HTTP Cache-Control Headers What no-cache, no-store, and max-age Actually Mean

You’ve probably written Cache-Control: no-cache and assumed the browser would skip the cache entirely. It won’t. no-cache means “revalidate before serving from cache.” If you actually want the response to never be stored, that’s no-store. This confusion causes stale data bugs in production that are annoying to diagnose because everything looks fine in the Network tab on first load.

Let’s go through each directive clearly — what browsers do with it, what CDNs do with it, and the mistakes that come from mixing them up.

The Full Cache-Control Directive Reference

指令	Browser behavior	CDN / proxy behavior	Typical use
max-age=N	Cache for N seconds	Cache for N seconds (unless s-maxage overrides)	Static assets, API responses
s-maxage=N	Ignored	Cache for N seconds	CDN TTL separate from browser
no-cache	Cache but revalidate every request	Cache but revalidate every request	Frequently-changing content with ETags
no-store	Do not store anywhere	Do not store anywhere	Auth responses, sensitive user data
must-revalidate	No stale serving — revalidate or fail	No stale serving — revalidate or fail	API responses where stale = broken
proxy-revalidate	Ignored	No stale serving on shared caches	CDN-specific must-revalidate
private	Browser may cache	Must not cache	User-specific pages
public	Any cache may store	May cache	Shared static resources
immutable	Never revalidate within max-age	Varies by CDN	Hashed / versioned assets
stale-while-revalidate=N	Serve stale for N seconds while fetching fresh	Varies by CDN	Speed without hard staleness

max-age and s-maxage: Browser vs CDN TTL

max-age=N tells both the browser and CDNs how many seconds the response is fresh. After N seconds, the cached response is stale and must be revalidated before use.

s-maxage=N is CDN-only. Browsers ignore it entirely. If you want a CDN to cache for an hour but the browser for only 5 minutes:

Cache-Control: max-age=300, s-maxage=3600

The browser caches for 5 minutes. CloudFront, Fastly, Nginx — they’ll use 3600 seconds. A common trap: setting max-age=0 thinking it disables caching. It doesn’t. max-age=0 means the response is immediately stale — the browser still caches it and revalidates, likely getting a 304. If you never want it cached, use no-store.

no-cache: Not What You Think

Cache-Control: no-cache does not mean “don’t use the cache.” It means “don’t serve from cache without revalidating with the server first.”

The sequence when a browser has a cached response with no-cache:

1. New request arrives for the cached resource
2. Browser sends the cached response’s ETag (via If-None-Match) or Last-Modified timestamp to the server
3. If content unchanged → server returns 304 Not Modified with no body
4. If content changed → server returns 200 OK with the new response

The benefit over no-store: you still get the bandwidth savings of 304 responses. If your content changes occasionally but must be fresh when it does, no-cache combined with an ETag is the right combination.

no-store: The Real “Don’t Cache This”

Cache-Control: no-store means the response must not be stored anywhere — not in the browser cache, not in a CDN, not in an intermediate proxy. No copy, full stop.

Use this for:

• Authentication responses (login tokens, session data)
• Sensitive personal data
• One-time content (payment confirmations, OTP pages)

One subtlety: no-store doesn’t prevent a page from appearing in the browser’s back-forward cache (bfcache). Browsers keep an in-memory snapshot for navigation performance that’s separate from the HTTP cache. If you need to handle post-logout back-button issues, hook into the pageshow event and check event.persisted.

must-revalidate: No Stale Grace

HTTP caching specs allow caches to serve stale responses when the origin is unreachable — a resilience feature most developers don’t know about. must-revalidate removes that leeway: once a cached response is stale, the cache must revalidate or return a 504. No stale serving under any circumstances.

# Without must-revalidate: CDN may serve stale if origin is slow or down
Cache-Control: max-age=3600

# With must-revalidate: stale = error, not a fallback
Cache-Control: max-age=3600, must-revalidate

Use this for API responses where serving stale data breaks functionality — inventory counts, pricing, auth state — rather than just looks slightly wrong.

private vs public: The CDN Bug That Leaks User Data

private means the response is intended for a specific user. Browsers can cache it; shared caches (CDNs, reverse proxies) must not.

public explicitly allows any cache — including shared — to store the response. Some caches only cache authenticated-request responses if you explicitly mark them public.

