Written by Harry Roberts on CSS Wizardry.
Table of Contents
Cache-Control
RecapCache-Control
as a Request Header- Refresh
- Hard Refresh
max-age=0
vsno-cache
- Revalidation
- When to Use a
Cache-Control
Request Header- Realtime Data
- Offline Applications
- Final Takeaways
I’ve written and
spoken many, many
times about the Cache-Control
response header and its many directives, but one
thing I haven’t covered before—and something I don’t think many developers are
even aware of—is the Cache-Control
request header. Unless you know your
caching well, those two links in the first sentence will make this article a lot
easier to understand. Maybe pop them open in another tab as a reference.
Let’s go!
Cache-Control
Recap
As developers, we’re most used to Cache-Control
as the preferred way of
instructing caches (usually browsers) on how they should store responses (if at
all), and what to do once their cache lifetime is up. Maybe something like this:
Cache-Control: max-age=2147483648, immutable
There’s a lot more to it than that, which you can read about in my 2019 piece,
Cache-Control for
Civilians
One thing we’re probably less used to is Cache-Control
’s employment as
a request header.
In a nutshell, the Cache-Control
request header determines whether the browser
retrieves content from the cache or forces a network request. It’s also used by
intermediaries such as CDNs to work out whether they should serve a response
themselves, or keep passing the request back upstream to origin.
It’s a way for the client to force freshness.
This means that the most common way you’re ever likely to see a Cache-Control
request header is when you refresh or hard refresh a page. Honestly, that’s
mostly it.
All browsers behave a little differently between refreshes, hard refreshes, and
run of the mill
revalidation.
Refresh
In Chrome, even if the page is still
fresh,
refreshing it will dispatch a request to the network with the following request
headers:
Cache-Control: max-age=0
[If-Modified-Since|If-None-Match]
Cache-Control: max-age=0
just means after zero seconds, revalidate this
resource. This isn’t incredibly strictly enforced so it’s technically a weak
instruction to revalidate. More on that later.- The
If-Modified-Since
orIf-None-Match
headers are revalidation request
headers that are used to compare the current version of the response with the
target version on the network.
All other subresources on the page are fetched as per their caching headers, so
there is no different or specific behaviour here.
In Firefox the behaviour is a little different. Refreshing a still-fresh
page results in the following request headers:
[If-Modified-Since|If-None-Match]
No Cache-Control
request header at all, just the relevant revalidation
headers.
Again, all of the page’s subresources are treated as normal.
Safari is different still, and generally seems much more aggressive with its
cache busting. Refreshing the same still-fresh page in Safari gives the
following request headers:
Cache-Control: no-cache
Pragma: no-cache
We have a Cache-Control
request header, this time with a no-cache
directive.
While this is functionally equivalent to max-age=0
, the spec speaks much more
clearly that no-cache
means that a cache MUST NOT use the response to
. We also have the first appearance of
satisfy a subsequent request without successful revalidation with the origin
serverPragma
, also carrying
no-cache
. Pragma
is an incredibly outdated header that serves as a backward
compatibility measure for HTTP/1.0 caches. Safari including this here is a very
defensive measure!
Again, all other subresources are treated as they would be normally.
All browsers exhibit some similarities and some differences.
- Even if the main document was still fresh in HTTP cache, a refresh will always
put a request out onto the network in all browsers. - Chrome and Firefox emit revalidation headers which mean that, even though
we’ve refreshed the page, we might still get served our locally cached version
(304
) if it’s still valid. - Firefox doesn’t emit a
Cache-Control
header, making it the least aggressive
of the three. - Safari is by far the most aggressive, emitting both
Cache-Control
and
Pragma
headers, and no revalidation headers for potential reuse. Safari will
always return a200
response to a refresh.
Hard Refresh
Things are a little different when it comes to a hard refresh. Hard refreshes
are usually a sign of user frustration and that something is badly broken or
outdated. To this end, browsers begin upping the ante here.
