Yes. In reverse-proxying scenarios one must rewrite the incoming headers before a second outbound request is made to the proxy target. Examples:
* `x-forwarded-*` headers * https:// to http:// support often requires rewriting location headers. * wss:// to ws:// WebSocket support: must rewrite the x-websocket-origin header (among others, don't have the spec off-hand) In the two most popular current implementations: * node-http-proxy: Reads the incoming headers off of the http.ServerRequest object, modifies them, and then makes an outbound request to the reverse proxy target via http.request * bouncy: Works behind a net.Server instance, but performs all HTTP header parsing in pure Javascript to work around the lack of framing from the core http-parser. mmalecki has fixed this and should be opening pull requests tonight or tomorrow. We'll be working on a new version of node-http-proxy which takes advantage of them after they're merged. Personally I consider this more of a bugfix than a feature. --Charlie On Sun, Feb 12, 2012 at 9:24 PM, Ben Noordhuis <[email protected]> wrote: > On Sun, Feb 12, 2012 at 23:50, Charlie Robbins > <[email protected]> wrote: > > All, > > > > I spoke with ry and substack about this issue when I was in San Francisco > > last month: https://github.com/joyent/node/issues/2612. If I remember > > correctly, it requires small changes to http-parser and the node > wrapper. I > > don't know this code base at all so I'm looking for someone to help > > implement it. > > > > The purpose here is to make HTTP proxying much simpler. The proxy logic > in > > node-http-proxy right now is implemented through an http.Server instance. > > This fix would allow it to be implemented through a net.Server instance > > using this logic: > > > > 1. Create a net server. > > (for each connection) > > 2. Buffer chunks and execute them against an HTTPParser instance. > > 3. Slice the HTTP headers from first one (or two) chunks, edit them, and > > serialize them. > > 4. Flush the buffer to the outgoing reverse proxy request. > > > > It is impossible to do (3) properly without this fix. I have a branch of > > node-http-proxy which performs a naive slice looking for the first > occurance > > of `\r\n\r\n`. > > > https://github.com/nodejitsu/node-http-proxy/blob/experimental/lib/balancing-proxy.js#L257-261 > . > > After some initial tests this approach is too naive and won't work in > > production. > > > > Any help here would be fantastic. > > > > Thanks, > > Charlie > > I don't quite understand the issue. You want the total size of the HTTP > headers? >
