This chapter is for the webdeveloper course only
This chapter covers:
- Protocol basics
- Requests and responses
- HTTP request/response control flow
- Statelessness and idempotence
- Cache related headers
HTTP is at the heart of Varnish, or rather the model HTTP represents.
This chapter will cover the basics of HTTP as a protocol, how it’s used in the wild, and delve into caching as it applies to HTTP.
- Hyper-Text Transfer Protocol, HTTP, is at the core of the web
- Specified by the IETF, the latest version (HTTP/1.1) is available from http://tools.ietf.org/html/rfc2616
- A request consists of a request method, headers and an optional request body.
- A response consists of a response status, headers and an optional response body.
- Multiple requests can be sent over a single connection, in serial.
- Clients will open multiple connections to fetch resources in parallel.
HTTP is a networking protocol for distributed systems. It is the foundation of data communication for the Web. The development of this standard is done by the IETF and the W3C. The latest version of the standard is HTTP/1.1.
A new version of HTTP called HTTP bis is under development, you can follow the work document at http://datatracker.ietf.org/wg/httpbis/charter/. Basically HTTP bis will be HTTP/1.1 with new features for example a better caching of web pages.
- Standard request methods are: GET, POST, HEAD, OPTIONS, PUT, DELETE, TRACE, or CONNECT.
- This is followed by a URI, e.g: /img/image.png or /index.html
- Usually followed by the HTTP version
- A new-line (CRLF), followed by an arbitrary amount of CRLF-separated headers (Accept-Language, Cookie, Host, User-Agent, etc).
- A single empty line, ending in CRLF.
- An optional message body, depending on the request method.
Each request has the same, strict and fairly simple pattern. A request method informs the web server what sort of request this is: Is the client trying to fetch a resource (GET), or update some data(POST)? Or just get the headers of a resource (HEAD)?
There are strict rules that apply to the request methods. For instance, a GET request can not contain a request body, but a POST request can.
Similarly, a web server can not attach a request body to a response to a HEAD body.
GET / HTTP/1.1 Host: localhost User-Agent: Mozilla/5.0 (Macintosh; U; Intel Mac OS X 10.5; fr; rv:184.108.40.206) Gecko/20110319 Firefox/3.6.16 Accept: text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8 Accept-Language: fr,fr-fr;q=0.8,en-us;q=0.5,en;q=0.3 Accept-Encoding: gzip,deflate Accept-Charset: ISO-8859-1,utf-8;q=0.7,*;q=0.7 Keep-Alive: 115 Connection: keep-alive Cache-Control: max-age=0
The above is a typical HTTP GET request for the
Note that the
Host-header contains the hostname as seen by the
browser. The above request was generated by entering http://localhost/
in the browser. The browser automatically adds a number of headers. Some
of these will vary depending on language settings, others will vary
depending on whether the client has a cached copy of the page already,
or if the client is doing a refresh or forced refresh.
Whether the server honors these headers will depend on both the server in question and the specific header.
The following is an example of a HTTP request using the POST method, which includes a request body:
POST /accounts/ServiceLoginAuth HTTP/1.1 Host: www.google.com User-Agent: Mozilla/5.0 (Macintosh; U; Intel Mac OS X 10.5; fr; rv:220.127.116.11) Gecko/20110319 Firefox/3.6.16 Accept: text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8 Accept-Language: fr,fr-fr;q=0.8,en-us;q=0.5,en;q=0.3 Accept-Encoding: gzip,deflate Accept-Charset: ISO-8859-1,utf-8;q=0.7,*;q=0.7 Keep-Alive: 115 Connection: keep-alive Referer: https://www.google.com/accounts/ServiceLogin Cookie: GoogleAccountsLocale_session=en;[...] Content-Type: application/x-www-form-urlencoded Content-Length: 288 ltmpl=default[...]&signIn=Sign+in&asts=
HTTP/1.1 200 OK Cache-Control: max-age=150 Content-Length: 150 [data]
- A HTTP response contains the HTTP versions, response code(e.g: 200) and response message (e.g: OK).
- CRLF as line separator
- A number of headers
- Headers are terminated with a blank line.
- Optional response body
The HTTP response is similar to the request itself. The response code informs the browser both whether the request succeeded and what type of response this is. The response message is a text-representation of the same information, and is often ignored by the browser itself.
