache keys
•Exclude: Single-site CDN configurations where host is redundant

URL Path Considerations

•Always Include: Path identifies the resource; essential for cache key
•Normalize Case: /Products/Shoes vs /products/shoes—many CDNs match insensitively
•Trailing Slash Normalization: /products/ vs /products—should these be same key?

Query String Handling: Query strings are where cache key configuration gets complex. Consider these URLs: /products?category=shoes&sort=price&utm_source=google /products?sort=price&category=shoes&fbclid=abc123 /products?category=shoes Should these share a cache key? category and sort affect content utm_source and fbclid are tracking parameters—they don't change content Parameter order technically makes URLs different, but content is the same Query String Strategies:

Query String Cache Key Strategies
Strategy	Description	Use Case
Include All	Full query string in cache key	Every parameter affects content (rare)
Exclude All	Ignore query string entirely	Static assets where params are irrelevant
Include Specific	Whitelist: only specified params in key	Known content-affecting params (category, locale)
Exclude Specific	Blacklist: remove specified params	Known tracking params (utm_*, fbclid)
Sort Parameters	Normalize param order before keying	Prevent order-based cache fragmentation

Headers in Cache Keys: SOme content varies based on HTTP headers:

Header	When to Include in Key
`Accept-Language`	Multi-language sites serving different content per locale
`Accept-Encoding`	Usually handled separately (content-encoding negotiation)
`Cookie`	Personalized/authenticated content (use carefully!)
`Accept`	Content negotiation (JSON vs HTML for same URL)
`User-Agent`	Rarely—mobile vs desktop sites if using same URL
`X-Forwarded-Proto`	HTTP vs HTTPS responses differ (usually normalized)

Cookie-Based Cache Keys: Including cookies in cache keys enables serving personalized content from cache: Cache-Key: host + path + cookie:locale + cookie:ab_variant But this creates per-user cache fragmentation. Best practice: Cache anonymous content aggressively Use edge compute for personalization (fetch from cache, transform at edge) Include only specific, structured cookies (locale, variant ID) not full cookie header

Start Simple, Add Complexity

Begin with the simplest cache key that produces correct behavior (host + path). Add query params only if needed for content differentiation. Add headers only with clear justification. Every addition to the cache key reduces cache efficiency—make each component earn its place.

Cache Key Configuration Patterns

Real-world applications require sophisticated cache key strategies. Here are common patterns and their implementations. Pattern 1: Static Asset Optimization For images, CSS, JS files: Path: /static/* Cache Key: Host + Path only Query String: Ignored entirely Headers: None Rationale: Static assets are immutable at a given path. Query strings are typically cache-busting versions or tracking parameters that shouldn't fragment the cache. Pattern 2: API Response Caching For cacheable API endpoints: Path: /api/products/* Cache Key: Host + Path + Selected Query Params (category, page, limit) Excluded Params: timestamp, requestId, tracking params Headers: Accept (if content negotiation used) Rationale: API responses vary by business-logic parameters but not by housekeeping parameters. Pattern 3: Localized Content For multi-language websites: Path: /* Cache Key: Host + Path + Query String + Accept-Language (normalized) Alternative: Host + Path + Query String + locale cookie Rationale: Same URL serves different content per locale. Normalization (en-US \to en) prevents over-fragmentation. Pattern 4: A/B Testing Variants For experiment-aware caching: Path: /* Cache Key: Host + Path + Query String + experiment_variant (from cookie or header) Rationale: Different experiment variants must be cached separately. Using a dedicated experiment identifier (not user ID) keeps fragmentation bounded.

Cloudflare Cache Key Configuration (Workers)
JavaScript
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
// Cloudflare Worker - Custom Cache Key
async function handleRequest(request) {
  const url = new URL(request.url);
  
  // Build custom cache key
  const cacheKeyParts = [
    url.hostname,
    url.pathname,
  ];
  
  // Include only content-affecting query params
  const relevantParams = ['category', 'sort', 'page', 'locale'];
  const searchParams = new URLSearchParams();
  
  relevantParams.forEach(param => {
    if (url.searchParams.has(param)) {
      searchParams.set(param, url.searchParams.get(param));
    }
  });
  
  // Sort params for consistent key generation
  searchParams.sort();
  if (searchParams.toString()) {
    cacheKeyParts.push(searchParams.toString());
  }
  
  // Include locale from Accept-Language if present
  const acceptLang = request.headers.get('Accept-Language');
  if (acceptLang) {
    const primaryLang = acceptLang.split(',')[0].split('-')[0]; // Normalize
    cacheKeyParts.push(`lang:${primaryLang}`);
  }
  
  const cacheKey = cacheKeyParts.join('|');
  
