=encoding utf8

=head1 NAME

Mojolicious::Guides::Cookbook - Cooking with Mojolicious

=head1 OVERVIEW

This document contains many fun recipes for cooking with L<Mojolicious>.

=head1 CONCEPTS

Essentials every L<Mojolicious> developer should know.

=head2 Blocking and non-blocking operations

A I<blocking> operation is a subroutine that blocks the execution of the calling subroutine until the subroutine is
finished.

  sub foo {
    my $result = blocking_subroutine();
    ...
  }

A I<non-blocking> operation on the other hand lets the calling subroutine continue execution even though the subroutine
is not yet finished. Instead of waiting, the calling subroutine passes along a callback to be executed once the
subroutine is finished, this is called continuation-passing style.

  sub foo {
    non_blocking_subroutine(sub ($result) {
      ...
    });
    ...
  }

While L<Mojolicious> has been designed from the ground up for non-blocking I/O and event loops, it is not possible to
magically make Perl code non-blocking. You have to use specialized non-blocking code available through modules like
L<Mojo::IOLoop> and L<Mojo::UserAgent>, or third-party event loops. You can wrap your blocking code in
L<subprocesses|Mojo::IOLoop/"subprocess"> though to prevent it from interfering with your non-blocking code.

=head2 Event loops

An event loop is basically a loop that continually tests for external events and executes the appropriate callbacks to
handle them, it is often the main loop in a program. Non-blocking tests for readability/writability of file descriptors
and timers are commonly used events for highly scalable network servers, because they allow a single process to handle
thousands of client connections concurrently.

  while (1) {
    my @readable = test_fds_for_readability();
    handle_readable_fds(@readable);

    my @writable = test_fds_for_writability();
    handle_writable_fds(@writable);

    my @expired = test_timers();
    handle_timers(@expired);
  }

In L<Mojolicious> this event loop is L<Mojo::IOLoop>.

=head2 Reverse proxy

A reverse proxy architecture is a deployment technique used in many production environments, where a I<reverse proxy>
server is put in front of your application to act as the endpoint accessible by external clients. It can provide a lot
of benefits, like terminating SSL connections from the outside, limiting the number of concurrent open sockets towards
the Mojolicious application (or even using Unix sockets), balancing load across multiple instances, or supporting
several applications through the same IP/port.

                   ..........................................
                   :                                        :
   +--------+      :  +-----------+      +---------------+  :
   |        |-------->|           |      |               |  :
   | client |      :  |  reverse  |----->|  Mojolicious  |  :
   |        |<--------|   proxy   |      |  application  |  :
   +--------+      :  |           |<-----|               |  :
                   :  +-----------+      +---------------+  :
                   :                                        :
                   .. system boundary (e.g. same host) ......

This setup introduces some problems, though: the application will receive requests from the reverse proxy instead of
the original client; the address/hostname where your application lives internally will be different from the one
visible from the outside; and if terminating SSL, the reverse proxy exposes services via HTTPS while using HTTP towards
the Mojolicious application.

As an example, compare a sample request from the client and what the Mojolicious application receives:

   client                       reverse proxy                Mojolicious app
    __|__              _______________|______________             ____|____
   /     \            /                              \           /         \
   1.2.3.4 --HTTPS--> api.example.com      10.20.30.39 --HTTP--> 10.20.30.40

   GET /foo/1 HTTP/1.1                |    GET /foo/1 HTTP/1.1
   Host: api.example.com              |    Host: 10.20.30.40
   User-Agent: Firefox                |    User-Agent: ShinyProxy/1.2
   ...                                |    ...

However, now the client address is no longer available (which might be useful for analytics, or Geo-IP) and URLs
generated via L<Mojolicious::Controller/"url_for"> will look like this:

   http://10.20.30.40/bar/2

instead of something meaningful for the client, like this:

   https://api.example.com/bar/2

To solve these problems, you can configure your reverse proxy to send the missing data (see L</Nginx> and
L</"Apache/mod_proxy">) and tell your application about it by setting the environment variable C<MOJO_REVERSE_PROXY>.
In more complex situations, usually involving multiple proxies or proxies that live outside your network, it can be
necessary to tell the application from which ip addresses to expect proxy requests by setting C<MOJO_TRUSTED_PROXIES>
to a list of comma separated addresses or CIDR networks. For even finer control, L</Rewriting> includes examples of how
the changes could be implemented manually.

=head1 DEPLOYMENT

Getting L<Mojolicious> and L<Mojolicious::Lite> applications running on different platforms. Note that many real-time
web features are based on the L<Mojo::IOLoop> event loop, and therefore require one of the built-in web servers to be
able to use them to their full potential.

=head2 Built-in web server

L<Mojolicious> contains a very portable non-blocking I/O HTTP and WebSocket server with L<Mojo::Server::Daemon>. It is
usually used during development and in the construction of more advanced web servers, but is solid and fast enough for
small to mid sized applications.

  $ ./script/my_app daemon
  Web application available at http://127.0.0.1:3000

It is available to every application through the command L<Mojolicious::Command::daemon>, which has many configuration
options and is known to work on every platform Perl works on with its single-process architecture.

  $ ./script/my_app daemon -h
  ...List of available options...

Another huge advantage is that it supports TLS and WebSockets out of the box, a development certificate for testing
purposes is built right in, so it just works, but you can specify all listen locations supported by
L<Mojo::Server::Daemon/"listen">.

  $ ./script/my_app daemon -l https://[::]:3000
  Web application available at https://[::]:3000

To manage the web server with systemd, you can use a unit configuration file like this.

  [Unit]
  Description=My Mojolicious application
  After=network.target

  [Service]
  Type=simple
  User=sri
  ExecStart=/home/sri/myapp/script/my_app daemon -m production -l http://*:8080

  [Install]
  WantedBy=multi-user.target

=head2 Pre-forking

For bigger applications L<Mojolicious> contains the UNIX optimized pre-forking web server L<Mojo::Server::Prefork>,
which can take advantage of multiple CPU cores and copy-on-write memory management to scale up to thousands of
concurrent client connections.

  Mojo::Server::Prefork
  |- Mojo::Server::Daemon [1]
  |- Mojo::Server::Daemon [2]
  |- Mojo::Server::Daemon [3]
  +- Mojo::Server::Daemon [4]

It is based on L<Mojo::Server::Daemon> and available to every application through the command
L<Mojolicious::Command::prefork>.

  $ ./script/my_app prefork
  Web application available at http://127.0.0.1:3000

Since all built-in web servers are based on the L<Mojo::IOLoop> event loop, they scale best with non-blocking
operations. But if your application for some reason needs to perform many blocking operations, you can improve
performance by increasing the number of worker processes and decreasing the number of concurrent connections each
worker is allowed to handle (often as low as C<1>).

  $ ./script/my_app prefork -m production -w 10 -c 1
  Web application available at http://127.0.0.1:3000

During startup your application is preloaded in the manager process, which does not run an event loop, so you can use
L<Mojo::IOLoop/"next_tick"> to run code whenever a new worker process has been forked and its event loop gets started.

  use Mojolicious::Lite;

  Mojo::IOLoop->next_tick(sub ($ioloop) {
    app->log->info("Worker $$ star...ALL GLORY TO THE HYPNOTOAD!");
  });

  get '/' => {text => 'Hello Wor...ALL GLORY TO THE HYPNOTOAD!'};

  app->start;

And to manage the pre-forking web server with systemd, you can use a unit configuration file like this.

  [Unit]
  Description=My Mojolicious application
  After=network.target

  [Service]
  Type=simple
  User=sri
  ExecStart=/home/sri/myapp/script/my_app prefork -m production -l http://*:8080

  [Install]
  WantedBy=multi-user.target

=head2 Morbo

After reading the L<Mojolicious::Guides::Tutorial>, you should already be familiar with L<Mojo::Server::Morbo>.

  Mojo::Server::Morbo
  +- Mojo::Server::Daemon

It is basically a restarter that forks a new L<Mojo::Server::Daemon> web server whenever a file in your project
changes, and should therefore only be used during development. To start applications with it you can use the L<morbo>
script.

