Proloxy - Prolog Reverse Proxy


A reverse proxy relays requests to different web services. The main advantage is isolation of concerns: You can run different and independent web services and serve them under a common umbrella URL.

Proloxy is a reverse proxy that is written in Prolog.

Source code:

You can track development via a public git repository:

Proloxy requires SWI-Prolog 7.5.8 or later.


Proloxy uses an extensible Prolog predicate to relay requests to different web services: For each arriving HTTP request, Proloxy calls the predicate request_prefix_target(+Request, -Prefix, -Target). Its arguments are: You configure Proloxy by providing a Prolog file that contains the definition of request_prefix_target/3 and any additional predicates and directives you need. For each web service you want to make available, add a clause of request_prefix_target/3 to relate an instantiated HTTP request to a prefix and the desired target. Each clause may use arbitrary Prolog code to analyse the request and form the target.

When dispatching an HTTP request, Proloxy considers the clauses of request_prefix_target/3 in the order they appear in your configuration file and commits to the first clause that succeeds. It relays the request to the computed target, and then sends the target's response to the client.

For example, the following clause relays all requests to a local web server on port 3031, passing along the original request path. The target server can for example host the site's main page, to be used if no other rules apply:
request_prefix_target(Request, '', Target) :-
        memberchk(request_uri(URI), Request),
        atomic_list_concat(['http://localhost:3031',URI], Target). shows a sample configuration file that uses Prolog rules to dispatch requests to two different web services.

Virtual hosts

The rule language is general enough to express virtual hosts. In the case of name-based virtual hosts, this means that you dispatch requests to different web services based on the domain name that is used to access your site.

For example, to dispatch all requests of users who access your server via to a web server running on port 4040 (while leaving the path unchanged), use:
request_prefix_target(Request, '', Target) :-
        memberchk(host(''), Request),
        memberchk(request_uri(URI), Request),
        atomic_list_concat(['http://localhost:4040',URI], Target).
Using this method, you can host multiple domains with a single Proloxy instance, dispatching requests to different underlying services.

Redirections and other replies

You can also use the predicates http_404/2 and http_redirect/3 from library(http/http_dispatch) in your configuration files.

For example, the following snippet responds with "HTTP 404 not found" if the URI contains .git:
:- use_module(library(http/http_dispatch)).

request_prefix_target(Request, _, _) :-
        memberchk(request_uri(Path), Request),
        sub_atom(Path, _, _, _, '.git/'),
        http_404([], Request).
In some cases, it is convenient to respond directly with plain text or HTML content instead of relaying the request to a different web service. If a clause of request_prefix_target/3 emits any text on standout output, then this output is sent to the client as the HTTP response. Such responses typically start with Content-type: text/plain (or text/html), followed by two newlines and the body of the reply. In rules that emit output, Target must be the atom - to avoid relaying the request to a different service. Proloxy provides the predicate output_from_process(+Program, +Args) to emit process output (from stdout and stderr) on standard output. For example, we can configure Proloxy to show the system's uptime when the URL /uptime is accessed:
request_prefix_target(Request, _, -) :-
        memberchk(request_uri('/uptime'), Request),
        format("Content-type: text/plain; charset=utf-8~n~n"),
        output_from_process('/usr/bin/uptime', []).

Auxiliary programs and scripts can be conveniently invoked with this method.

Relaying header fields

The extensible predicate transmit_header_field/1 allows you to relay header fields that the target service emits to the client. The argument is the name of the header field you want to transmit if it exists in the target's response. For example, you can put the following in
The name of the header field is matched case-insensitively and underscore (_) matches hyphen (-).

By default, Proloxy does not relay any response header fields.

Here is a recommended configuration:
Further, the extensible predicate add_header/1 lets you add custom header fields to the response. An example is:
add_header(strict_transport_security('max-age=63072000; includeSubdomains')).
This enables HTTP Strict Transport Security (HSTS), which is useful when running HTTPS servers.