  // Use custom cache key for fetching
  const cache = caches.default;
  const cacheKeyRequest = new Request(
    `https://cache-key/${encodeURIComponent(cacheKey)}`,
    request
  );
  
  let response = await cache.match(cacheKeyRequest);
  if (response) {
    return response;
  }
  
  // Cache miss - fetch from origin
  response = await fetch(request);
  
  // Store with custom cache key
  const responseToCache = response.clone();
  event.waitUntil(cache.put(cacheKeyRequest, responseToCache));
  
  return response;
}

CDN-Specific Configuration

Cache key configuration syntax varies by CDN provider. Cloudflare uses Page Rules, Cache Rules, and Workers. Fastly uses VCL. Akamai uses Property Manager behaviors. AWS CloudFront uses Cache Policies. The concepts are universal; the implementation details differ.

Understanding Time-To-Live (TTL)

Time-To-Live (TTL) defines how long cached content remains valid before the CDN must revalidate or re-fetch it from the origin. TTL is the primary mechanism for balancing cache efficiency (longer TTL = more hits) with content freshness (shorter TTL = more timely updates). TTL Sources (Priority Order): CDNs determine TTL from multiple sources, typically evaluated in this order: CDN Override Rules : CDN configuration that overrides origin headers (highest priority) Cache-Control: s-maxage : Shared cache directive from origin response Cache-Control: max-age : Cache directive from origin response Expires Header : Specific expiration timestamp (legacy) CDN Default TTL : Fallback when origin provides no caching directives

HTTP Cache Control Directives for CDNs
Directive	Purpose	CDN Behavior
max-age=N	Content valid for N seconds	Cache for N seconds (unless s-maxage present)
s-maxage=N	Shared cache validity (N seconds)	Cache for N seconds (overrides max-age for CDN)
public	Response can be cached by shared caches	Explicitly cacheable
private	Response is user-specific	Do NOT cache at CDN
no-cache	Must revalidate before use	Cache but always revalidate
no-store	Never cache	Must not cache (bypass CDN)
stale-while-revalidate=N	Serve stale for N seconds while revalidating	Serve stale, async revalidate in background
stale-if-error=N	Serve stale if origin errors	Fallback to stale if origin returns 5xx

The TTL Trade-off:

Concern	Short TTL (e.g., 60s)	Long TTL (e.g., 1 day)
Cache Hit Rate	Lower (more revalidation)	Higher (more hits)
Origin Load	Higher (more requests)	Lower (fewer requests)
Content Freshness	More current	More stale
Update Propagation	~60 seconds	~24 hours (without purge)
Bandwidth Costs	Higher (duplicate transfers)	Lower

Understanding Stale Content: When TTL expires, content becomes stale . The CDN's behavior with stale content depends on configuration: Strict Freshness : Stale content is never served; CDN blocks until revalidation completes Stale-While-Revalidate : Stale content served immediately; revalidation happens async Stale-If-Error : Stale content served if origin returns error (failover) Modern CDNs strongly favor stale-while-revalidate for performance—serving slightly stale content is almost always preferable to adding latency.

s-maxage vs max-age

Use s-maxage to set CDN cache duration independent of browser cache duration. Example: Cache-Control: max-age=60, s-maxage=3600 tells browsers to cache for 1 minute but CDN can cache for 1 hour. This allows long CDN caching with short browser caching for faster updates (browser fetches from CDN after 60s, CDN still has content).

TTL Strategy Patterns

Different content types warrant different TTL strategies. Here's a comprehensive framework for TTL configuration. Immutable Content Strategy: For versioned/fingerprinted assets (content that never changes at a given URL): URL Pattern: /static/app.a1b2c3d4.js, /images/product-v2.png TTL: 1 year (31536000 seconds) Headers: Cache-Control: public, max-age=31536000, immutable Rationale: Content is immutable—the filename changes when content changes. Maximum caching is safe and optimal. Frequently Updated Content Strategy: For content that updates regularly (news, stock prices, dashboards): URL Pattern: /api/prices, /news/latest TTL: 60-300 seconds Headers: Cache-Control: public, s-maxage=60, stale-while-revalidate=300 Rationale: Balance freshness with caching. stale-while-revalidate ensures fast responses even during revalidation. Semi-Static Content Strategy: For content that changes occasionally (product pages, articles): URL Pattern: /products/*, /articles/* TTL: 3600-86400 seconds (1 hour to 1 day) Headers: Cache-Control: public, s-maxage=3600, stale-while-revalidate=86400 Invalidation: Event-driven purge when content updates Rationale: Long cache duration with on-demand invalidation for updates. stale-while-revalidate provides buffer for invalidation propagation. Never Cache Strategy: For personalized or sensitive content: URL Pattern: /api/me, /dashboard, /checkout TTL: 0 (no caching) Headers: Cache-Control: private, no-store Rationale: User-specific content must never be cached at CDN layer.

