Kelp::Manual - Reference to web development with Kelp


First ...

    # lib/
    package MyApp;
    use parent 'Kelp';

    sub build {
        my $self = shift;
        my $r = $self->routes;
        $r->add( "/hello", sub { "Hello, world!" } );
        $r->add( '/hello/:name', 'greet' );

    sub greet {
        my ( $self, $name ) = @_;
        "Hello, $name!";


Then ...

    # app.psgi
    use lib 'lib';
    use MyApp;
    my $app = MyApp->new;

Finally ...

    > plackup app.psgi

Or, for quick prototyping use Kelp::Less:

    # app.psgi
    use Kelp::Less;

    get '/hello/?name' => sub {
        my ( $self, $name ) = @_;
        "Hello " . $name // 'world';



If you're going to be deploying a Perl based web application, chances are that you will be using Plack. Plack has almost all necessary tools to create and maintain a healthy web app. Tons of middleware is written for it, and there are several very well tested high performance preforking servers, such as Gazelle.

Plack, however, is not a web framework, hence its creators have intentionally omitted adding certain components. This is where Kelp gets to shine. It provides a layer on top of Plack and puts everything together into a complete web framework.

Kelp provides:

  • Advanced Routing. Create intricate, yet simple ways to capture HTTP requests and route them to their designated code. Use explicit and optional named placeholders, wildcards, or just regular expressions.

  • Flexible Configuration. Use different configuration file for each environment, e.g. development, deployment, etc. Merge a temporary configuration into your current one for testing and debugging purposes.

  • Enhanced Logging. Log messages at different levels of emergency. Log to a file, screen, or anything supported by Log::Dispatch.

  • Powerful Rendering. Use the built-in auto-rendering logic, or the template module of your choice to return rich text, html and JSON responses.

  • JSON encoder/decoder. Kelp comes with JSON, which will automatically choose the best (fastest, most secure) backend available (using JSON::MaybeXS).

  • Extendable Core. Kelp uses pluggable modules for everything. This allows anyone to add a module for a custom interface. Writing Kelp modules is easy.

  • Sleek Testing. Kelp takes Plack::Test and wraps it in an object oriented class of convenience methods. Testing is done via sending requests to your routes, then analyzing the response.


What makes Kelp different from the other Perl micro web frameworks? There are a number of fine web frameworks on CPAN, and most of them provide a complete platform for web app building. Most of them, however, bring their deployment code, and aim to write their own processing mechanisms. Kelp, on the other hand, is heavily Plack-centric. It uses Plack as its foundation layer, and it builds the web framework on top of it. Kelp::Request is an extension of Plack::Request, Kelp::Response is an extension of Plack::Response.

This approach of extending current CPAN code puts familiar and well tested tools in the hands of the application developer, while keeping familiar syntax and work flow.

Kelp is a team player and it uses several popular, trusted CPAN modules for its internals. At the same time it doesn't include modules that it doesn't need, just because they are considered trendy. It does its best to keep a lean profile and a small footprint, and it's completely object manager agnostic.


Using the Kelp script

The easiest way to create the directory structure and a general application skeleton is by using the Kelp script, which comes with this package.

    > Kelp MyApp

This will create lib/, app.psgi and some other files (explained below).

To create a Kelp::Less app, use:

    > Kelp --type=less MyApp

Get help by typing:

    > Kelp --help

Directory structure

Before you begin writing the internals of your app, you need to create the directory structure either by hand, or by using the above described Kelp utility script.

     |   |
     |   |--/MyApp
     |   |
     |   |
     |   |
     |   |

The lib folder contains your application modules and any local modules that you want your app to use.


The conf folder is where Kelp will look for configuration files. You need one main file, named You can also add other files that define different running environments, if you name them Replace environment with the actual name of the environment. To change the running environment, you can specify the app mode, or you can set the PLACK_ENV environment variable.

    my $app = MyApp->new( mode => 'development' );


    > PLACK_ENV=development plackup app.psgi

This is where the Template module will look for template files.


This is where the Logger module will create error.log, debug.log and any other log files that were defined in the configuration.


The t folder is traditionally used to hold test files. It is up to you to use it or not, although we strongly recommend that you write some automated test units for your web app.