  $ morbo ./script/my_app
  Web application available at http://127.0.0.1:3000

=head2 Containers

There are many ways to go cloud-native with L<Mojolicious>. To get you started with containerizing your web applications
we will explore one of them in this recipe. First, you will need to declare the CPAN dependencies of your application,
for example in a C<Makefile.PL> file. This should always include at the very least L<Mojolicious> itself.

  use strict;
  use warnings;

  use ExtUtils::MakeMaker;

  WriteMakefile(
    VERSION   => '0.01',
    PREREQ_PM => {
      'Mojolicious' => '8.65',
      'Mojolicious::Plugin::Status' => '1.12'
    },
    test => {TESTS => 't/*.t'}
  );

The helper command L<Mojolicious::Command::Author::generate::makefile> can also generate a minimal C<Makefile.PL> for
you.

  $ ./myapp.pl generate makefile
  ...

And then we are going to need a C<Dockerfile> describing the container. A very simple one will do for now.

  FROM perl
  WORKDIR /opt/myapp
  COPY . .
  RUN cpanm --installdeps -n .
  EXPOSE 3000
  CMD ./myapp.pl prefork

It uses the latest L<Perl container|https://hub.docker.com/_/perl> from Docker Hub, copies all the contents of your
application directory into the container, installs CPAN dependencies with L<App::cpanminus>, and then starts the
application on port C<3000> with the pre-forking web server. With L<Mojolicious::Command::Author::generate::dockerfile>
there is also a helper command to generate a minimal C<Dockerfile> for you.

  $ ./myapp.pl generate dockerfile
  ...

To build and deploy our container there are also many options available, here we will simply use Docker.

  $ docker build -t myapp_image .
  ...
  $ docker run -d -p 3000:3000 --name myapp_container myapp_image
  ...

And now your web application should be deployed as a container under C<http://127.0.0.1:3000>. For more information and
many more container deployment options we recommend the L<Docker|https://docs.docker.com/> and
L<Kubernetes|https://kubernetes.io/docs/> documentation.

=head2 Hypnotoad

L<Hypnotoad|Mojo::Server::Hypnotoad> is based on the L<Mojo::Server::Prefork> web server, and adds some features
especially optimized for high availability non-containerized production environments. To start applications with it you
can use the L<hypnotoad> script, which listens on port C<8080>, automatically daemonizes the server process and defaults
to C<production> mode for L<Mojolicious> and L<Mojolicious::Lite> applications.

  $ hypnotoad ./script/my_app

Many configuration settings can be tweaked right from within your application with L<Mojolicious/"config">, for a full
list see L<Mojo::Server::Hypnotoad/"SETTINGS">.

  use Mojolicious::Lite;

  app->config(hypnotoad => {listen => ['http://*:80']});

  get '/' => {text => 'Hello Wor...ALL GLORY TO THE HYPNOTOAD!'};

  app->start;

Or just add a C<hypnotoad> section to your L<Mojolicious::Plugin::Config>, L<Mojolicious::Plugin::JSONConfig> or
L<Mojolicious::Plugin::NotYAMLConfig> configuration file.

  # myapp.conf
  {
    hypnotoad => {
      listen  => ['https://*:443?cert=/etc/server.crt&key=/etc/server.key'],
      workers => 10
    }
  };

But one of its biggest advantages is the support for effortless zero downtime software upgrades (hot deployment). That
means you can upgrade L<Mojolicious>, Perl or even system libraries at runtime without ever stopping the server or
losing a single incoming connection, just by running the command above again.

  $ hypnotoad ./script/my_app
  Starting hot deployment for Hypnotoad server 31841.

You might also want to enable proxy support if you're using L<Hypnotoad|Mojo::Server::Hypnotoad> behind a reverse
proxy. This allows L<Mojolicious> to automatically pick up the C<X-Forwarded-For> and C<X-Forwarded-Proto> headers.

  # myapp.conf
  {hypnotoad => {proxy => 1}};

To manage L<Hypnotoad|Mojo::Server::Hypnotoad> with systemd, you can use a unit configuration file like this.

  [Unit]
  Description=My Mojolicious application
  After=network.target

  [Service]
  Type=forking
  User=sri
  PIDFile=/home/sri/myapp/script/hypnotoad.pid
  ExecStart=/path/to/hypnotoad /home/sri/myapp/script/my_app
  ExecReload=/path/to/hypnotoad /home/sri/myapp/script/my_app
  KillMode=process

  [Install]
  WantedBy=multi-user.target

=head2 Zero downtime software upgrades

L<Hypnotoad|Mojo::Server::Hypnotoad> makes zero downtime software upgrades (hot deployment) very simple, as you can see
above, but on modern operating systems that support the C<SO_REUSEPORT> socket option, there is also another method
available that works with all built-in web servers.

  $ ./script/my_app prefork -P /tmp/first.pid -l http://*:8080?reuse=1
  Web application available at http://127.0.0.1:8080

All you have to do, is to start a second web server listening to the same port, and stop the first web server
gracefully afterwards.

  $ ./script/my_app prefork -P /tmp/second.pid -l http://*:8080?reuse=1
  Web application available at http://127.0.0.1:8080
  $ kill -s TERM `cat /tmp/first.pid`

Just remember that both web servers need to be started with the C<reuse> parameter.

=head2 Nginx

One of the most popular setups these days is L<Hypnotoad|Mojo::Server::Hypnotoad> behind an L<Nginx|https://nginx.org>
reverse proxy, which even supports WebSockets in newer versions.

  upstream myapp {
    server 127.0.0.1:8080;
  }
  server {
    listen 80;
    server_name localhost;
    location / {
      proxy_pass http://myapp;
      proxy_http_version 1.1;
      proxy_set_header Upgrade $http_upgrade;
      proxy_set_header Connection "upgrade";
      proxy_set_header Host $host;
      proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
      proxy_set_header X-Forwarded-Proto $scheme;
    }
  }

=head2 Apache/mod_proxy

Another good reverse proxy is L<Apache|https://httpd.apache.org> with C<mod_proxy>, the configuration looks quite
similar to the Nginx one above. And if you need WebSocket support, newer versions come with C<mod_proxy_wstunnel>.

  <VirtualHost *:80>
    ServerName localhost
    <Proxy *>
      Require all granted
    </Proxy>
    ProxyRequests Off
    ProxyPreserveHost On
    ProxyPass /echo ws://localhost:8080/echo
    ProxyPass / http://localhost:8080/ keepalive=On
    ProxyPassReverse / http://localhost:8080/
    RequestHeader set X-Forwarded-Proto "http"
  </VirtualHost>

=head2 Apache/CGI

C<CGI> is supported out of the box and your L<Mojolicious> application will automatically detect that it is executed as
a C<CGI> script. Its use in production environments is discouraged though, because as a result of how C<CGI> works, it
is very slow and many web servers are making it exceptionally hard to configure properly. Additionally, many real-time
web features, such as WebSockets, are not available.

  ScriptAlias / /home/sri/my_app/script/my_app/

=head2 Envoy

L<Mojolicious> applications can be deployed on cloud-native environments that use L<Envoy|https://www.envoyproxy.io>,
such as with this reverse proxy configuration similar to the Apache and Nginx ones above.

  static_resources:
    listeners:
    - name: listener_0
      address:
        socket_address: { address: 0.0.0.0, port_value: 80 }
      filter_chains:
      - filters:
        - name: envoy.filters.network.http_connection_manager
          typed_config:
            "@type": type.googleapis.com/envoy.extensions.filters.network.http_connection_manager.v3.HttpConnectionManager
            codec_type: auto
            stat_prefix: index_http
            route_config:
              name: local_route
              virtual_hosts:
              - name: service
                domains: ["*"]
                routes:
                - match:
                    prefix: "/"
                  route:
                    cluster: local_service
            upgrade_configs:
            - upgrade_type: websocket
            http_filters:
            - name: envoy.filters.http.router
              typed_config:
    clusters:
    - name: local_service
      connect_timeout: 0.25s
      type: strict_dns
      lb_policy: round_robin
      load_assignment:
        cluster_name: local_service
        endpoints:
        - lb_endpoints:
          - endpoint:
              address:
                socket_address: { address: mojo, port_value: 8080 }

While this configuration works for simple applications, Envoy's typical use case is for implementing proxies of
applications as a "service mesh" providing advanced filtering, load balancing, and observability features, such as
seen in L<Istio|https://istio.io/latest/docs/ops/deployment/architecture/>. For more examples, visit the
L<Envoy documentation|https://www.envoyproxy.io/docs/envoy/latest/start/start>.

