HTTP headers are an important way of controlling how caches and browsers process your web content. But many are used incorrectly or pointlessly, which adds overhead at a critical time in the loading of your page, and may not work as you intended. In this first of a series of posts about header best practice, we’ll look at unnecessary headers.
Most developers know about and depend on a variety of HTTP headers to make their content work. Those that are best known include
Content-Length, which are both almost universal. But more recently, headers such as
Strict-Transport-Security have started to improve security, and
Link rel=preload headers to improve performance. Few sites use these, despite their wide support across browsers.
At the same time, there are lots of headers that are hugely popular but aren’t new and aren’t actually all that useful. We can prove this using HTTP Archive, a project run by Google and sponsored by Fastly that loads 500,000 websites every month using WebPageTest, and makes the results available in BigQuery.
From the HTTP Archive data, here are the 30 most popular response headers (based on the number of domains in the archive which are serving each header), and roughly how useful each one is:
|date||48779277||535621||Required by protocol|
|content-type||47185627||533636||Usually required by browser|
|content-encoding||16194121||409159||Required for compressed content|
|x-robots-tag||179177||24911||Not relevant to browsers|
Let’s look at the unnecessary headers and see why we don’t need them, and what we can do about it.
Vanity (server, x-powered-by, via)
You may be very proud of your choice of server software, but most people couldn’t care less. At worst, these headers might be divulging sensitive data that makes your site easier to attack.
Server: apache X-Powered-By: PHP/5.1.1 Via: 1.1 varnish, 1.1 squid
RFC7231 allows for servers to include a
Server header in the response, identifying the software used to serve the content. This is most commonly a string like “apache” or “nginx”. While it’s allowed, it’s not mandatory, and offers very little value to either developers or end users. Nevertheless, this is the third most popular HTTP response header on the web today.
X-Powered-By is the most popular header in our list that is not defined in any standard, and has a similar purpose, though normally refers to the application platform that sits behind the web server. Common values include “ASP.net”, “PHP” and “Express”. Again this isn’t providing any tangible benefit and is taking up space.
More debatable perhaps is
Via, which is required (by RFC7230) to be added to the request by any proxy through which it passes to identify the proxy. This can be the proxy’s hostname, but is more likely to be a generic identifier like “vegur”, “varnish”, or “squid”. Removing (or not setting) this header on a request can cause proxy forwarding loops. However, interestingly it is also copied into the response on the way back to the browser, and here it’s just informational and no browsers do anything with it, so it’s reasonably safe to get rid of it if you want to.
Deprecated standards (P3P, Expires, X-Frame-Options, X-UA-Compatible)
Another category of headers is those that do have an effect in the browser but are not (or are no longer) the best way of achieving that effect.
P3P: cp="this is not a p3p policy" Expires: Thu, 01 Dec 1994 16:00:00 GMT X-Frame-Options: SAMEORIGIN X-UA-Compatible: IE=edge
P3P is a curious animal. I had no idea what this was, and even more curiously, one of the most common values is “this is not a p3p policy”. Well, is it, or isn’t it?
P3P didn’t trigger any visual effect to the user; it just needs to be present to permit access to third party cookies in iframes. Some sites even set a non-conforming P3P policy like the one above – even though doing so is on shaky legal ground.
Needless to say, reading third party cookies is generally a bad idea, so if you don’t do it, then you won’t need to set a
Expires is almost unbelievably popular, considering that
Cache-Control has been preferred over
Expires for 20 years. Where a
Cache-Control header includes a
max-age directive, any
Expires header on the same response will be ignored. But there are a massive number of sites setting both, and the
Expires header is most commonly set to
Thu, 01 Dec 1994 16:00:00 GMT, because you want your content to not be cached and copy-pasting the example date from the spec is certainly one way of doing that.
But there is simply no reason to do this. If you have an
Expires header with a date in the past, replace it with:
Cache-Control: no-store, private
no-store is a very strong directive not to write the content to persistent storage, so depending on your use case you might actually prefer
no-cache for better performance, for example when using back/forward navigation or resuming hibernated tabs)
Some of the tools that audit your site will tell you to add an
X-Frame-Options header with a value of ‘SAMEORIGIN’. This tells browsers that you are refusing to be framed by another site, and is generally a good defense against clickjacking. However, the same effect can be achieved, with more consistent support and more robust definition of behaviour, by doing:
Content-Security-Policy: frame-ancestors 'self'
This has the additional benefit of being part of a header (CSP) which you should have anyway for other reasons (more on that later). So you can probably do without
X-Frame-Options these days.
Finally, back in their IE9 days, Microsoft introduced ‘compatibility view’, and would potentially render a page using the IE8 or IE7 engine, even when the user was browsing with IE9, if the browser thought that the page might require the earlier version to work properly. Those heuristics were not always correct, and developers were able to override them by using an
X-UA-Compatible header or meta tag. In fact, this increasingly became a standard part of frameworks like Bootstrap. These days, this header achieves very little - very few people are using browsers that would understand it, and if you are actively maintaining your site it’s very unlikely that you are using technologies that would trigger compatibility view.
Debug data (X-ASPNet-Version, X-Cache)
It’s kind of astonishing that some of the most popular headers in common use are not in any standard. Essentially this means that somehow, thousands of websites seem to have spontaneously agreed to use a particular header in a particular way.
X-Cache: HIT X-Request-ID: 45a336c7-1bd5-4a06-9647-c5aab6d5facf X-ASPNet-Version: 3.2.32 X-AMZN-RequestID: 0d6e39e2-4ecb-11e8-9c2d-fa7ae01bbebc
In reality, these ‘unknown’ headers are not separately and independently minted by website developers. They are typically artefacts of using particular server frameworks, software or specific vendors’ services (in this example set, the last header is a common AWS header).
X-Cache, in particular, is actually added by Fastly (other CDNs also do this), along with other Fastly-related headers like
X-Served-By. When debugging is enabled, we add even more, such as
These headers are not recognised by any browser, and removing them makes no difference to how your pages are rendered. However, since these headers might provide you, the developer, with useful information, you might like to keep a way to turning them on.
I didn’t expect to be in 2018 writing a post about the
Pragma header, but according to our HTTP Archive data it’s still the 11th most popular. Not only was Pragma deprecated as long ago as 1997, but it was never intended to be a response header anyway - as specified, it only has meaning as part of a request.
Nevertheless it’s use as a response header is so widespread that some browsers recognise it in this context as well. Today the probability that your response will transit a cache that understands
Pragma in a response context, and doesn’t understand
Cache-Control, is vanishingly small. If you want to make sure that something isn’t cached,
Cache-Control: no-store, private is all you need.
One header in our top 30 is a non-browser header.
X-Robots-Tag is intended to be consumed by a crawler, such as Google or Bing’s bots. Since it has no meaning to a browser, you could choose to only set it when the requesting user-agent is a crawler. Equally, you might decide that this makes testing harder, or perhaps that it violates the terms of service of the search engine.
Finally, it’s worth finishing on an honourable mention for simple bugs. In a request, a
Host header makes sense, but seeing it on a response probably means your server is misconfigured somehow (I’d love to know how, exactly). Nevertheless, 68 domains in HTTP archive are returning a
Host header in their responses.
Removing headers at the edge
Fortunately, if your site is behind Fastly, removing headers is pretty easy using VCL. It makes sense that you might want to keep the genuinely useful debug data available to your dev team, but hide it for public users, so that’s easily done by detecting a cookie or inbound HTTP header:
In the next post in this series, I’ll be talking about best practices for headers that you should be setting, and how to enable them at the edge.