In all browsers, a hard refresh causes both the main document and all of its
subresources to be requested with the following:
Cache-Control: no-cache
Pragma: no-cache
Key things to note:
- No browser emits a revalidation header, meaning a
304
is not possible. We’re
always guaranteed a fresh response. - Chrome switched from
max-age=0
tono-cache
. This is a clear signal of
intent that a hard refresh is more aggressive than a regular one. - Safari’s behaviour remains unchanged, which means that as far as the main
document is concerned, a refresh and a hard refresh are equivalent.
Note that this all applies to the main document and all of its subresources—even
immutable
assets—so
everything on the page is now guaranteed fresh. 304
responses are not
possible.
max-age=0
vs no-cache
Both of these directives behave incredibly similarly: max-age=0
means the
response is considered stale after zero seconds and therefore should be
revalidated, and no-cache
means don’t fetch this response from cache without
revalidating it first.
Where they differ is that max-age=0
permits caches to reuse a response if
revalidation isn’t possible (e.g. no network access); no-cache
is much
stricter—it means the cache must always revalidate before releasing a response,
or return an error if revalidation fails.
Revalidation
In the case that a user hasn’t refreshed the page, but instead they have a file
in their cache that is now considered
stale,
the browser needs to check with the server whether or not it needs a new copy,
or if it can reuse and renew the previously cached version. This is called
revalidation and is when the If-Modified-Since
or If-None-Match
headers
come into play.
If-Modified-Since
is used to check a file against itsLast-Modified
response header.If-None-Match
is used to check a file against itsEtag
response
header.
When a file needs revalidating, all browsers behave the same:
[If-Modified-Since|If-None-Match]
They attach the relevant revalidation header, which will result in either
a 200
or 304
response in most cases. This is unremarkable other than the
fact that no browser attaches a Cache-Control
request header at this point.
Each of these use cases was browser-defined, very much out of our hands as web
developers, but there are scenarios when we might want to (and can!) add our own
Cache-Control
request headers. Think of these scenarios as incredibly
aggressive, incredibly defensive bidirectional caching rules to absolutely
guarantee that no caches anywhere along with request–response lifecycle will
retain a copy of a response. By setting Cache-Control
at both ends, we have
a double-pronged approach to our strategy. A very cautious approach.
Realtime Data
Imagine you’re building a sports betting site or stock trading app: realtime
price updates are incredibly important, and all data must be up to date,
always. You’d serve your responses with something like:
…and make your requests with something like:
fetch("https://api.website.com/data", {
method: "GET",
headers: {
"Cache-Control": "no-store",
}
})
This is the bare minimum for modern and compliant caches, but the
hyper-defensive version would be more like:
Cache-Control: no-store, no-cache, max-age=0, must-revalidate
Pragma: no-cache
…on your responses, and this in your requests:
fetch("https://api.website.com/data", {
method: "GET",
headers: {
"Cache-Control": "no-store, no-cache, max-age=0",
"Pragma": "no-cache"
}
})
The latter two examples are overkill and do contain a lot of redundancy, but
they also won’t do any harm.
Note that if the data is also potentially sensitive and contains
user-specific data, you’d want to add private
to your Cache-Control
response
headers.
Offline Applications
If you have an offline application, you can use only-if-cached
to only serve
a response if it’s in cache, otherwise returning a 504
.
fetch("https://api.website.com/offline-data", {
method: "GET",
headers: {
"Cache-Control": "only-if-cached"
}
})
Adding this request header ensures that the request would never hit the network.
While only-if-cached
might not be useful for most web pages, it can be handy
for offline-first applications, such as PWAs or news readers that prefer using
stored content rather than attempting a network request that might fail.
Final Takeaways
- Browsers automatically send a mix
Cache-Control
,Pragma
, or revalidation
headers in refresh and hard refresh scenarios. max-age=0
andno-cache
both trigger revalidation, butno-cache
is much
stricter and requires a fresh response.- You can manually use
Cache-Control
in requests when you need realtime data
or offline-first apps. - To force freshness, use:
Cache-Control: no-store, no-cache, max-age=0
- To build offline-first apps, consider:
Cache-Control: only-if-cached