TTL Best Practices

•Use Immutable for Versioned Assets — If filenames include hashes, set max-age to 1 year with immutable directive. This is the most efficient caching pattern.
•Leverage stale-while-revalidate — Almost always preferable to blocking on revalidation. Users get fast responses; freshness is maintained async.
•Separate CDN and Browser TTL — Use s-maxage for CDN and max-age for browsers. Long CDN TTL + short browser TTL = fast CDN updates with revalidation.
•Default to Shorter TTL, Extend with Monitoring — Start conservative, increase TTL as you gain confidence in invalidation and monitor cache hit rates.
•Plan for Invalidation — Long TTL only works if you have reliable invalidation for updates. Without invalidation, TTL is your only freshness mechanism.

The Uncacheable Mistake

A common mistake is making content uncacheable when it should be cached with short TTL. Even TTL=10 seconds provides 90%+ cache hit rate under sustained traffic. Don't default to no-cache when s-maxage=60 with stale-while-revalidate would provide good freshness and dramatically better performance.

TTL Configuration Examples

Here are concrete examples of TTL configuration for common CDN platforms. Origin Server Headers (Node.js/Express):

Express.js Cache Control Middleware
JavaScript
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
const express = require('express');
const app = express();
 
// Middleware to set cache control headers based on path
function cacheControl(req, res, next) {
  const path = req.path;
  
  // Immutable static assets (fingerprinted filenames)
  if (path.match(/\.(js|css)$/) && path.includes('.')) {
    res.set('Cache-Control', 'public, max-age=31536000, immutable');
    return next();
  }
  
  // Images and fonts - long cache with revalidation
  if (path.match(/\.(png|jpg|jpeg|gif|svg|woff2|woff)$/)) {
    res.set('Cache-Control', 'public, max-age=86400, s-maxage=604800, stale-while-revalidate=86400');
    return next();
  }
  
  // API responses - short cache, async revalidation
  if (path.startsWith('/api/')) {
    res.set('Cache-Control', 'public, s-maxage=60, stale-while-revalidate=300, stale-if-error=86400');
    return next();
  }
  
  // HTML pages - moderate cache
  if (path.match(/\.html$/) || path === '/') {
    res.set('Cache-Control', 'public, s-maxage=300, stale-while-revalidate=3600');
    return next();
  }
  
  // User-specific pages - no caching
  if (path.startsWith('/dashboard') || path.startsWith('/account')) {
    res.set('Cache-Control', 'private, no-store');
    return next();
  }
  
  // Default
  res.set('Cache-Control', 'public, s-maxage=60');
  next();
}
 
app.use(cacheControl);

CDN Configuration Examples:

Cloudflare Cache Rules (Terraform)

HCL

# Cloudflare Ruleset for Cache Rules
resource "cloudflare_ruleset" "cache_rules" {
  zone_id     = var.zone_id
  name        = "Cache Rules"
  description = "Caching configuration"
  kind        = "zone"
  phase       = "http_request_cache_settings"
 
  # Immutable static assets
  rules {
    action = "set_cache_settings"
    action_parameters {
      edge_ttl {
        mode    = "override_origin"
        default = 31536000  # 1 year
      }
      browser_ttl {
        mode    = "override_origin"
        default = 31536000
      }
    }
    expression  = "(http.request.uri.path.extension in {\"js\" \"css\"} and http.request.uri.path contains \"-\")"
    description = "Fingerprinted assets - immutable caching"
    enabled     = true
  }
 
  # API responses
  rules {
    action = "set_cache_settings"
    action_parameters {
      edge_ttl {
        mode    = "override_origin"
        default = 60
      }
      serve_stale {
        disable_stale_while_updating = false
      }
    }
    expression  = "starts_with(http.request.uri.path, \"/api/\")"
    description = "API responses - short TTL with stale-while-revalidate"
    enabled     = true
  }
 
  # Bypass cache for authenticated routes
  rules {
    action      = "set_cache_settings"
    action_parameters {
      cache = false
    }
    expression  = "starts_with(http.request.uri.path, \"/dashboard\") or starts_with(http.request.uri.path, \"/account\")"
    description = "User-specific pages - bypass cache"
    enabled     = true
  }
}

Origin Headers vs CDN Override

Best practice: Set correct Cache-Control headers at the origin and let the CDN respect them. Use CDN overrides only when you can't modify origin behavior or need CDN-specific caching policies. Origin-controlled caching is more maintainable and works consistently across CDN providers.