This is the PSGI file, of the app, which you will deploy. In it's most basic form it should look like this:

    use lib './lib';
    use MyApp;

    my $app = MyApp->new;

The application classes

Your application's classes should be put in the lib/ folder. The main class, in our example, initializes any modules and variables that your app will use. Here is an example that uses Moose to create lazy attributes and initialize a database connection:

    package MyApp;

    use parent Kelp;
    use Moose;

    has dbh => (
        is      => 'ro',
        isa     => 'DBI',
        lazy    => 1,
        default => sub {
            my $self   = shift;
            my @config = @{ $self->config('dbi') };
            return DBI->connect(@config);

    sub build {
        my $self = shift;
        $self->routes->add("/read/:id", "read");

    sub read {
        my ( $self, $id ) = @_;
            SELECT * FROM problems
            WHERE id = ?
        ], $id);


What is happening here?

  • First, we create a lazy attribute and instruct it to connect to DBI. Notice that we have access to the current app and all of its internals via the $self variable. Notice also that the reason we define dbh as a lazy attribute is that config will not yet be initialized. All modules are initialized upon the creation of the object instance, e.g. when we call MyApp->new;

  • Then, we override Kelp's "build" in Kelp subroutine to create a single route /read/:id, which is assigned to the subroutine read in the current class.

  • The read subroutine, takes $self and $id (the named placeholder from the path), and uses $self->dbh to retrieve data.

A note about object managers: The above example uses Moose. It is entirely up to you to use Moose, another object manager, or no object manager at all. The above example will be just as successful if you used our own little Kelp::Base:

    package MyApp;
    use Kelp::Base 'Kelp';

    attr dbi => sub {



Kelp uses a powerful and very flexible router. Traditionally, it is also light and consists of less than 300 lines of code (comments included). You are encouraged to read Kelp::Routes, but here are some key points. All examples are assumed to be inside the "build" in Kelp method and $r is equal to $self->routes:


You can direct HTTP paths to subroutines in your classes or, you can use inline code.

    $r->add( "/home", "home" );  # goes to sub home
    $r->add( "/legal", "Legal::view" ); # goes to MyApp::Legal::view
    $r->add( "/about", sub { "Content for about" }); # inline

Restrict HTTP methods

Make a route only catch a specific HTTP method:

    $r->add( [ POST => '/update' ], "update_user" );

Named captures

Using regular expressions is so Perl. Sometimes, however, it gets a little overwhelming. Use named paths if you anticipate that you or someone else will ever want to maintain your code.


    $r->add( "/update/:id", "update" );

    # Later
    sub update {
        my ( $self, $id ) = @_;
        # Do something with $id


    $r->add( "/person/?name", sub {
        my ( $self, $name ) = @_;
        return "I am " . $name // "nobody";

This will handle /person, /person/ and /person/jack.


    $r->add( '/*article/:id', 'Articles::view' );

This will handle /bar/foo/baz/500 and send it to MyApp::Articles::view with parameters $article equal to bar/foo/baz and $id equal to 500.

Placeholder restrictions

Paths' named placeholders can be restricted by providing regular expressions.

    $r->add( '/user/:id', {
        check => { id => '\d+' },
        to    => "Users::get"

    # Matches /user/1000, but not /user/abc

Placeholder defaults

This only applies to optional placeholders, or those prefixed with a question mark. If a default value is provided for any of them, it will be used in case the placeholder value is missing.

    $r->add( '/:id/?other', defaults => { other => 'info' } );

    # GET /100;
    # { id => 100, other => 'info' }

    # GET /100/delete;
    # { id => 100, other => 'delete' }


A bridge is a route that has to return a true value in order for the next route in line to be processed.

    $r->add( '/users', { to => 'Users::auth', bridge => 1 } );
    $r->add( '/users/:action' => 'Users::dispatch' );

See "BRIDGES" in Kelp::Routes for more information.

URL building

Each path can be given a name and later a URL can be built using that name and the necessary arguments.

    $r->add( "/update/:id", { name => 'update', to => 'User::update' } );

    # Later

    my $url = $self->route->url('update', id => 1000); # /update/1000

Reblessing the app into a controller class

All of the examples here show routes which take an instance of the web application as a first parameter. This is true even if those routes live in another class. To rebless the app instance into the controller class instance, use the custom router class Kelp::Routes::Controller.

