the Proxy Interface¶

Module: mod_proxy_core

• Table of contents
• the Proxy Interface
• Description
• Installation
• Load Balancing
• Fail-Over Handling
• Timeouts
• Options
• Example

Description¶

FastCGI, SCGI, HTTP, ... are doing the same simple job: they forward the HTTP-request to a backend and send the response back to the server. They just use different protocols to encode the data.

• FastCGI, developed by Open Market and documented on http://www.fastcgi.com/, is a binary container around HTTP requests which reduces the parsing overhead. In addition to normal response generation it supports authorization queries
• SCGI is similar to HTTP and only adds a length header for the HTTP-header. http://www.mems-exchange.org/software/scgi/ says "The SCGI protocol is a replacement for the Common Gateway Interface (CGI) protocol. It is a standard for applications to interface with HTTP servers. It is similar to FastCGI but is designed to be easier to implement."
• for HTTP we support HTTP/1.0 and HTTP/1.1 on the backend side
• AJP13 is the Apache JServ Protocol version 1.3 (see http://tomcat.apache.org/tomcat-3.3-doc/AJPv13.html), implemented by mod_jk and mod_proxy_ajp (Apache 2.2+) in the Apache world.

A small guide shall help you to choose the right protocol for your needs.

Protocol	preferred by	binary protocol	keep-alive
HTTP	mongrel	no	yes
SCGI	WSGI (python)	not really	no
FastCGI	PHP, rails	yes	yes
AJP13	Tomcat	yes	yes

Installation¶

mod-proxy-core requires lighttpd 1.5 and is intended to replace mod_fastcgi.

If you want to use mod-proxy-core you have to load it first. Each protocol is a module too and has to be loaded
after the core-module.:

  server.modules = (
    ..., 
    "mod_proxy_core",
    "mod_proxy_backend_http",
    "mod_proxy_backend_fastcgi",
    ... )

Load Balancing¶

mod-proxy-core supports 4 different load balancers:

• the static load balancer (static) does no load balancing, only fail-over in case the first listed backend is not available. See http://permalink.gmane.org/gmane.comp.web.lighttpd/4686 for an example (basically local and remote backends).
• in Round Robin (round-robin) the requests are distributed equally over all backends. In our implementation we are not enforcing true round robin, but do a random selection of a backend.
• Shortest Queue First (sqf) is similar to round-robin and prefers the backend with the shortest wait-queue.
• the Cache Array Routing Protocol (carp) is a bit different as it is the only balancing protocol which improves the locality of the backends. It hashes the URL and send the same URL always to the same backend. http://icp.ircache.net/carp.txt explains the full spec.

Currently not implemented:

• sticky-session
• a lot of the http://www.ietf.org/proceedings/99mar/I-D/draft-melve-intercache-comproto-00.txt protocols

Fail-Over Handling¶

In case a request to the backend fails (connections refused, connection timeout, ...) the request is sent to another backend. If no backend is a available, ``504 Gateway Timeout`` is returned. (see http://forum.lighttpd.net/topic/4993)

Timeouts¶

If your incoming requests is exceeding the number of backends you have the request is put into a backlog queue. As soon as a backend is available again the first connection from the queue is woken up and is processed.

In case a request is waiting for more than 10 seconds in the queue, the request is terminated and ``504 Gateway timeout`` is returned.

Options¶

proxy-core.balancer
might be one of 'round-robin', 'sqf', 'carp' or 'static'.

See:

• http://blog.lighttpd.net/articles/2006/11/14/mod-proxy-core-and-sqf

proxy-core.protocol
might be one of 'http', 'fastcgi', 'ajp13' or 'scgi'. Make sure you load the backend modules (see mod_proxy_backend_<name>)

proxy-core.backends
tell the module where to send Proxy requests to.

It is a list of hostname/ip address/unix-domain socket

e.g.

    proxy-core.backends = ( 
      "10.0.0.1:80",     ## IPv4 address
      "unix:/tmp/php.socket", ## unix domain socket
      "[::1]:80",        ## IPv6 addresss
      "google.com:80"    ## hostname, resolved at startup
    )

proxy-core.max-pool-size
max size for pool of connections to a single backend.