Examples of status codes are 200 OK, 404 File Not Found, 304 Not Modified and so fort. They are all defined in the HTTP standard, and grouped into the following categories:
- 1xx: Informational - Request received, continuing process
- 2xx: Success - The action was successfully received, understood, and accepted
- 3xx: Redirection - Further action must be taken in order to complete the request
- 4xx: Client Error - The request contains bad syntax or cannot be fulfilled
- 5xx: Server Error - The server failed to fulfill an apparently valid request
HTTP/1.1 200 OK Server: Apache/2.2.14 (Ubuntu) X-Powered-By: PHP/5.3.2-1ubuntu4.7 Cache-Control: public, max-age=86400 Last-Modified: Mon, 04 Apr 2011 04:13:41 +0000 Expires: Sun, 11 Mar 1984 12:00:00 GMT Vary: Cookie,Accept-Encoding ETag: "1301890421" Content-Type: text/html; charset=utf-8 Content-Length: 23562 Date: Mon, 04 Apr 2011 09:02:26 GMT X-Varnish: 1886109724 1886107902 Age: 17324 Via: 1.1 varnish Connection: keep-alive (data)
HTTP request/response control flow¶
The client sends an HTTP request to the server which returns an HTTP response with the message body.
Statelesness and idempotence¶
- HTTP is by definition a stateless protocol which means that in theory your browser has to reconnect to the server for every request. In practice there is a header called Keep-Alive you may use if you want to keep the connection open between the client (your browser) and the server.
- Imdempotence means that an operation can be applied multiple times without changing the result. GET and PUT HTTP request are expected to be idempotent whereas POST requests are not. In other words, you can not cache POST HTTP responses.
Exercise : Test various Cache headers¶
Before we talk about all the various cache headers and cache mechanisms, we will use httpheadersexample.php to experiment and get a sense of what it’s all about.
Try both clicking the links twice, hitting refresh and forced refresh (usually done by hitting control-F5, depending on browser).
- Try out the Expires-header and see how the browser and Varnish behave.
- What happens when both Expires and Cache-Control is present?
- Test the If-Modified-Since request too. Does the browser issue a request to Varnish? If the item was in cache, does Varnish query the web-server?
- Try the Vary-test by using two different browsers at the same time.
When performing this exercise, try to see if you can spot the patterns. There are many levels of cache on the Web, and you have to think about more than just Varnish.
If it hasn’t already, it’s likely that browser cache will confuse you at least a few times through this course. When that happens, pull up varnishlog or another browser.
The Expires response header field gives the date/time after which the response is considered stale. A stale cache item will not be returned by any cache (proxy cache or client cache).
The syntax for this header is:
Expires: GMT formatted date
It is recommended not to define Expires too far in the future. Setting it to 1 year is usually enough.
Using Expires does not prevent the cached resource to be updated. If a resource is updated changing its name (by using a version number for instance) is possible.
The Cache-Control header field specifies directives that must be applied by all caching mechanisms (from proxy cache to browser cache). Cache-Control accepts the following arguments (only the most relevant are described):
- public: The response may be cached by any cache.
- no-store: The response body must not be stored by any cache mechanism;
- no-cache: Authorizes a cache mechanism to store the response in its cache but it must not reuse it without validating it with the origin server first. In order to avoid any confusion with this argument think of it as a “store-but-do-no-serve-from-cache-without-revalidation” instruction.
- max-age: Specifies the period in seconds during which the cache will be considered fresh;
- s-maxage: Like max-age but it applies only to public caches;
- must-revalidate: Indicates that a stale cache item can not be serviced without revalidation with the origin server first;
Unlike Expires, Cache-Control is both a request and a response header, here is the list of arguments you may use for each context:
Example of a Cache-Control header:
Cache-Control: public, must-revalidate, max-age=2592000
As you might have noticed Expires and Cache-Control do more or less the same job, Cache-Control gives you more control though. There is a significant difference between these two headers:
- Cache-Control uses relative times in seconds, cf (s)max-age
- Expires always returns an absolute date
Cache-Control always overrides Expires.
By default, Varnish does not care about the Cache-Control request header. If you want to let users update the cache via a force refresh you need to do it yourself.
The Last-Modified response header field indicates the date and time at which the origin server believes the variant was last modified. This headers may be used in conjunction with If-Modified-Since and If-None-Match.
Example of a Last-Modified header:
Last-Modified: Wed, 01 Sep 2004 13:24:52 GMT
The If-Modified-Since request header field is used with a method to make it conditional:
- if the requested variant has not been modified since the time specified in this field, an entity will not be returned from the server;
- instead, a 304 (not modified) response will be returned without any message-body.