=head2 PSGI/Plack

L<PSGI> is an interface between Perl web frameworks and web servers, and L<Plack> is a Perl module and toolkit that
contains L<PSGI> middleware, helpers and adapters to web servers. L<PSGI> and L<Plack> are inspired by Python's WSGI
and Ruby's Rack. L<Mojolicious> applications are ridiculously simple to deploy with L<Plack>, but be aware that many
real-time web features, such as WebSockets, are not available.

  $ plackup ./script/my_app

L<Plack> provides many server and protocol adapters for you to choose from, such as C<FCGI>, C<uWSGI> and C<mod_perl>.

  $ plackup ./script/my_app -s FCGI -l /tmp/myapp.sock

The C<MOJO_REVERSE_PROXY> environment variable can be used to enable proxy support, this allows L<Mojolicious> to
automatically pick up the C<X-Forwarded-For> and C<X-Forwarded-Proto> headers.

  $ MOJO_REVERSE_PROXY=1 plackup ./script/my_app

If an older server adapter is unable to correctly detect the application home directory, you can simply use the
C<MOJO_HOME> environment variable.

  $ MOJO_HOME=/home/sri/my_app plackup ./script/my_app

There is no need for a C<.psgi> file, just point the server adapter at your application script, it will automatically
act like one if it detects the presence of a C<PLACK_ENV> environment variable.

=head2 Plack middleware

Wrapper scripts like C<myapp.fcgi> are a great way to separate deployment and application logic.

  #!/usr/bin/env plackup -s FCGI
  use Plack::Builder;

  builder {
    enable 'Deflater';
    require './script/my_app';
  };

L<Mojo::Server::PSGI> can be used directly to load and customize applications in the wrapper script.

  #!/usr/bin/env plackup -s FCGI
  use Mojo::Server::PSGI;
  use Plack::Builder;

  builder {
    enable 'Deflater';
    my $server = Mojo::Server::PSGI->new;
    $server->load_app('./script/my_app');
    $server->app->config(foo => 'bar');
    $server->to_psgi_app;
  };

But you could even use middleware right in your application.

  use Mojolicious::Lite -signatures;
  use Plack::Builder;

  get '/welcome' => sub ($c) {
    $c->render(text => 'Hello Mojo!');
  };

  builder {
    enable 'Deflater';
    app->start;
  };

=head2 Rewriting

Sometimes you might have to deploy your application in a blackbox environment where you can't just change the server
configuration or behind a reverse proxy that passes along additional information with C<X-Forwarded-*> headers. In such
cases you can use the hook L<Mojolicious/"before_dispatch"> to rewrite incoming requests.

  # Change scheme if "X-Forwarded-HTTPS" header is set
  $app->hook(before_dispatch => sub ($c) {
    $c->req->url->base->scheme('https')
      if $c->req->headers->header('X-Forwarded-HTTPS');
  });

Since reverse proxies generally don't pass along information about path prefixes your application might be deployed
under, rewriting the base path of incoming requests is also quite common. This allows
L<Mojolicious::Controller/"url_for"> for example, to generate portable URLs based on the current environment.

  # Move first part and slash from path to base path in production mode
  $app->hook(before_dispatch => sub ($c) {
    push @{$c->req->url->base->path->trailing_slash(1)},
      shift @{$c->req->url->path->leading_slash(0)};
  }) if $app->mode eq 'production';

L<Mojo::URL> objects are very easy to manipulate, just make sure that the URL (C<foo/bar?baz=yada>), which represents
the routing destination, is always relative to the base URL (C<http://example.com/myapp/>), which represents the
deployment location of your application.

=head2 Deployment specific plugins

Deployment specific 3rd party plugins such as L<Mojolicious::Plugin::SetUserGroup> do not need to be included in your
application code. They can also be loaded later on via the reserved C<plugins> value for L<Mojolicious> applications
that are using any one of the built-in configuration plugins L<Mojolicious::Plugin::Config>,
L<Mojolicious::Plugin::JSONConfig> or L<Mojolicious::Plugin::NotYAMLConfig>.

  # myapp.conf
  {
    plugins => [
      {SetUserGroup => {user => 'sri', group => 'staff'}}
    ]
  };

=head2 Application embedding

From time to time you might want to reuse parts of L<Mojolicious> applications like configuration files, database
connection or helpers for other scripts, with this little L<Mojo::Server> based mock server you can just embed them.

  use Mojo::Server;

  # Load application with mock server
  my $server = Mojo::Server->new;
  my $app = $server->load_app('./myapp.pl');

  # Access fully initialized application
  say for @{$app->static->paths};
  say $app->config->{secret_identity};
  say $app->dumper({just => 'a helper test'});
  say $app->build_controller->render_to_string(template => 'foo');

The plugin L<Mojolicious::Plugin::Mount> uses this functionality to allow you to combine multiple applications into one
and deploy them together.

  use Mojolicious::Lite;

  app->config(hypnotoad => {listen => ['http://*:80']});

  plugin Mount => {'test1.example.com' => '/home/sri/myapp1.pl'};
  plugin Mount => {'test2.example.com' => '/home/sri/myapp2.pl'};

  app->start;

=head2 Web server embedding

You can also use L<Mojo::IOLoop/"one_tick"> to embed the built-in web server L<Mojo::Server::Daemon> into alien
environments like foreign event loops that for some reason can't just be integrated with a new reactor backend.

  use Mojolicious::Lite;
  use Mojo::IOLoop;
  use Mojo::Server::Daemon;

  # Normal action
  get '/' => {text => 'Hello World!'};

  # Connect application with web server and start accepting connections
  my $daemon = Mojo::Server::Daemon->new(app => app, listen => ['http://*:8080']);
  $daemon->start;

  # Call "one_tick" repeatedly from the alien environment
  Mojo::IOLoop->one_tick while 1;

=head1 REAL-TIME WEB

The real-time web is a collection of technologies that include Comet (long polling), EventSource and WebSockets, which
allow content to be pushed to consumers with long-lived connections as soon as it is generated, instead of relying on
the more traditional pull model. All built-in web servers use non-blocking I/O and are based on the L<Mojo::IOLoop>
event loop, which provides many very powerful features that allow real-time web applications to scale up to thousands
of concurrent client connections.

=head2 Backend web services

Since L<Mojo::UserAgent> is also based on the L<Mojo::IOLoop> event loop, it won't block the built-in web servers when
used non-blocking, even for high latency backend web services.

  use Mojolicious::Lite -signatures;

  # Search MetaCPAN for "mojolicious"
  get '/' => sub ($c) {
    $c->ua->get('fastapi.metacpan.org/v1/module/_search?q=mojolicious' => sub ($ua, $tx) {
      $c->render('metacpan', hits => $tx->result->json->{hits}{hits});
    });
  };

  app->start;
  __DATA__

  @@ metacpan.html.ep
  <!DOCTYPE html>
  <html>
    <head><title>MetaCPAN results for "mojolicious"</title></head>
    <body>
      % for my $hit (@$hits) {
        <p><%= $hit->{_source}{release} %></p>
      % }
    </body>
  </html>

The callback passed to L<Mojo::UserAgent/"get"> will be executed once the request to the backend web service has been
finished, this is called continuation-passing style.