• for mongrel this should be 1
• for PHP this should match the PHP_FCGI_CHILDREN setting
• for WSGI it should match the number of threads of the backend
• for HTTP proxying it is up to you how many parallel requests you want to allow to the same backend

Default: 1

proxy-core.allow-x-sendfile
enables use of "X-Sendfile/X-LIGHTTPD-Sendfile/X-LIGHTTPD-send-tempfile" headers.

Note: If you're running PHP with zlib compression, turn it off using ini_set("zlib.output_compression", "off") or the file you're trying to send will be just 1 byte and not work.

See:

• http://blog.lighttpd.net/articles/2006/07/22/mod_proxy_core-got-x-sendfile-support
• http://blog.lighttpd.net/articles/2006/11/29/faster-fastcgi

proxy-core.allow-x-rewrite
enables use of X-Rewrite

Gives you the ability to redirect requests from one backend to another without the client noticing it by using the response headers "X-Rewrite-URI" and "X-Rewrite-Host"

Suppose you want to have some kind of loadbalancing but with a more complex logic behind it than roundrobin or weighted randomization.
The client requests http://host.tld/ which is handled by your fastcgi backend which does some computation to decide which backend should be used.
All your backend has to do now is to send a 200 OK together with e.g. "X-Rewrite-URI: /foo" and "X-Rewrite-Host: internal3.host.tld"
What that does is that it will cause mod_proxy_core to handle the current request from the client again but with URI "/foo" instead of "/" and host "internal3.host.tld" instead of "host.tld"

See:

• http://blog.lighttpd.net/articles/2006/07/22/x-sendfiles-new-friend-x-rewrite

proxy-core.rewrite-request
rewrite request headers or request uri.

The following keys are currently supported:

• "_uri": The full path + querystring of the request, e.g. "/some/file/in/some/dir.fcgi?request&uri=1"
• "_docroot": Path to the document root, useful if the request goes to another server with a different file system layout
• "_pathinfo": CGI Environment variable PATH_INFO (e.g. /foobar for a request to /myfcgiapp/foobar)
• "_scriptname": CGI Environment variable SCRIPT_NAME (e.g. /myfcgiapp for a request to /myfcgiapp)

Note, see #1600.

Example:

trac needs a guessed PATH_INFO:

      $HTTP["url"] =~ "^/trac/" {
        proxy-core.backends = ( "127.0.0.1:9090" )
        proxy-core.protocol = "fastcgi" 
        proxy-core.rewrite-request = (
           "_pathinfo" => ( "^/trac(/.*)" => "$1" ),
           "_scriptname" => ( "^(/trac/)" => "$1" )
        )
      }

We build the PATH_INFO from the request-uri (the same as in the $HTTP["url"]) and extract the part after the trac-base from it as PATH_INFO. The SCRIPT_NAME is shortened accordingly.

trac will only see:
      SCRIPT_NAME=/trac
      PATH_INFO=/wiki

     For proxies can make a relative part of your URL:
      $HTTP["url"] =~ "^/blog" {
        proxy-co...

        proxy-core.rewrite-request = (
          "_uri" => ( "^/blog/?(.*)" => "/$1" ),
          "Host" => ( ".*" => "blog.example.org" ),
        )
      }

    All requests to http://example.org/blog will be fetched from http://blog.example.org/ transparently for the user.

proxy-core.rewrite-response
rewrite response headers.

proxy-core.max-keep-alive-requests
number of request before closing a keep-alive connection.

Default: 0

proxy-core.split-hostnames
If the backends uses DNS hostnames each with multiple ip address, split each address into a separate backend.
Disable this option to put the ip addresses into an address pool for that backend.

Default: enabled

Example¶

Using lighttpd + mod_proxy_core in front of 8 Squids which handle the caching of dynamic content for you. All requests for the host www.example.org should be forwarded to the proxy. All proxies listen on port 80 for requests.

  server.modules  += ( "mod_proxy_backend_http" )
  $HTTP["host"] == "www.example.org" {
    proxy-core.protocol = "http" 
    proxy-core.balancer = "carp" 
    proxy-core.backends = ( "10.0.0.10",
                            "10.0.0.11",
                            "10.0.0.12",
                            "10.0.0.13",
                            "10.0.0.14",
                            "10.0.0.15",
                            "10.0.0.16",
                            "10.0.0.17" )
  }

If one of the hosts goes down the all requests for this one server are moved equally to the other servers. If you want to know more about the algorithm used here google for 'Microsoft CARP'.