Example of an If-Modified-Since header:
If-Modified-Since: Wed, 01 Sep 2004 13:24:52 GMT
The If-None-Match request header field is used with a method to make it conditional.
A client that has one or more entities previously obtained from the resource can verify that none of those entities is current by including a list of their associated entity tags in the If-None-Match header field.
The purpose of this feature is to allow efficient updates of cached information with a minimum amount of transaction overhead. It is also used to prevent a method (e.g. PUT) from inadvertently modifying an existing resource when the client believes that the resource does not exist.
Example of an If-None-Match header :
The ETag response header field provides the current value of the entity tag for the requested variant. The idea behind Etag is to provide a unique value for a resource’s contents.
Example of an Etag header:
The Pragma request header is a legacy header and should no longer be used.
Some applications still send headers like
Pragma: no-cache but this is
for backwards compatibility reasons only.
Any proxy cache should treat
Pragma: no-cache as
no-cache, and should not be seen as a reliable header especially when
used as a response header.
The Vary response header indicates the response returned by the origin server may vary depending on headers received in the request.
The most common usage of Vary is to use
Vary: Accept-Encoding, which
tells caches (Varnish included) that the content might look different
depending on the Accept-Encoding-header the client sends. In other words:
The page can be delivered compressed or uncompressed depending on the
The Vary-header is one of the trickiest headers to deal with for a cache. A cache, like Varnish, does not necessarily understand the semantics of a header, or what part triggers different variants of a page.
As a result, using
Vary: User-Agent for instance tells a cache that
for ANY change in the User-Agent-header, the content might look
different. Since there are probably thousands of User-Agent strings
out there, this means you will drastically reduce the efficiency of any
An other example is using
Vary: Cookie which is actually not a bad
idea. Unfortunately, you can’t issue
Vary: Cookie(but only THESE
cookies: ...). And since a client will send you a great deal of
cookies, this means that just using
Vary: Cookie is not necessarily
sufficient. We will discuss this further in the Content Composition
From Varnish version 3, Varnish handles Accept-Encoding and
Accept-Encoding for you. This is because Varnish 3 has support for
gzip compression. In Varnish 2 it was necessary to normalize the
Accept-Encoding-header, but this is redundant in Varnish 3.
- A cache server can send an additional response header, Age, to indicate the age of the response.
- Varnish (and other caches) does this.
- Browsers (and Varnish) will use the Age-header to determine how long to cache.
- E.g: for a max-age-based equation: cache duration = max-age - Age
- If you allow Varnish to cache for a long time, the Age-header could effectively disallow client-side caches.
Consider what happens if you let Varnish cache content for a week, because you can easily invalidate the cache Varnish keeps. If you do not change the Age-header, Varnish will happily inform clients that the content is, for example, two days old, and that the maximum age should be no more than fifteen minutes.
Browsers will obey this. They will use the reply, but they will also realize that it has exceeded its max-age, so they will not cache it.
Varnish will do the same, if your web-server emits and Age-header (or if you put one Varnish-server in front of another).
We will see in later chapters how we can handle this in Varnish.
Header availability summary¶
The table below lists HTTP headers seen above and wether they are a request header or a response one.
Cache-hit and misses¶
There is a cache-hit when Varnish returns a page from its cache instead of forwarding the request to the origin server.
There is a cache-miss when Varnish has to forward the request to the origin server so the page can be serviced.
Exercise: Use article.php to test Age¶
- Modify the article.php-script to send an Age header that says 30 and
- Send a request to Varnish for article.php. See what Age-Header Varnish replies with.
- Is the Age-header an accurate method to determine if Varnish made a cache hit or not?
- How long does Varnish cache the reply? How long would a browser cache it?
Also consider how you would avoid issues like this to begin with. We do not yet know how to modify Varnish’ response headers, but hopefully you will understand why you may need to do that.
Varnish is not the only part of your web-stack that parses and honors cache-related headers. The primary consumer of such headers are the web browsers, and there might also be other caches along the way which you do not control, like a company-wide proxy server.
By using s-maxage instead of max-age we limit the number of consumers to cache servers, but even s-maxage will be used by caching proxies which you do not control.
In the next few chapters, you will learn how to modify the response headers Varnish sends. That way, your web-server can emit response headers that are only seen and used by Varnish.