=head2 Synchronizing non-blocking operations

Multiple non-blocking operations, such as concurrent requests, can be easily synchronized with promises and
L<Mojo::Promise/"all">. You create L<Mojo::Promise> objects manually or use methods like L<Mojo::UserAgent/"get_p">
that create them for you.

  use Mojolicious::Lite -signatures;
  use Mojo::Promise;
  use Mojo::URL;

  # Search MetaCPAN for "mojo" and "minion"
  get '/' => sub ($c) {

    # Create two promises
    my $url   = Mojo::URL->new('http://fastapi.metacpan.org/v1/module/_search');
    my $mojo   = $c->ua->get_p($url->clone->query({q => 'mojo'}));
    my $minion = $c->ua->get_p($url->clone->query({q => 'minion'}));

    # Render a response once both promises have been resolved
    Mojo::Promise->all($mojo, $minion)->then(sub ($mojo, $minion) {
      $c->render(json => {
        mojo   => $mojo->[0]->result->json('/hits/hits/0/_source/release'),
        minion => $minion->[0]->result->json('/hits/hits/0/_source/release')
      });
    })->catch(sub ($err) {
      $c->reply->exception($err);
    })->wait;
  };

  app->start;

To create promises manually you just wrap your continuation-passing style APIs in functions that return promises.
Here's an example for how L<Mojo::UserAgent/"get_p"> works internally.

  use Mojo::UserAgent;
  use Mojo::Promise;

  # Wrap a user agent method with a promise
  my $ua = Mojo::UserAgent->new;
  sub get_p {
    my $promise = Mojo::Promise->new;
    $ua->get(@_ => sub ($ua, $tx) {
      my $err = $tx->error;
      $promise->resolve($tx) if !$err || $err->{code};
      $promise->reject($err->{message});
    });
    return $promise;
  }

  # Use our new promise generating function
  get_p('https://mojolicious.org')->then(sub ($tx) {
    say $tx->result->dom->at('title')->text;
  })->wait;

Promises have three states, they start out as C<pending> and you call L<Mojo::Promise/"resolve"> to transition them to
C<fulfilled>, or L<Mojo::Promise/"reject"> to transition them to C<rejected>.

=head2 async/await

And if you have L<Future::AsyncAwait> installed you can make using promises even easier. The C<async> and C<await>
keywords are enabled with the C<-async_await> flag of L<Mojo::Base>, and make the use of closures with promises
completely optional.

  use Mojo::Base -strict, -async_await;

The C<async> keyword is placed before the C<sub> keyword, and means that this function always returns a promise.
Returned values that are not L<Mojo::Promise> objects will be wrapped in a resolved promise automatically. And if an
exception gets thrown in the function it will result in a rejected promise.

  use Mojo::Base -strict, -async_await;

  async sub hello_p {
    return 'Hello Mojo!';
  }

  hello_p()->then(sub { say @_ })->wait;

The C<await> keyword on the other hand makes Perl wait for the promise to be settled. It then returns the fulfillment
values or throws an exception with the rejection reason. While waiting, the event loop is free to perform other tasks
however, so no resources are wasted.

  use Mojo::Base -strict, -signatures, -async_await;
  use Mojo::UserAgent;
  use Mojo::URL;

  my $ua = Mojo::UserAgent->new;

  # Search MetaCPAN non-blocking for multiple terms sequentially
  async sub search_cpan_p ($terms) {
    my $cpan = Mojo::URL->new('http://fastapi.metacpan.org/v1/module/_search');
    my @urls = map { $cpan->clone->query(q => $_) } @$terms;

    for my $url (@urls) {
      my $tx = await $ua->get_p($url);
      say $tx->result->json('/hits/hits/0/_source/release');
    }
  }

  search_cpan_p(['mojo', 'minion'])->wait;

The loop above performs all requests sequentially, awaiting a result before sending the next request. But you can also
perform those requests concurrently instead, by using methods like L<Mojo::Promise/"all"> to combine multiple promises
before awaiting the results.

  use Mojo::Base -strict, -signatures, -async_await;
  use Mojo::Promise;
  use Mojo::UserAgent;
  use Mojo::URL;

  my $ua = Mojo::UserAgent->new;

  # Search MetaCPAN non-blocking for multiple terms concurrently
  async sub search_cpan_p ($terms) {
    my $cpan = Mojo::URL->new('http://fastapi.metacpan.org/v1/module/_search');
    my @urls = map { $cpan->clone->query(q => $_) } @$terms;

    my @promises = map { $ua->get_p($_) } @urls;
    my @results  = await Mojo::Promise->all(@promises);
    for my $result (@results) {
      say $result->[0]->result->json('/hits/hits/0/_source/release');
    }
  }

  search_cpan_p(['mojo', 'minion'])->wait;

All of this also means that you can use normal Perl exception handling again. Even many 3rd party exception handling
modules from CPAN work just fine.

  use Mojo::Base -strict, -async_await;
  use Mojo::Promise;

  # Catch a non-blocking exception
  async sub hello_p {
    eval { await Mojo::Promise->reject('This is an exception') };
    if (my $err = $@) { warn "Error: $err" }
  }

  hello_p()->wait;

And it works just the same in L<Mojolicious> and L<Mojolicious::Lite> applications. Just declare your actions with the
C<async> keyword and use C<await> to wait for promises to be C<fulfilled> or C<rejected>.

  use Mojolicious::Lite -signatures, -async_await;

  # Request HTML titles from two sites non-blocking
  get '/' => async sub ($c) {
    my $mojo_tx    = await $c->ua->get_p('https://mojolicious.org');
    my $mojo_title = $mojo_tx->result->dom->at('title')->text;
    my $cpan_tx    = await $c->ua->get_p('https://metacpan.org');
    my $cpan_title = $cpan_tx->result->dom->at('title')->text;

    $c->render(json => {mojo => $mojo_title, cpan => $cpan_title});
  };

  app->start;

Promises returned by actions will automatically get the default L<Mojolicious> exception handler attached. Making it
much harder to ever miss a non-blocking exception again, even if you forgot to handle it yourself.

=head2 Timers

Timers, another primary feature of the event loop, are created with L<Mojo::IOLoop/"timer"> and can, for example, be
used to delay rendering of a response, and unlike C<sleep>, won't block any other requests that might be processed
concurrently.

  use Mojolicious::Lite -signatures;
  use Mojo::IOLoop;

  # Wait 3 seconds before rendering a response
  get '/' => sub ($c) {
    Mojo::IOLoop->timer(3 => sub ($ioloop) {
      $c->render(text => 'Delayed by 3 seconds!');
    });
  };

  app->start;

Recurring timers created with L<Mojo::IOLoop/"recurring"> are slightly more powerful, but need to be stopped manually,
or they would just keep getting emitted.

  use Mojolicious::Lite -signatures;
  use Mojo::IOLoop;

  # Count to 5 in 1 second steps
  get '/' => sub ($c) {

    # Start recurring timer
    my $i = 1;
    my $id = Mojo::IOLoop->recurring(1 => sub ($ioloop) {
      $c->write_chunk($i);
      $c->finish if $i++ == 5;
    });

    # Stop recurring timer
    $c->on(finish => sub ($c) { Mojo::IOLoop->remove($id) });
  };

  app->start;

Timers are not tied to a specific request or connection, and can even be created at startup time.

  use Mojolicious::Lite -signatures;
  use Mojo::IOLoop;

  # Check title in the background every 10 seconds
  my $title = 'Got no title yet.';
  Mojo::IOLoop->recurring(10 => sub ($ioloop) {
    app->ua->get('https://mojolicious.org' => sub ($ua, $tx) {
      $title = $tx->result->dom->at('title')->text;
    });
  });

  # Show current title
  get '/' => sub ($c) {
    $c->render(json => {title => $title});
  };

  app->start;

Just remember that all these non-blocking operations are processed cooperatively, so your callbacks shouldn't block for
too long.

=head2 Subprocesses

You can also use subprocesses, created with L<Mojo::IOLoop/"subprocess">, to perform computationally expensive
operations without blocking the event loop.

  use Mojolicious::Lite -signatures;
  use Mojo::IOLoop;

  # Operation that would block the event loop for 5 seconds
  get '/' => sub ($c) {
    Mojo::IOLoop->subprocess->run_p(sub {
      sleep 5;
      return '♥', 'Mojolicious';
    })->then(sub (@results) {
      $c->render(text => "I $results[0] $results[1]!");
    })->catch(sub ($err) {
      $c->reply->exception($err);
    });
  };

  app->start;

The callback passed to L<Mojo::IOLoop::Subprocess/"run_p"> will be executed in a child process, without blocking the
event loop of the parent process. The results of the callback will then be shared between both processes, and the
promise fulfilled or rejected in the parent process.

