2013 twenty-four merry days of Perl Feed

Web done better!

PSGI/Plack - 2013-12-07

Remembering the horror

If you've done any web programming in Perl for more than a few years, you probably know how difficult and annoying server-side web programming can be. If you're new, just hold on to your hat for a few minutes.

So you want to write a server-side web application?

Welcome! How would you like to write your web application? Do you want to use CGI.pm, mod_perl, FCGI, Catalyst, Jifty? Maybe you want to write it yourself manually? Each form means a different output. Each server (Apache, Lighttpd, Nginx, etc.) supports some technologies but not others, and requires data in a different format.

You're probably beginning to see the problem. In web programming you're basically marrying yourself to a specific server environment. Beyond that, each framework usually settles on a specific server environment protocol, whether it is CGI or mod_perl. Then, if the web server doesn't work, or if you want to move to a different protocol, your web application needs rewriting. Yay! Oh wait, no. That sucks.

Great artists are inspired

Picasso said that good artists copy, great artists steal, but a better paraphrase might be that great artists are inspired. As such, Perl superhero Tatsuhiko Miyagawa was inspired by specifications available in other languages (Ruby's Rack, Python's WSGI) that provide an interface layer between different server environment protocols. He created Perl's PSGI.

PSGI finally provides a clear specification for interfacing with different layers. This effectively separates between the web application, the interface (HTTP::Engine, CGI, FCGI), and the web server (Apache, Nginx, HTTP::Server::Simple), which up until now were mashed together like a an awful software train wreck.

Do you want to write your application in Catalyst? Maybe in Web::Simple? As long as they support the PSGI protocol, you can do that. Now you can deploy them on any web server, use any technology (whether it's CGI, FastCGI, or anything else), and even run them on Perl-based web servers as a backend to a reverse-proxy server (Nginx, Apache + mod_proxy, and so on).

PSGI comes with a reference implementation called Plack, which includes a range of utilities, and implementations for servers (under the Plack::Server namespace).

Welcome to a Modern Perl web

All modern Perl web frameworks now support the PSGI protocol (Catalyst, Dancer, Web::Simple, Mojolicious, Mason, OX, Tatsumaki, and more), and you can deploy on any web server, including Perl web servers (Starman, Starlet, Twiggy, Corona).

PSGI also specifies different layers of action called middleware, allowing you to add blocks on top of your web application providing a range of services, such as caching, privacy and authorization control, REPL, debugging screens, session management and more.

Words are Cheap

To anyone with an appetite for understanding, this is probably not enough. Seeing is believing, isn't it? What does a PSGI app really look like?

In essence, a PSGI application is just a code reference:

my $app = sub {
# your application

That reference is then sent to the server. For every new HTTP request, the server will call the code reference and send it the environment variables of the request in a hash structure, conventionally called $env.

$env = {
   HTTP_ACCEPT => "*/*",
   HTTP_HOST => "localhost:5000",
   HTTP_USER_AGENT => "curl/7.27.0",
   PATH_INFO => "/",
   psgi.errors => *main::STDERR,
   psgi.input => *HTTP::Server::PSGI::input,
   psgi.multiprocess => "",
   psgi.multithread => "",
   psgi.nonblocking => "",
   psgi.run_once => "",
   psgi.streaming => 1,
   psgi.url_scheme => "http",
   psgi.version => [ 1, 1 ],
   psgix.harakiri => 1,
   psgix.input.buffered => 1,
   psgix.io => *Symbol::GEN1,
   QUERY_STRING => "",
   REMOTE_ADDR => "",
   REMOTE_PORT => 37063,
   REQUEST_URI => "/",
   SCRIPT_NAME => "",
   SERVER_NAME => 0,
   SERVER_PORT => 5000,

You can use this environment hash to understand the request. Alternatively you could use one of the available PSGI-supported web frameworks. They abstract the code that handled these environment variables and allows you to define code paths for your web application.

Next comes the reply. With CGI, you would return two chunks of output to STDOUT: one of headers to the browser, and the other of the actual content.

PSGI needs the same information from you, but in a structured form:

my $app = sub {
    my $env = shift;

    return [
        '200', # HTTP code
            'Content-Type' => 'application/json',
        [ @content_hunks ],

Once the server calls the code reference with the environment, it awaits this reply. It will then use this reply to return the proper data to the user.

And the middlewares?

Oh right! The middlewares! Since it is all code references, we can wrap them in another code reference. We use Plack::Builder for that:

use Plack::Builder;
use Plack::Middleware::BufferedStreaming;

my $webapp = sub { ... };
my $app = builder {
    enable 'BufferedStreaming',

The builder function (from Plack::Builder) creates a code references that receives any number of code references and sends the request it receives to each one in turn. This allows us to aggregate more and more PSGI applications. Our web application is a PSGI app, but so is a middleware. when we call enable, it will receive the middleware code reference and register it.

Thus we created a code reference that aggregates multiple code references, providing our software with a layer on top that will add buffering. We can stack more and more middlewares to provide even more features before it even reaches our actual web application code.

The future seems bright

Now that we have PSGI and Plack, we can finally program in any web framework we want without worrying about painting ourselves into a corner, limiting ourselves with regards to the web server, the technology, and the platform. No more! The future is open, and web programming is so much better.

See Also

Gravatar Image This article contributed by: Sawyer X <xsawyerx@cpan.org>