Proposing HTTP Request Forwarding

I’ve been monitoring the ietf-http-wg mailing list and have noticed there is renewed actively around revising RFC 2616, the HTTP/1.1 specification. This renewed activity got me thinking it was time to discuss the need for HTTP Request Forwarding.

I’ll start by saying it is possible that Request Forwarding is already part of the HTTP spec and that I just overlooked it. Lord knows I’ve read enough W3C and IETF specs lately to raze a small forest if printed, so I could easily have missed that part as I read bleary-eyed through the specs. But I’ve asked the question privately of a few people that should know and they all said that HTTP does not support request forwarding; one of them pointed out that that is why VOIP needed SIP.

If the term “HTTP Request Forwarding” isn’t obvious from context, let me give a use-case to illustrate:

USE-CASE: Retaining control of URL Interfaces when outsourcing image hosting

  1. http://example.com
  2. http://image-servers-r-us.net
  3. http://example.com/index.html
  4. http://example.com/images/splash.png
  5. http://image-servers-r-us.net/example.com/splash.png

Assume client “A”, server “B” mapped to [1] and server “C” mapped to [2] (servers “B” and “C” are different computers with different IP addresses probably at different locations.) For this use case client “A” requests the HTML file at [3] which includes an <img> tag pointing to a graphic at [4].

The request from client “A” for the .html file at [3] goes to server “B” and the response is returne to client “A.” After parsing [3] client “A” realizes it also needs to download the image at [4] and requests the .png file from server “B” at [1].

HTTP Request Forwarding (Note use of HTTP 1.2 to make URL in example explicit)

However, server “B” knows that the .png file [3] is actually located on server “C” at [5] so it forwards the request to server “C” at [2]. Server “C” responds by returning the .png file directly back to client “A” and client “A” is none the wiser; i.e. client “A” still thinks that the image was returned by [4].

In addition, if client “A” is a browser and the user inspects the properties of the image, the user would find the URL to be [4], not [5]. No where in the response given to the client would there be information that the image came from [5] instead of [4] except that the IP address differs from the server that requested it. It is possible that a header could contain the forwarding information, but the client would not need to act on it except potentially for debugging.

There are several other use-cases as well, such as making it possible to fully virtualize a domain authority’s URL interface. However, the use-case of outsourcing images or any large static content illustrates a benefit that I believe should be obvious to most people. And for those that are interested, especially if you are involved in updating the spec, I could detail other use-cases if people are interested.

Now it is possible there are security implications with this that I have not considered. If so, I would hope that we could at least explore potential ways to mitigate those issues as opposed to immediately dismissing HTTP Request Forwarding out of hand.

In summary, adding a Request Forwarding functionality to HTTP would allow servers to maintain control of their URL interfaces while still being able to distribute loads and services to appropriate servers. Having such a feature could improve consistency in URL design and make it easier for website owners to restructure their websites without the level of broken links seen on the web today.

Interestingly, on Tuesday Scott Hanselman blogged about just this problem from a slightly different perspective; I present Scott’s post for your consideration.

This entry was posted in Call to Action, For Comment, HTTP, Potential, Standards Participants. Bookmark the permalink.

3 Responses to Proposing HTTP Request Forwarding

  1. I notice that you did not mention redirects at all in this post.

    What’s the difference between this, and server “B” issuing a redirect response for splash.png? In the image-hosting case, the user will never see an URL that points to the image-hosting service. The URLs in the document will point to “B”, and the browser never puts the URL for an “img src” into the location bar.

  2. Tom Barta says:

    The problem with doing something like this is the underlying TCP mechanisms. When client A establishes a connection with server B, there is an open TCP connection. A can be behind a firewall or a NAT, and B’s response will come in fine because the connection was already established.

    C, on the other hand, cannot directly send content to A (at least, not reliably). At present, it’s up to A to re-request the content from C, because an HTTP response must be followed by a request in order to work with the current model of the Internet (unless everyone wants to switch to IPv6 right now).

    Because HTTP is client-server and not peer-peer, there needs to be a URL at C to access the content. Unless you have a rewriting proxy sitting somewhere in the middle, you’ll always have to deal with secondary sub-par URLs being visible to an HTTP client.

  3. chary says:

    http://wiki.welldesignedurls.org/URL_Interfaces

    Parse error: syntax error, unexpected T_STRING, expecting T_OLD_FUNCTION or T_FUNCTION or T_VAR or ‘}’ in /home/welldesi/public_html/wiki/includes/Exception.php on line 114

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>