Hey, I didn't write the darn things, no shoot the messenger :-) Still, after a little research ... this all appears to be perfectly legal ... RFC2616 13.5.2 lists the "end to end" headers that a transparent proxy is not allowed to modify, and Host isn't on the list. As I think about it though, both Host and other aspects of the resource location are often mutated in the course of my more, er, creative ... proxying. For example, I can use Squid or mod_proxy (or Restlet Redirector!) to proxy http://demosomething.solertium.com to http://mydesktophost.internal.solertium.com:8182/demo ... here the proxy needs to rewrite the Host as well as the resource URI, so my desktop computer can figure out which of the umpty-odd services I'm currently running to serve it with. Great for this kind of demo/dev application. Performance is not a plus of the setup. I think you are looking for more of a "content switch" which I don't think is usually implemented as a true HTTP proxy ... like F5's BigIP products. Here -- based on my limited experience with BigIP -- requests are not mutated and repeated, they are preserved transparently end-to-end as much as possible, but selectively sent to the best available working node in a cluster. It's highly magical, and lovely if you can afford one. Which, for the record, I cannot.
Not that you can't use Squid or mod_proxy or whatever to front a cluster ... we have a lot of clients who do this in their enterprise ... but typically this works best for ordinary content traffic (GET html, jpg, etc.) and where the cost of generating that traffic far exceeds the cost of delivering it. So, if one has a cluster of poky AquaLogic portal servers that need to spend a minute getting out of their own way to render a Web page (*), putting a speedy and aggressively caching Squid in front of the mess can effectively hide the ugly. But I doubt it would improve upon your properly constructed Restlet OSGi applications that ideally leverage the web(tm) ;-) John Mitchell should be along presently to straighten both of us out on this subject. It's Friday, I'm gonna go out and play. - R * anti JEE prejudice is showing On Fri, Apr 24, 2009 at 6:38 PM, David Fogel <carrotsa...@gmail.com> wrote: > Hi Rob- > > I'm sure I'm just revealing my own ignorance about proxies here, but > I'm wondering if you could go into greater detail regarding why > reverse proxies would change the Host header. Seems to me that the > Host header is a fundamental part of the client's request, and is > independent from IP-level network routing. > ------------------------------------------------------ http://restlet.tigris.org/ds/viewMessage.do?dsForumId=4447&dsMessageId=1899907