Testing the configuration

Since each configuration file is also a valid Prolog program, you can easily test your configuration. Consulting the Prolog program in SWI-Prolog lets you detect syntax errors and singleton variables in your configuration file. To test whether HTTP requests are dispatched as you intend, query request_prefix_target/3. For example:
$ swipl
Welcome to SWI-Prolog (Multi-threaded, 64 bits, Version 7.3.14)

?- once(request_prefix_target([request_uri(/)], P, T)).
P = '',
T = 'http://localhost:3031/'.

?- once(request_prefix_target([request_uri('/rits/demo.html')], P, T)).
P = '/rits',
T = 'http://localhost:4040/demo.html'.
Note that:
  1. we are using once/1 to commit to the first clause that succeeds.
  2. we are simulating an actual HTTP request, using a list of header fields.
  3. the answers tell us how the given HTTP requests are dispatched.
The ability to conveniently test your configuration is a nice property, and a natural consequence of using Prolog as the configuration language. You can also write unit tests for your configuration and therefore easily detect regressions.

Running Proloxy

You can run Proloxy as a Unix daemon. See the SWI-Prolog documentation for invocation options.

In the following, assume that your proxy rules are stored in the file called To run Proloxy as a Unix daemon on the standard HTTP port (80) as user web, use:
$ sudo swipl --user=web
To run the process in the foreground and with a Prolog toplevel, use:
$ sudo swipl --user=web --interactive
You can also use a different port that does not need root privileges:
$ swipl --interactive --port=3040

Launching Proloxy on system startup

proloxy.service is a sample systemd unit file that runs Proloxy on system startup. Adapt the paths and options as needed, copy the file to /etc/systemd/system/ and install it using:
$ sudo systemctl enable /etc/systemd/system/proloxy.service
$ sudo systemctl start proloxy.service

Security: Run HTTPS servers

You can run Proloxy as an HTTPS server and thus encrypt traffic for all hosted services at once. See LetSWICrypt for more information. A common use case when using HTTPS is to run a second Proloxy instance as a regular HTTP server on port 80 to redirect each request for http://X to https://X. You can do this with the following configuration file for the HTTP server:
:- use_module(library(http/http_dispatch)).

/* - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
   Redirect each request X to https://X
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - */

request_prefix_target(Request, _, _) :-
        memberchk(request_uri(URI), Request),
        memberchk(host(Host), Request),
        atomic_list_concat(['https://',Host,URI], Target),
        http_redirect(moved, Target, Request).

WebSocket connections

Proloxy supports proxying of WebSocket connections.

As an example, consider making noVNC available via /vnc/, assuming that noVNC listens on port 6080. The following clauses accomplish the configuration:
request_prefix_target(Request, '/vnc', Target) :-
        memberchk(request_uri(URI), Request),
        atom_concat('/vnc', Rest, URI),
        atomic_list_concat(['http://localhost:6080',Rest], Target).

request_prefix_target(Request, '/websockify', Target) :-
        memberchk(request_uri(URI), Request),
        atom_concat('/websockify', Rest, URI),
        atomic_list_concat(['ws://localhost:6080',Rest], Target).
WebSocket connections are automatically detected via the Upgrade: websocket and other header fields.

Serving very large files

In typical use cases, Proloxy relays requests to other web servers, and sends their answers to the client. The overhead is typically negligible, since the other web services usually reside on the same machine.

However, if a web server sends very large files in response to some requests, Proloxy may not have enough global stack space to collect the response. In such cases, one solution is to configure Proloxy so that such large files are sent directly by Proloxy, without involving a different web service. For example, the following snippet configures Proloxy to directly send any files (such as ISO images) that are located in /home/web/iso, and are accessed via /iso/.
:- use_module(library(http/http_dispatch)).

request_prefix_target(Request, '', _) :-
        memberchk(request_uri(URI), Request),
        atom_concat('/iso/', Rest, URI),
        http_safe_file(Rest, []),
        atom_concat('/home/web/iso', URI, Path),
        http_reply_file(Path, [unsafe(true)], Request).

More about Prolog: The Power of Prolog

In particular: Web Applications

Main page