For php using unix-domain socket "/tmp/php-fastcgi.sock"

  server.modules  += ( "mod_proxy_backend_fastcgi" )
  $PHYSICAL["existing-path"] =~ "\.php$" {
    proxy-core.balancer = "round-robin" 
    proxy-core.protocol = "fastcgi" 
    proxy-core.allow-x-sendfile = "enable" 
    proxy-core.backends = ( "unix:/tmp/php-fastcgi.sock" )
    proxy-core.max-pool-size = 16
    proxy-core.rewrite-request = (
      "_pathinfo" => ( "\.php(/.*)" => "$1" )
    )
  }

For SCGI

  server.modules  += ( "mod_proxy_backend_scgi" )
  $PHYSICAL["existing-path"] =~ "\.scgi$" {
    proxy-core.balancer = "round-robin" 
    proxy-core.protocol = "scgi" 
    proxy-core.allow-x-sendfile = "enable" 
    proxy-core.backends = ( "127.0.0.1:9090" )
    proxy-core.max-pool-size = 16
  }

For http-proxy with host and file-extension conditionals

  server.modules  += ( "mod_proxy_backend_http" )
  $HTTP["host"] == "www.example.org" {
    proxy-core.protocol = "http" 
    proxy-core.balancer = "carp" 
    $HTTP["url"] =~ "\.php$" {
      proxy-core.backends = ( "10.0.0.10:80" )
    }
    else $HTTP["url"] =~ "\.scgi$" {
      proxy-core.backends = ( "10.0.0.11:80" )
    }
  }

Reverse-Proxying

  server.modules  += ( "mod_proxy_backend_http" )
  $HTTP["url"] =~ "^/proxyme(/|$)" {
    proxy-core.balancer = "round-robin" 
    proxy-core.protocol = "http" 
    proxy-core.backends = ( "en.wikipedia.org" )
    proxy-core.rewrite-response = (
      "Location" => ( "^http://en.wikipedia.org/(.*)" => "http://127.0.0.1:1025/proxyme/$1" ),
    )
    proxy-core.rewrite-request = (
      "_uri" => ( "^/proxyme/?(.*)" => "/$1" ),
      "Host" => ( ".*" => "en.wikipedia.org" ),
    )
  }

For proxying to Tomcat using AJP13 protocol

  server.modules  += ( "mod_proxy_backend_ajp13" )
  $HTTP["url"] =~ "^/tomcat/" {
    proxy-core.balancer = "round-robin" 
    proxy-core.protocol = "ajp13" 
    proxy-core.backends = ( "localhost:8009" )
    proxy-core.max-pool-size = 16
  }

Example of allow-x-rewrite config

  server.modules  += ( "mod_proxy_backend_fastcgi" )
  $PHYSICAL["existing-path"] =~ "\.php$" {
    proxy-core.balancer = "round-robin" 
    proxy-core.protocol = "fastcgi" 
    proxy-core.allow-x-rewrite = "enable" 
    proxy-core.backends = ( "unix:/tmp/php-fastcgi.sock" )
    proxy-core.max-pool-size = 16
  }
  server.modules  += ( "mod_proxy_backend_http" )
  $HTTP["host"] == "backend-cluster" {
    proxy-core.balancer = "round-robin" 
    proxy-core.protocol = "http" 
    proxy-core.max-pool-size = 4
  }

Example X-Rewrite script

  <?php
  header("X-Rewrite-URI: /");
  header("X-Rewrite-Host: backend-cluster");
  header("X-Rewrite-Backend: " . $_REQUEST['backend']); // dynamicaly add/select a backend to the "backend-cluster" proxy-core
  ?>

Proxying requests for a Rails app to a locally hosted pack of mongrels.

  $HTTP["host"] == "rails.dom.ain" { # or, e.g., =~ "^([^.])+\.rails\.tld$" 
    proxy-core.protocol = "http" 
    proxy-core.balancer = "round-robin" 
    proxy-core.allow-x-sendfile = "enable" # avoid send_file memory ballooning by using the x_send_file plugin
    proxy-core.backends = (
      "127.0.0.1:8000",
      "127.0.0.1:8001",
      "127.0.0.1:8002",
      "127.0.0.1:8003" 
    )
  }