Cro::HTTP::ReverseProxy
A reverse proxy accepts incoming requests and delegates them to another HTTP server, passing that server's response back to the client. A reverse proxy may modify the request and/or response, and route different requests to different target servers.
Potential applications of a HTTP reverse proxy include:
Providing a common endpoint that delegates to multiple backend services
Inserting extra headers (for example, mapping a session token from the browser into a JSON Web Token expected by backend services)
Load-balancing requests over multiple backend services
Aiding A/B testing by routing requests to partiuclar versions of a service
Providing authenticated access to a non-authenticated service (assuming the non-authenticated service is not exposed directly)
SSL termination
Caching
Content transformation
Proxying all incoming requests§
The class Cro::HTTP::ReverseProxy is a Cro::Transform
that consumes a Cro::HTTP::Request and produces a Cro::HTTP::Response. Therefore, it may be used as the application
parameter to Cro::HTTP::Server. This setup is ideal if all the incoming requests should be proxied, since it avoids the overhead of Cro::HTTP::Router.
First, set up the reverse proxy, specifying to
, which is the target base URL (that is, the incoming target will be appended to it):
my $proxy = Cro::HTTP::ReverseProxy.new:
to => 'http://the.target.base/url/';
Then host it:
my $server = Cro::HTTP::Server.new:
port => 10101,
application => $proxy;
$server.start;
react whenever signal(SIGINT) {
$server.stop;
}
Proxying just some routes§
Since Cro::HTTP::ReverseProxy is a Cro::Transform
from HTTP requests to HTTP responses, it may be used with Cro::HTTP::Router's delegate
function. This is useful when wanting to directly handle some requests, while proxying others.
my $app = route {
get -> {
content 'text/plain', 'Unproxied URL';
}
# /user/foo proxied to http://user-service/foo
delegate <user *> => Cro::HTTP::ReverseProxy.new:
to => 'http://user-service/';
# /product/foo proxied to http://product-service/foo
delegate <product *> => Cro::HTTP::ReverseProxy.new:
to => 'http://product-service/';
}
Proxying without appending the target URL§
Sometimes, one might wish to proxy all requests to a single target URL, without appending the target of the request. This is primarily useful in conjunction with other features (such as manipulating the request). Instead of to
, pass to-absolute
:
my $app = route {
# Everything under /images/ proxied to a really cute cat picture
# (provide your own, or there's just a few on the internet...)
delegate <images *> => Cro::HTTP::ReverseProxy.new:
to-absolute => 'http://really-cute.cat/picture.gif';
}
Controlling the target URL§
Sometimes, it's desirable to proxy different requests to different URLs. This can be achieved by passing a Code
object to to
or to-absolute
.
For example, picking two different servers to proxy to at random could be achieved with:
my @servers = 'http://replica-a/', 'http://replica-b/';
my $proxy = Cro::HTTP::ReverseProxy.new:
to => { @servers.pick };
The request object is passed in to
, allowing the target to be determined based upon request properties:
my $proxy = Cro::HTTP::ReverseProxy.new:
to => {
.has-header('authorization')
?? 'http://private-service/'
!! 'http://public-service/'
};
If doing significant amounts of work in the code block, for example doing a database query, and if the proxy is to support HTTP/2.0 requests, then return an Awaitable
object (for example, a Promise
) that will be completed with the target URL. This avoids blocking processing of other HTTP/2.0 requests multiplexed on the same client connection.
Everything works in the same way for to-absolute
, except that the result will be used as the absolute target URL.
Transforming the request§
To transform the request before it is proxied to the target, pass the code to transform it using the request
named argument. The code will be called with the Cro::HTTP::Request object.
my $proxy = Cro::HTTP::ReverseProxy.new:
to => 'http://the.target.base/url/',
request => {
.add-header('X-Experiment', <A B>.pick);
};
If wishing to do something involving the network or some other time-consuming operation, write this asynchronously to avoid blocking processing of HTTP/2.0 multiplexed requests:
my $proxy = Cro::HTTP::ReverseProxy.new:
to => 'http://the.target.base/url/',
request => -> $req {
start {
my $jwt = await jwt-lookup($req.cookie-value('session-id'));
$req.add-header('Authorization', "Bearer $jwt");
}
};
It is also possible to obtain the request body and transform that, using set-body
to put in place the new request body.
my $asciify-proxy = Cro::HTTP::ReverseProxy.new:
to => 'http://the.target.base/url/',
request => -> $req {
start {
if $req.content-type.type-and-subtype eq 'application/json' {
my $body = await $req.body-text;
$req.set-body:
$body.subst(/<-[\x00..\x7F]>/, '?', :g);
}
}
};
Transforming the response§
To transform the response from the target before it is returned to the client, pass the code to transform it using the response
named argument. The Cro::HTTP::Response object will be passed to the code as an argument.
my $proxy = Cro::HTTP::ReverseProxy.new:
to => 'http://the.target.base/url/',
response => {
.remove-header('server');
};
If doing a long-running operation or transforming the body, then it is better that the transform code works asynchronously. Any Awaitable
returned will be automatically awaited.
my $analytics-insertion-proxy = Cro::HTTP::ReverseProxy.new:
to => 'http://the.target.base/url/',
response => -> $res {
start {
if $res.content-type.type-and-subtype eq 'text/html' {
my $body = await $res.body-text;
$res.set-body:
$body.subst(/<?before '</body>'>/, $analytics-code);
}
}
};
Per-request state§
Sometimes it's desirable to keep some state for each request that is being proxied, and refer to it in the request, response, or dynamic to
callbacks. The $*PROXY-STATE
dynamic variable is set up when each of the callbacks are made, and can be assigned to and accessed as needed.
For example, an A/B testing proxy could be set up as follows:
my %experiments =>
A => 'http://service-version-a/',
B => 'http://service-version-b/';
my class ProxyState {
has Str $.experiment is required;
}
my $proxy = Cro::HTTP::ReverseProxy.new:
to => sub ($request) {
my $exp;
with $request.cookie-value('ab-experiment') {
$exp = $_ if %experiments{$_}:exists;
}
$exp //= %experiments.keys.pick;
$*PROXY-STATE = ProxyState.new(experiment => $exp);
return %experiments{$exp};
},
response => {
.set-cookie('ab-experiment', $*PROXY-STATE.experiment);
};
Thread safety§
The callbacks passed to Cro::HTTP::ReverseProxy may be called on multiple threads simultaneously. However, they are called one at a time for a given request (so the to
and request
callbacks can never run at the same time for a particular request), meaning that use of the per-request $*PROXY-STATE
is always safe.
If the proxy needs to deal with other state, then that state should be given suitable protection (for example, by using OO::Monitors
and placing it in a monitor
).