Step 1: Specify the custom router class in the config

        modules_init => {
            Routes => {
                router => 'Controller'

Step 2: Create a main controller class

This class must inherit from Kelp.

    # lib/MyApp/
    package MyApp::Controller;
    use Kelp::Base 'MyApp';

    # Now $self is an instance of 'MyApp::Controller';
    sub service_method {
        my $self = shift;


Step 3: Create any number of controller classes

They all must inherit from your main controller class.

    # lib/MyApp/Controller/
    package MyApp::Controller::Users;
    use Kelp::Base 'MyApp::Controller';

    # Now $self is an instance of 'MyApp::Controller::Users'
    sub authenticate {
        my $self = shift;


Quick development using Kelp::Less

For writing quick experimental web apps and to reduce the boiler plate, one could use Kelp::Less. In this case all of the code can be put in app.psgi: Look up the POD for Kelp::Less for many examples, but to get you started off, here is a quick one:

    # app.psgi
    use Kelp::Less;

    module 'JSON';

    get '/api/:user/?action' => sub {
        my ( $self, $user, $action ) = @_;
        my $json = {
            success => \1,
            user    => $user,
            action  => $action // 'ask'
        return $json;


Adding middleware

Kelp, being Plack-centric, will let you easily add middleware. There are three possible ways to add middleware to your application, and all three ways can be used separately or together.

Using the configuration

Adding middleware in your configuration is probably the easiest and best way for you. This way you can load different middleware for each running mode, e.g. Debug in development only.

Add middleware names to the middleware array in your configuration file and the corresponding initializing arguments in the middleware_init hash:

    # conf/
        middleware      => [qw/Session Debug/],
        middleware_init => {
            Session => { store => 'File' }

The middleware will be added in the order you specify in the middleware array.

In app.psgi:

    # app.psgi
    use MyApp;
    use Plack::Builder;

    my $app = MyApp->new();

    builder {
        enable "Plack::Middleware::ContentLength";

By overriding the "run" in Kelp subroutine in lib/

Make sure you call SUPER first, and then wrap new middleware around the returned app.

    # lib/
    sub run {
        my $self = shift;
        my $app = $self->SUPER::run(@_);
        $app = Plack::Middleware::ContentLength->wrap($app);
        return $app;

Note that any middleware defined in your config file will be added first.


Kelp provides a test class called Kelp::Test. It is object oriented, and all methods return the Kelp::Test object, so they can be chained together. Testing is done by sending HTTP requests to an already built application and analyzing the response. Therefore, each test usually begins with the "request" in Kelp::Test method, which takes a single HTTP::Request parameter. It sends the request to the web app and saves the response as an HTTP::Response object.

    # file t/test.t
    use MyApp;
    use Kelp::Test;
    use Test::More;
    use HTTP::Request::Common;

    my $app = MyApp->new( mode => 'test' );
    my $t = Kelp::Test->new( app => $app );

    $t->request( GET '/path' )
      ->content_is("It works");

    $t->request( POST '/api' )
      ->json_cmp({auth => 1});


What is happening here?

  • First, we create an instance of the web application class, which we have previously built and placed in the lib/ folder. We set the mode of the app to test, so that file conf/ overrides the main configuration. The test configuration can contain anything you see fit. Perhaps you want to disable certain modules, or maybe you want to make DBI connect to a different database.

  • Second, we create an instance of the Kelp::Test class and tell it that it will perform all tests using our $app instance.

  • At this point we are ready to send requests to the app via the request method. It takes only one argument, an HTTP::Request object. It is very convenient to use the HTTP::Request::Common module here, because you can create common requests using abridged syntax, i.e. GET, POST, etc. The line $t->request( GET '/path' ) first creates a HTTP::Request GET object, and then passes it to the request method.

  • After we send the request, we can test the response using any of the Test:: modules, or via the methods provided by Kelp::Test. In the above example, we test if we got a code 200 back from /path and if the returned content was It works.

Run the rest as usual, using prove:

    > prove -l t/test.t

Take a look at the Kelp::Test for details and more examples.