=head2 Exceptions in non-blocking operations

Since timers and other non-blocking operations are running solely in the event loop, outside of the application,
exceptions that get thrown in callbacks can't get caught and handled automatically. But you can handle them manually by
subscribing to the event L<Mojo::Reactor/"error"> or catching them inside the callback.

  use Mojolicious::Lite -signatures;
  use Mojo::IOLoop;

  # Forward error messages to the application log
  Mojo::IOLoop->singleton->reactor->on(error => sub ($reactor, $err) {
    app->log->error($err);
  });

  # Exception only gets logged (and connection times out)
  get '/connection_times_out' => sub ($c) {
    Mojo::IOLoop->timer(2 => sub ($ioloop) {
      die 'This request will not be getting a response';
    });
  };

  # Exception gets caught and handled
  get '/catch_exception' => sub ($c) {
    Mojo::IOLoop->timer(2 => sub ($ioloop) {
      eval { die 'This request will be getting a response' };
      $c->reply->exception($@) if $@;
    });
  };

  app->start;

A default subscriber that turns all errors into warnings will usually be added by L<Mojo::IOLoop> as a fallback.

  Mojo::IOLoop->singleton->reactor->unsubscribe('error');

During development or for applications where crashing is simply preferable, you can also make every exception that gets
thrown in a callback fatal by removing all of its subscribers.

=head2 WebSocket web service

The WebSocket protocol offers full bi-directional low-latency communication channels between clients and servers.
Receive messages just by subscribing to events such as L<Mojo::Transaction::WebSocket/"message"> with
L<Mojolicious::Controller/"on"> and return them with L<Mojolicious::Controller/"send">.

  use Mojolicious::Lite -signatures;

  # Template with browser-side code
  get '/' => 'index';

  # WebSocket echo service
  websocket '/echo' => sub ($c) {

    # Opened
    $c->app->log->debug('WebSocket opened');

    # Increase inactivity timeout for connection a bit
    $c->inactivity_timeout(300);

    # Incoming message
    $c->on(message => sub ($c, $msg) {
      $c->send("echo: $msg");
    });

    # Closed
    $c->on(finish => sub ($c, $code, $reason = undef) {
      $c->app->log->debug("WebSocket closed with status $code");
    });
  };

  app->start;
  __DATA__

  @@ index.html.ep
  <!DOCTYPE html>
  <html>
    <head><title>Echo</title></head>
    <body>
      <script>
        const ws = new WebSocket('<%= url_for('echo')->to_abs %>');

        // Incoming messages
        ws.onmessage = function (event) {
          document.body.innerHTML += event.data + '<br/>';
        };

        // Outgoing messages
        ws.onopen = function (event) {
          window.setInterval(function () { ws.send('Hello Mojo!') }, 1000);
        };
      </script>
    </body>
  </html>

The event L<Mojo::Transaction::WebSocket/"finish"> will be emitted right after the WebSocket connection has been
closed.

  $c->tx->with_compression;

You can activate C<permessage-deflate> compression with L<Mojo::Transaction::WebSocket/"with_compression">, this can
result in much better performance, but also increases memory usage by up to 300KiB per connection.

  my $proto = $c->tx->with_protocols('v2.proto', 'v1.proto');

You can also use L<Mojo::Transaction::WebSocket/"with_protocols"> to negotiate a subprotocol.

=head2 EventSource web service

EventSource is a special form of long polling where you can use L<Mojolicious::Controller/"write"> to directly send DOM
events from servers to clients. It is uni-directional, that means you will have to use Ajax requests for sending data
from clients to servers, the advantage however is low infrastructure requirements, since it reuses the HTTP protocol
for transport.

  use Mojolicious::Lite -signatures;

  # Template with browser-side code
  get '/' => 'index';

  # EventSource for log messages
  get '/events' => sub ($c) {

    # Increase inactivity timeout for connection a bit
    $c->inactivity_timeout(300);

    # Change content type and finalize response headers
    $c->res->headers->content_type('text/event-stream');
    $c->write;

    # Subscribe to "message" event and forward "log" events to browser
    my $cb = $c->app->log->on(message => sub ($log, $level, @lines) {
      $c->write("event:log\ndata: [$level] @lines\n\n");
    });

    # Unsubscribe from "message" event again once we are done
    $c->on(finish => sub ($c) {
      $c->app->log->unsubscribe(message => $cb);
    });
  };

  app->start;
  __DATA__

  @@ index.html.ep
  <!DOCTYPE html>
  <html>
    <head><title>LiveLog</title></head>
    <body>
      <script>
        const events = new EventSource('<%= url_for 'events' %>');

        // Subscribe to "log" event
        events.addEventListener('log', function (event) {
          document.body.innerHTML += event.data + '<br/>';
        }, false);
      </script>
    </body>
  </html>

The event L<Mojo::Log/"message"> will be emitted for every new log message and the event L<Mojo::Transaction/"finish">
right after the transaction has been finished.

=head2 Streaming multipart uploads

L<Mojolicious> contains a very sophisticated event system based on L<Mojo::EventEmitter>, with ready-to-use events on
almost all layers, and which can be combined to solve some of the hardest problems in web development.

  use Mojolicious::Lite -signatures;
  use Scalar::Util qw(weaken);

  # Intercept multipart uploads and log each chunk received
  hook after_build_tx => sub ($tx, $app) {

    # Subscribe to "upgrade" event to identify multipart uploads
    weaken $tx;
    $tx->req->content->on(upgrade => sub ($single, $multi) {
      return unless $tx->req->url->path->contains('/upload');

      # Subscribe to "part" event to find the right one
      $multi->on(part => sub ($multi, $single) {

        # Subscribe to "body" event of part to make sure we have all headers
        $single->on(body => sub ($single) {

          # Make sure we have the right part and replace "read" event
          return unless $single->headers->content_disposition =~ /example/;
          $single->unsubscribe('read')->on(read => sub ($single, $bytes) {

            # Log size of every chunk we receive
            $app->log->debug(length($bytes) . ' bytes uploaded');
          });
        });
      });
    });
  };

  # Upload form in DATA section
  get '/' => 'index';

  # Streaming multipart upload
  post '/upload' => {text => 'Upload was successful.'};

  app->start;
  __DATA__

  @@ index.html.ep
  <!DOCTYPE html>
  <html>
    <head><title>Streaming multipart upload</title></head>
    <body>
      %= form_for upload => (enctype => 'multipart/form-data') => begin
        %= file_field 'example'
        %= submit_button 'Upload'
      % end
    </body>
  </html>

=head2 More event loops

Internally, the L<Mojo::IOLoop> event loop can use multiple reactor backends, L<EV> for example, will be automatically
used if possible. Which in turn allows other event loops like L<AnyEvent> to just work.

  use Mojolicious::Lite -signatures;
  use EV;
  use AnyEvent;

  # Wait 3 seconds before rendering a response
  get '/' => sub ($c) {
    my $w;
    $w = AE::timer 3, 0, sub {
      $c->render(text => 'Delayed by 3 seconds!');
      undef $w;
    };
  };

  app->start;

=head1 USER AGENT

When we say L<Mojolicious> is a web framework we actually mean it, with L<Mojo::UserAgent> there's a full featured HTTP
and WebSocket user agent built right in.

=head2 REST web services

Requests can be performed very comfortably with methods like L<Mojo::UserAgent/"get">, and always result in a
L<Mojo::Transaction::HTTP> object, which has many useful attributes and methods. You can check for connection errors
with L<Mojo::Transaction/"result">, or access HTTP request and response information directly through
L<Mojo::Transaction/"req"> and L<Mojo::Transaction/"res">.

  use Mojo::UserAgent;

  # Request a resource and make sure there were no connection errors
  my $ua = Mojo::UserAgent->new;
  my $tx = $ua->get('https://docs.mojolicious.org/Mojo' => {Accept => 'text/plain'});
  my $res = $tx->result;

  # Decide what to do with its representation
  if    ($res->is_success)  { say $res->body }
  elsif ($res->is_error)    { say $res->message }
  elsif ($res->code == 301) { say $res->headers->location }
  else                      { say 'Whatever...' }

While methods like L<Mojo::Message::Response/"is_success"> and L<Mojo::Message::Response/"is_error"> serve as building
blocks for more sophisticated REST clients.