Cache Variance and the Vary Header

The Vary header is a critical mechanism for indicating that responses differ based on request headers. It directly impacts how CDNs construct cache keys. How Vary Works: When an origin responds with Vary: Accept-Language, it's telling caches: "Responses for this URL differ based on the Accept-Language header. Include Accept-Language in your cache key." Common Vary Header Values:

Vary Header Usage Patterns
Vary Value	Use Case	CDN Impact
Accept-Encoding	Compressed vs uncompressed responses	CDN maintains gzip, br, identity variants
Accept-Language	Localized content	Cache per language preference
Accept	Content negotiation (HTML vs JSON)	Cache per content type
Cookie	Personalized content	DANGER: Per-user cache fragmentation
User-Agent	Device-specific responses	DANGER: Massive cache fragmentation
Origin	CORS preflight caching	Cache per requesting origin

The Vary Pitfalls: Vary: Cookie — Technically correct for personalized content, but creates per-user cache entries. Every unique cookie value creates a separate cache entry. For anonymous users with session cookies, this effectively disables caching. Vary: User-Agent — User-Agent strings are nearly unique per browser version. Vary: User-Agent creates separate cache entries for Firefox 98.0, Firefox 98.0.1, Chrome 99, Chrome 100, etc. This fragments the cache into unusability. The Solution: Explicit Header Normalization Instead of varying on high-cardinality headers, normalize to low-cardinality values: # Instead of: Vary: Cookie # Normalize cookie to specific values: X-Cache-Vary: locale=en|ab-variant=control # Instead of: Vary: User-Agent # Normalize to device class: X-Device-Type: mobile | tablet | desktop Vary: X-Device-Type Many CDNs can perform this normalization at the edge or through configuration.

Vary Header Normalization (Cloudflare Worker)
JavaScript
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
// Normalize high-cardinality headers for caching
async function handleRequest(request) {
  const modifiedHeaders = new Headers(request.headers);
  
  // Normalize Accept-Language to primary language only
  const acceptLang = request.headers.get('Accept-Language') || 'en';
  const primaryLang = acceptLang.split(',')[0].split(';')[0].split('-')[0].toLowerCase();
  modifiedHeaders.set('X-Normalized-Lang', primaryLang);
  
  // Normalize User-Agent to device class
  const userAgent = request.headers.get('User-Agent') || '';
  let deviceType = 'desktop';
  if (/Mobile|Android|iPhone/i.test(userAgent)) {
    deviceType = 'mobile';
  } else if (/iPad|Tablet/i.test(userAgent)) {
    deviceType = 'tablet';
  }
  modifiedHeaders.set('X-Device-Type', deviceType);
  
  // Normalize cookie to cache-relevant values only
  const cookies = request.headers.get('Cookie') || '';
  const locale = cookies.match(/locale=([^;]+)/)?.[1] || 'en';
  const abVariant = cookies.match(/ab_variant=([^;]+)/)?.[1] || 'control';
  modifiedHeaders.set('X-Cache-Key-Cookie', `locale=${locale}|variant=${abVariant}`);
  
  const modifiedRequest = new Request(request.url, {
    method: request.method,
    headers: modifiedHeaders,
    body: request.body,
  });
  
  return fetch(modifiedRequest);
}

Vary Header Validation

Audit your origin responses for Vary headers. A single Vary: Cookie response on your homepage can destroy your cache hit rate. Use CDN analytics to identify responses with problematic Vary headers and fix them at the origin or normalize at the edge.

Cache Keys and TTL: Controlling What Gets Cached and For How Long

The Cache Identity Problem

Understanding Cache Keys

Cache Key Components: A Deep Dive

Cache Key Configuration Patterns

Understanding Time-To-Live (TTL)

TTL Strategy Patterns

TTL Configuration Examples

Cache Variance and the Vary Header

There’s more to understand.

Cache Keys and TTL: Controlling What Gets Cached and For How Long

The Cache Identity Problem

Understanding Cache Keys

Cache Key Components: A Deep Dive

Cache Key Configuration Patterns

Understanding Time-To-Live (TTL)

TTL Strategy Patterns

TTL Configuration Examples

Cache Variance and the Vary Header

There’s more to understand.