Building an HTTP response

Kelp contains an elegant module, called Kelp::Response, which extends Plack::Response with several useful methods. Most methods return $self after they do the required job. For the sake of the examples below, let's assume that all of the code is located inside a route definition.

Automatic content type

Your routes don't always have to set the response object. You could just return a simple scalar value or a reference to a hash, array or anything that can be converted to JSON.

    # Content-type automatically set to "text/html"
    sub text_route {
        return "There, there ...";

    # Content-type automatically set to "application/json"
    sub json_route {
        return { error => 1,  message => "Fail" };

Rendering text

    # Render simple text
    $self->res->text->render("It works!");

Rendering HTML

    $self->res->html->render("<h1>It works!</h1>");

Custom content type


Return 404 or 500 errors

    sub some_route {
        my $self = shift;
        if ($missing) {
            return $self->res->render_404;
        if ($broken) {
            return $self->res->render_500;


    sub hello {
        my ( $self, $name ) = @_;
        $self->res->template( '', { name => $name } );

The above example will render the contents of, and it will set the content-type to text/html. To set a different content-type, use set_content_type or any of its aliases:

    sub hello_txt {
        my ( $self, $name ) = @_;
        $self->res->text->template( '', { name => $name } );


    $self->set_header( "X-Framework", "Kelp" )->render( { success => \1 } );

Serving static files

If you want to serve static pages, you can use the Plack::Middleware::Static middleware that comes with Plack. Here is an example configuration that serves files in your public folder (under the Kelp root folder) from URLs that begin with /public:

    # conf/
        middleware      => [qw/Static/],
        middleware_init => {
            Static => {
                path => qr{^/public/},
                root => '.',

Uploading files

File uploads are handled by Kelp::Request, which inherits Plack::Request and has its uploads|Plack::Request/uploads property. The uploads property returns a reference to a hash containing all uploads.

    sub upload {
        my $self = shift;
        my $uploads  = $self->req->uploads;

        # Now $uploads is a hashref to all uploads

For Kelp::Less, then you can use the req reserved word:

    get '/upload' => sub {
        my $uploads = req->uploads;

Delayed responses

To send a delayed response, have your route return a subroutine.

    sub delayed {
        my $self = shift;
        return sub {
            my $responder = shift;
            $self->res->text->body("Better late than never.");

See the PSGI pod for more information and examples.

Pluggable modules

Kelp can be extended using custom modules. Each new module must be a subclass of the Kelp::Module namespace. Modules' job is to initialize and register new methods into the web application class. The following is the full code of the Kelp::Module::JSON for example:

    package Kelp::Module::JSON;

    use Kelp::Base 'Kelp::Module';
    use JSON;

    sub build {
        my ( $self, %args ) = @_;
        my $json = JSON::MaybeXS->new(
            map { $_ => $args{$_} } keys %args
        $self->register( json => $json );


What is happening here?

  • First we create a class Kelp::Module::JSON which inherits Kelp::Module.

  • Then, we override the build method (of Kelp::Module), create a new JSON object and register it into the web application via the register method.

If we instruct our web application to load the JSON::MaybeXS module, it will have a new method json which will be a link to the JSON::MaybeXS object initialized in the module.

See more examples and POD at Kelp::Module.

How to load modules using the config

There are two modules that are always loaded by each application instance. Those are Config and Routes. The reason behind this is that each and every application always needs a router and configuration. All other modules must be loaded either using the "load_module" in Kelp method, or using the modules key in the configuration. The default configuration already loads these modules: Template, Logger and JSON. Your configuration can remove some and/or add others. The configuration key modules_init may contain hashes with initialization arguments. See Kelp::Module for configuration examples.


To learn more, see the documentation for the following modules:







Stefan Geneshky - minimal <at>


In no particular order:

Julio Fraire

Maurice Aubrey

David Steinbrunner

Gurunandan Bhat


Ruslan Zakirov

Christian Froemmel (senfomat)

Ivan Baidakou (basiliscos)


Konstantin Yakunin (@yakunink)

Benjamin Hengst (notbenh)

Nikolay Mishin (@mishin)

Bartosz Jarzyna (@brtastic)


This module and all the modules in this package are governed by the same license as Perl itself.