=head2 Web scraping

Scraping information from websites has never been this much fun before. The built-in HTML/XML parser L<Mojo::DOM> is
accessible through L<Mojo::Message/"dom"> and supports all CSS selectors that make sense for a standalone parser, it
can be a very powerful tool especially for testing web application.

  use Mojo::UserAgent;

  # Fetch website
  my $ua = Mojo::UserAgent->new;
  my $res = $ua->get('https://docs.mojolicious.org')->result;

  # Extract title
  say 'Title: ', $res->dom->at('head > title')->text;

  # Extract headings
  $res->dom('h1, h2, h3')->each(sub ($dom, $i) {
    say 'Heading: ', $dom->all_text;
  });

  # Visit all nodes recursively to extract more than just text
  for my $n ($res->dom->descendant_nodes->each) {

    # Text or CDATA node
    print $n->content if $n->type eq 'text' || $n->type eq 'cdata';

    # Also include alternate text for images
    print $n->{alt} if $n->type eq 'tag' && $n->tag eq 'img';
  }

For a full list of available CSS selectors see L<Mojo::DOM::CSS/"SELECTORS">.

=head2 JSON web services

Most web services these days are based on the JSON data-interchange format. That's why L<Mojolicious> comes with the
possibly fastest pure-Perl implementation L<Mojo::JSON> built right in, which is accessible through
L<Mojo::Message/"json">.

  use Mojo::UserAgent;
  use Mojo::URL;

  # Fresh user agent
  my $ua = Mojo::UserAgent->new;

  # Search MetaCPAN for "mojolicious" and list latest releases
  my $url = Mojo::URL->new('http://fastapi.metacpan.org/v1/release/_search');
  $url->query({q => 'mojolicious', sort => 'date:desc'});
  for my $hit (@{$ua->get($url)->result->json->{hits}{hits}}) {
    say "$hit->{_source}{name} ($hit->{_source}{author})";
  }

=head2 Basic authentication

You can just add username and password to the URL, an C<Authorization> header will be automatically generated.

  use Mojo::UserAgent;

  my $ua = Mojo::UserAgent->new;
  say $ua->get('https://sri:secret@example.com/hideout')->result->body;

If you're using L<Mojo::URL> to build the URL, be aware that the userinfo part will not be included if the object is
stringified. You'll have to pass the object itself to L<Mojo::UserAgent> or use L<Mojo::URL/"to_unsafe_string">.

  use Mojo::UserAgent;
  use Mojo::URL;

  my $ua  = Mojo::UserAgent->new;
  my $url = Mojo::URL->new('https://example.com/hideout')->userinfo('sri:secret');
  say $ua->get($url)->result->body;

=head2 Decorating follow-up requests

L<Mojo::UserAgent> can automatically follow redirects, the event L<Mojo::UserAgent/"start"> allows you direct access to
each transaction right after they have been initialized and before a connection gets associated with them.

  use Mojo::UserAgent;

  # User agent following up to 10 redirects
  my $ua = Mojo::UserAgent->new(max_redirects => 10);

  # Add a witty header to every request
  $ua->on(start => sub ($ua, $tx) {
    $tx->req->headers->header('X-Bender' => 'Bite my shiny metal ass!');
    say 'Request: ', $tx->req->url->clone->to_abs;
  });

  # Request that will most likely get redirected
  say 'Title: ', $ua->get('google.com')->result->dom->at('head > title')->text;

This even works for proxy C<CONNECT> requests.

=head2 Content generators

Content generators can be registered with L<Mojo::UserAgent::Transactor/"add_generator"> to generate the same type of
content repeatedly for multiple requests.

  use Mojo::UserAgent;
  use Mojo::Asset::File;

  # Add "stream" generator
  my $ua = Mojo::UserAgent->new;
  $ua->transactor->add_generator(stream => sub ($transactor, $tx, $path) {
    $tx->req->content->asset(Mojo::Asset::File->new(path => $path));
  });

  # Send multiple files streaming via PUT and POST
  $ua->put('http://example.com/upload'  => stream => '/home/sri/mojo.png');
  $ua->post('http://example.com/upload' => stream => '/home/sri/minion.png');

The C<json>, C<form> and C<multipart> content generators are always available.

  use Mojo::UserAgent;

  # Send "application/json" content via PATCH
  my $ua = Mojo::UserAgent->new;
  my $tx = $ua->patch('http://api.example.com' => json => {foo => 'bar'});

  # Send query parameters via GET
  my $tx2 = $ua->get('search.example.com' => form => {q => 'test'});

  # Send "application/x-www-form-urlencoded" content via POST
  my $tx3 = $ua->post('http://search.example.com' => form => {q => 'test'});

  # Send "multipart/form-data" content via PUT
  my $tx4 = $ua->put('upload.example.com' => form => {test => {content => 'Hello World!'}});

  # Send custom multipart content via PUT
  my $tx5 = $ua->put('api.example.com' => multipart => ['Hello', 'World!']);

For more information about available content generators see also L<Mojo::UserAgent::Transactor/"tx">.

=head2 Large file downloads

When downloading large files with L<Mojo::UserAgent> you don't have to worry about memory usage at all, because it will
automatically stream everything above 250KiB into a temporary file, which can then be moved into a permanent file with
L<Mojo::Message/"save_to">.

  use Mojo::UserAgent;

  # Fetch the latest Mojolicious tarball
  my $ua = Mojo::UserAgent->new(max_redirects => 5);
  my $tx = $ua->get('https://www.github.com/mojolicious/mojo/tarball/main');
  $tx->result->save_to('mojo.tar.gz');

To protect you from excessively large files there is also a limit of 2GiB by default, which you can tweak with the
attribute L<Mojo::UserAgent/"max_response_size">.

  # Increase limit to 10GiB
  $ua->max_response_size(10737418240);

=head2 Large file upload

Uploading a large file is even easier.

  use Mojo::UserAgent;

  # Upload file via POST and "multipart/form-data"
  my $ua = Mojo::UserAgent->new;
  $ua->post('example.com/upload' => form => {image => {file => '/home/sri/hello.png'}});

And once again you don't have to worry about memory usage, all data will be streamed directly from the file.

=head2 Streaming response

Receiving a streaming response can be really tricky in most HTTP clients, but L<Mojo::UserAgent> makes it actually
easy.

  use Mojo::UserAgent;

  # Accept responses of indefinite size
  my $ua = Mojo::UserAgent->new(max_response_size => 0);

  # Build a normal transaction
  my $tx = $ua->build_tx(GET => 'http://example.com');

  # Replace "read" events to disable default content parser
  $tx->res->content->unsubscribe('read')->on(read => sub ($content, $bytes) {
    say "Streaming: $bytes";
  });

  # Process transaction
  $tx = $ua->start($tx);

The event L<Mojo::Content/"read"> will be emitted for every chunk of data that is received, even chunked transfer
encoding and gzip content encoding will be handled transparently if necessary.

=head2 Streaming request

Sending a streaming request is almost just as easy.

  use Mojo::UserAgent;

  # Build a normal transaction
  my $ua = Mojo::UserAgent->new;
  my $tx = $ua->build_tx(POST => 'http://example.com');

  # Prepare body
  my $body = 'Hello World!';
  $tx->req->headers->content_length(length $body);

  # Start writing directly with a drain callback
  my $drain = sub ($content) {
    my $chunk = substr $body, 0, 1, '';
    $content->write($chunk, length $body ? __SUB__ : undef);
  };
  $tx->req->content->$drain;

  # Process transaction
  $tx = $ua->start($tx);

The drain callback passed to L<Mojo::Content/"write"> will be executed whenever the entire previous chunk of data has
actually been written.