The real-world bug this causes: a developer copy-pastes Cache-Control: public, max-age=3600 from a static asset onto a page that includes user-specific data. The CDN caches the response. User B makes the same request and gets User A’s page from cache. This isn’t theoretical — GitHub had a variant of this in 2018. Mark authenticated or user-specific responses private explicitly, even if you think your CDN “knows” not to cache them.

ETags and Conditional Requests

ETags are how the server says “here’s a fingerprint of this response.” The browser stores the ETag alongside the cached response and sends it back on the next request via If-None-Match. If content hasn’t changed, the server returns 304 Not Modified with no body — same freshness enforcement as no-cache, much less bandwidth.

这 no-cache + ETag flow:

→ GET /api/config HTTP/1.1

← HTTP/1.1 200 OK
   Cache-Control: no-cache
   ETag: "abc123"
   [full response body]

→ GET /api/config HTTP/1.1
   If-None-Match: "abc123"

← HTTP/1.1 304 Not Modified
   [no body — browser uses its cached copy]

Two ETag types:

• Strong ETags ("abc123") — byte-for-byte identical. Required for CDN range request support.
• Weak ETags (W/"abc123") — semantically equivalent but not necessarily byte-identical. Fine for browser revalidation, not for range requests.

Nginx generates ETags automatically from file mtime and size. Express doesn’t add them by default — use app.set('etag', 'strong') 或 etag middleware explicitly.

Last-Modified and If-Modified-Since

Same concept as ETags but coarser — timestamp-based rather than content-hash-based. The server includes Last-Modified; the browser sends If-Modified-Since on subsequent requests.

The problem: if you redeploy and file modification times update even though content didn’t change, caches invalidate unnecessarily. A content-hash-based ETag won’t have this problem. Use ETags where possible and treat Last-Modified as a fallback for servers that don’t support ETags.

Vary: The Header That Silently Multiplies Your Cache

这 Vary header tells caches that the response may differ based on other request headers. Each unique combination of those headers gets its own cache entry.

Vary: Accept-Encoding

This tells caches to store separate responses for gzip, brotli, and identity encoding. Correct and common. The dangerous one: Vary: Cookie. Every user has a unique cookie set, so every user gets their own cache entry — effectively disabling shared caching. Many frameworks add Vary: Cookie silently. If your CDN cache hit rate is inexplicably low despite generous max-age values, check your response headers for Vary: Cookie sneaking in from session middleware.

Vary: * means “don’t cache this at all” in practice — every request is treated as unique. It’s equivalent to no-store for CDNs.

Cache-Busting with Query Parameters

When you need to force fresh downloads of versioned assets, appending a query param is the standard approach — the query string is part of the URL, so it’s treated as a new resource by both browsers and CDNs:

/app.js?v=2.1.4
/styles.css?hash=a1b2c3d4e5f6

If you’re constructing cache-busting params dynamically from content hashes or version strings that might contain special characters, make sure to percent-encode them before appending. The URL编码器/解码器 handles that quickly if you’re testing or building URLs manually.

The Three Mistakes That Bite Most Developers

1. Using no-cache when you mean no-store. If you’re handling auth responses, logout endpoints, or anything with PII, you want no-store. no-cache leaves data in the browser cache (just flagged as stale); no-store removes the footprint entirely. The difference matters when users share devices.

2. Not setting s-maxage for CDN control. Without s-maxage, your CDN uses max-age. If max-age is short for browser freshness (say, 60 seconds), your CDN caches for 60 seconds too — which probably isn’t what you want. Separate the two TTLs explicitly.

3. public on endpoints that return user data. This one is the security incident, not just a performance bug. Any response that’s personalized or authenticated should be private. Default to private and opt into public only for genuinely shared resources.

整合应用

The mental model: no-cache is about freshness enforcement — the response lives in the cache, it just needs a server stamp of approval before use. no-store is about leaving no trace. max-age is your browser TTL. s-maxage is your CDN’s separate TTL. ETags are what make revalidation cheap.

Get the private/public distinction right on any endpoint that touches user data. That single mistake — copying a cache header from a static asset onto an authenticated endpoint — is the one that turns into a cross-user data leak when your CDN starts caching personalized responses.

你可能也会喜欢