=head2 Non-blocking

L<Mojo::UserAgent> has been designed from the ground up to be non-blocking, the whole blocking API is just a simple
convenience wrapper. Especially for high latency tasks like web crawling this can be extremely useful, because you can
keep many concurrent connections active at the same time.

  use Mojo::UserAgent;
  use Mojo::IOLoop;

  # Concurrent non-blocking requests
  my $ua = Mojo::UserAgent->new;
  $ua->get('https://metacpan.org/search?q=mojo' => sub ($ua, $mojo) {
    say $mojo->result->dom->at('title')->text;
  });
  $ua->get('https://metacpan.org/search?q=minion' => sub ($ua, $minion) {
    say $minion->result->dom->at('title')->text;
  });

  # Start event loop if necessary
  Mojo::IOLoop->start unless Mojo::IOLoop->is_running;

But don't try to open too many connections to one server at the same time, it might get overwhelmed. Better use a queue
to process requests in smaller batches.

  use Mojo::Promise;
  use Mojo::UserAgent;

  my @urls = (
    'https://docs.mojolicious.org/Mojo/DOM',  'https://docs.mojolicious.org/Mojo',
    'https://docs.mojolicious.org/Mojo/File', 'https://docs.mojolicious.org/Mojo/URL'
  );

  # User agent with a custom name, following up to 5 redirects
  my $ua = Mojo::UserAgent->new(max_redirects => 5);
  $ua->transactor->name('MyParallelCrawler 1.0');

  # Use a promise to keep the event loop running until we are done
  my $promise = Mojo::Promise->new;
  my $count = 0;
  my $fetch = sub {

    # Stop if there are no more URLs
    return unless my $url = shift @urls;

    # Fetch the next title
    $ua->get($url => sub ($ua, $tx) {
      say "$url: ", $tx->result->dom->at('title')->text;

      # Next request
      __SUB__->();
      $promise->resolve if --$count == 0;
    });
    $count++;
  };

  # Process two requests at a time
  $fetch->() for 1 .. 2;
  $promise->wait;

It is also strongly recommended to respect every sites C<robots.txt> file as well as terms of service, and to wait a
little before reopening connections to the same host, or the operators might be forced to block your access.

=head2 Concurrent blocking requests

You might have seen L<Mojo::Promise/"wait"> already in some examples above. It is used to make non-blocking operations
portable, allowing them to work inside an already running event loop or start one on demand.

  use Mojo::UserAgent;
  use Mojo::Promise;

  # Synchronize non-blocking requests with promises
  my $ua = Mojo::UserAgent->new;
  my $mojo_promise   = $ua->get_p('https://metacpan.org/search?q=mojo');
  my $minion_promise = $ua->get_p('https://metacpan.org/search?q=minion');
  Mojo::Promise->all($mojo_promise, $minion_promise)->then(sub ($mojo, $minion) {
    say $mojo->[0]->result->dom->at('title')->text;
    say $minion->[0]->result->dom->at('title')->text;
  })->wait;

=head2 WebSockets

WebSockets are not just for the server-side, you can use L<Mojo::UserAgent/"websocket_p"> to open new connections,
which are always non-blocking. The WebSocket handshake uses HTTP, and is a normal C<GET> request with a few additional
headers. It can even contain cookies, and is followed by a C<101> response from the server, notifying our user agent
that the connection has been established and it can start using the bi-directional WebSocket protocol.

  use Mojo::UserAgent;
  use Mojo::Promise;

  # Open WebSocket to echo service
  my $ua = Mojo::UserAgent->new;
  $ua->websocket_p('wss://ws.postman-echo.com/raw')->then(sub ($tx) {

    # Prepare a followup promise so we can wait for messages
    my $promise = Mojo::Promise->new;

    # Wait for WebSocket to be closed
    $tx->on(finish => sub ($tx, $code, $reason) {
      say "WebSocket closed with status $code.";
      $promise->resolve;
    });

    # Close WebSocket after receiving one message
    $tx->on(message => sub ($tx, $msg) {
      say "WebSocket message: $msg";
      $tx->finish;
    });

    # Send a message to the server
    $tx->send('Hi!');

    # Insert a new promise into the promise chain
    return $promise;
  })->catch(sub ($err) {

    # Handle failed WebSocket handshakes and other exceptions
    warn "WebSocket error: $err";
  })->wait;

=head2 UNIX domain sockets

Not just TCP/IP sockets are supported, but also UNIX domain sockets, which can have significant security and
performance benefits when used for inter-process communication. Instead of C<http://> and C<ws://> you can use the
C<http+unix://> and C<ws+unix://> schemes, and pass along a percent encoded path (C</> becomes C<%2F>) instead of a
hostname.

  use Mojo::UserAgent;
  use Mojo::Promise;

  # GET request via UNIX domain socket "/tmp/foo.sock"
  my $ua = Mojo::UserAgent->new;
  say $ua->get('http+unix://%2Ftmp%2Ffoo.sock/index.html')->result->body;

  # GET request with HOST header via UNIX domain socket "/tmp/bar.sock"
  my $tx = $ua->get('http+unix://%2Ftmp%2Fbar.sock' => {Host => 'example.com'});
  say $tx->result->body;

  # WebSocket connection via UNIX domain socket "/tmp/baz.sock"
  $ua->websocket_p('ws+unix://%2Ftmp%2Fbaz.sock/echo')->then(sub ($tx) {

    my $promise = Mojo::Promise->new;
    $tx->on(finish => sub ($tx) { $promise->resolve });

    $tx->on(message => sub ($tx, $msg) {
      say "WebSocket message: $msg";
      $tx->finish;
    });
    $tx->send('Hi!');

    return $promise;
  })->catch(sub ($err) {
    warn "WebSocket error: $err";
  })->wait;

You can set the C<Host> header manually to pass along a hostname.

=head2 Command line

Don't you hate checking huge HTML files from the command line? Thanks to the command L<Mojolicious::Command::get> that
is about to change. You can just pick the parts that actually matter with the CSS selectors from L<Mojo::DOM> and JSON
Pointers from L<Mojo::JSON::Pointer>.

  $ mojo get https://mojolicious.org 'head > title'

How about a list of all id attributes?

  $ mojo get https://mojolicious.org '*' attr id

Or the text content of all heading tags?

  $ mojo get https://mojolicious.org 'h1, h2, h3' text

Maybe just the text of the third heading?

  $ mojo get https://mojolicious.org 'h1, h2, h3' 3 text

You can also extract all text from nested child elements.

  $ mojo get https://mojolicious.org '#mojobar' all

The request can be customized as well.

  $ mojo get -M POST -H 'X-Bender: Bite my shiny metal ass!' http://google.com

Store response data by redirecting C<STDOUT>.

  $ mojo get mojolicious.org > example.html

Pass request data by redirecting C<STDIN>.

  $ mojo get -M PUT mojolicious.org < example.html

Or use the output of another program.

  $ echo 'Hello World' | mojo get -M PUT https://mojolicious.org

Submit forms as C<application/x-www-form-urlencoded> content.

  $ mojo get -M POST -f 'q=Mojo' -f 'size=5' https://metacpan.org/search

And upload files as C<multipart/form-data> content.

  $ mojo get -M POST -f 'upload=@example.html' mojolicious.org

You can follow redirects and view the headers for all messages.

  $ mojo get -r -v http://google.com 'head > title'

Extract just the information you really need from JSON data structures.

  $ mojo get https://fastapi.metacpan.org/v1/author/SRI /name

This can be an invaluable tool for testing your applications.

  $ ./myapp.pl get /welcome 'head > title'

=head2 One-liners

For quick hacks and especially testing, L<ojo> one-liners are also a great choice.

  $ perl -Mojo -E 'say g("mojolicious.org")->dom->at("title")->text'

=head1 APPLICATIONS

Fun L<Mojolicious> application hacks for all occasions.

=head2 Basic authentication

Basic authentication data will be automatically extracted from the C<Authorization> header.

  use Mojolicious::Lite -signatures;
  use Mojo::Util qw(secure_compare);

  get '/' => sub ($c) {

    # Check for username "Bender" and password "rocks"
    return $c->render(text => 'Hello Bender!') if secure_compare $c->req->url->to_abs->userinfo, 'Bender:rocks';

    # Require authentication
    $c->res->headers->www_authenticate('Basic');
    $c->render(text => 'Authentication required!', status => 401);
  };

  app->start;

This can be combined with TLS for a secure authentication mechanism.

  $ ./myapp.pl daemon -l 'https://*:3000?cert=./server.crt&key=./server.key'

=head2 Adding a configuration file

Adding a configuration file to your application is as easy as adding a file to its home directory and loading the
plugin L<Mojolicious::Plugin::Config>. The default name is based on the value of L<Mojolicious/"moniker"> (C<myapp>),
appended with a C<.conf> extension (C<myapp.conf>).

  $ mkdir myapp
  $ cd myapp
  $ touch myapp.pl
  $ chmod 744 myapp.pl
  $ echo '{name => "my Mojolicious application"};' > myapp.conf

Configuration files themselves are just Perl scripts that return a hash reference with configuration settings of your
choice. All those settings are then available through the method L<Mojolicious/"config"> and the helper
L<Mojolicious::Plugin::DefaultHelpers/"config">.

  use Mojolicious::Lite;

  plugin 'Config';

  my $name = app->config('name');
  app->log->debug("Welcome to $name");

  get '/' => 'with_config';

  app->start;
  __DATA__
  @@ with_config.html.ep
  <!DOCTYPE html>
  <html>
    <head><title><%= config 'name' %></title></head>
    <body>Welcome to <%= config 'name' %></body>
  </html>

Alternatively you can also use configuration files in the JSON format with L<Mojolicious::Plugin::JSONConfig>.

=head2 Adding a plugin to your application

To organize your code better and to prevent helpers from cluttering your application, you can use application specific
plugins.

  $ mkdir -p lib/MyApp/Plugin
  $ touch lib/MyApp/Plugin/MyHelpers.pm

They work just like normal plugins and are also subclasses of L<Mojolicious::Plugin>. Nested helpers with a prefix
based on the plugin name are an easy way to avoid conflicts.

  package MyApp::Plugin::MyHelpers;
  use Mojo::Base 'Mojolicious::Plugin', -signatures;

  sub register ($self, $app, $conf) {
    $app->helper('my_helpers.render_with_header' => sub ($c, @args) {
      $c->res->headers->header('X-Mojo' => 'I <3 Mojolicious!');
      $c->render(@args);
    });
  }

  1;

You can have as many application specific plugins as you like, the only difference to normal plugins is that you load
them using their full class name.

  use Mojolicious::Lite -signatures;

  use lib qw(lib);

  plugin 'MyApp::Plugin::MyHelpers';

  get '/' => sub ($c) {
    $c->my_helpers->render_with_header(text => 'I ♥ Mojolicious!');
  };

  app->start;

Of course these plugins can contain more than just helpers, take a look at L<Mojolicious::Plugins/"PLUGINS"> for a few
ideas.

=head2 Adding commands to Mojolicious

By now you've probably used many of the built-in commands described in L<Mojolicious::Commands>, but did you know that
you can just add new ones and that they will be picked up automatically by the command line interface if they are
placed in a directory from C<@INC>?

  package Mojolicious::Command::spy;
  use Mojo::Base 'Mojolicious::Command', -signatures;

  has description => 'Spy on application';
  has usage       => "Usage: APPLICATION spy [TARGET]\n";

  sub run ($self, @args) {

    # Leak secret passphrases
    if ($args[0] eq 'secrets') { say for @{$self->app->secrets} }

    # Leak mode
    elsif ($args[0] eq 'mode') { say $self->app->mode }
  }

  1;

Command line arguments are passed right through and there are many useful attributes and methods in
L<Mojolicious::Command> that you can use or overload.

  $ mojo spy secrets
  HelloWorld

  $ ./script/myapp spy secrets
  secr3t

And to make your commands application specific, just add a custom namespace to L<Mojolicious::Commands/"namespaces">
and use a class name like C<MyApp::Command::spy> instead of C<Mojolicious::Command::spy>.

  # Application
  package MyApp;
  use Mojo::Base 'Mojolicious', -signatures;

  sub startup ($self) {

    # Add another namespace to load commands from
    push @{$self->commands->namespaces}, 'MyApp::Command';
  }

  1;

The options C<-h>/C<--help>, C<--home> and C<-m>/C<--mode> are handled automatically by L<Mojolicious::Commands> and
are shared by all commands.

  $ ./script/myapp spy -m production mode
  production

For a full list of shared options see L<Mojolicious::Commands/"SYNOPSIS">.

=head2 Running code against your application

Ever thought about running a quick one-liner against your L<Mojolicious> application to test something? Thanks to the
command L<Mojolicious::Command::eval> you can do just that, the application object itself can be accessed via C<app>.

  $ mojo generate lite-app myapp.pl
  $ ./myapp.pl eval 'say for @{app->static->paths}'
  $ ./myapp.pl eval 'say for sort keys %{app->renderer->helpers}'

The C<verbose> options will automatically print the return value or returned data structure to C<STDOUT>.

  $ ./myapp.pl eval -v 'app->static->paths->[0]'
  $ ./myapp.pl eval -V 'app->static->paths'

=head2 Making your application installable

Ever thought about releasing your L<Mojolicious> application to CPAN? It's actually much easier than you might think.

  $ mojo generate app MyApp
  $ cd my_app
  $ mv public lib/MyApp/
  $ mv templates lib/MyApp/

The trick is to move the C<public> and C<templates> directories so they can get automatically installed with the
modules. Additionally author commands from the C<Mojolicious::Command::Author> namespace are not usually wanted by an
installed application so they can be excluded.

  # Application
  package MyApp;
  use Mojo::Base 'Mojolicious', -signatures;

  use Mojo::File qw(curfile);
  use Mojo::Home;

  # Every CPAN module needs a version
  our $VERSION = '1.0';

  sub startup ($self) {

    # Switch to installable home directory
    $self->home(Mojo::Home->new(curfile->sibling('MyApp')));

    # Switch to installable "public" directory
    $self->static->paths->[0] = $self->home->child('public');

    # Switch to installable "templates" directory
    $self->renderer->paths->[0] = $self->home->child('templates');

    # Exclude author commands
    $self->commands->namespaces(['Mojolicious::Command']);

    my $r = $self->routes;
    $r->get('/')->to('example#welcome');
  }

  1;

Finally there is just one small change to be made to the application script. The shebang line becomes the recommended
C<#!perl>, which the toolchain can rewrite to the proper shebang during installation.

  #!perl

  use strict;
  use warnings;

  use Mojo::File qw(curfile);
  use lib curfile->dirname->sibling('lib')->to_string;
  use Mojolicious::Commands;

  # Start command line interface for application
  Mojolicious::Commands->start_app('MyApp');

That's really everything, now you can package your application like any other CPAN module.

  $ ./script/my_app generate makefile
  $ perl Makefile.PL
  $ make test
  $ make manifest
  $ make dist

And if you have a PAUSE account (which can be requested at L<http://pause.perl.org>) even upload it.

  $ mojo cpanify -u USER -p PASS MyApp-0.01.tar.gz

=head2 Proxy

While every L<Mojolicious> application has the built-in user agent L<Mojolicious::Plugin::DefaultHelpers/"ua"> for you
to perform requests to backend web services, this is not always the most efficient solution. The specialized proxy
helpers L<Mojolicious::Plugin::DefaultHelpers/"proxy-E<gt>get_p"> and
L<Mojolicious::Plugin::DefaultHelpers/"proxy-E<gt>start_p"> can stream response content straight to the client, as soon
as a new chunk of data is received from the backend web service. Additionally they will take care of removing hop-by-hop
headers and protect you automatically from backpressure issues. Which can happen in situations where the connection to
the backend web service is faster than the connection to the client and data forwarding needs to be throttled. And the
best of all, everything happens non-blocking, that means your web server can process other requestes concurrently while
waiting for I/O.

  use Mojolicious::Lite -signatures;

  # Just forward the response
  get '/' => sub ($c) {
    $c->proxy->get_p('https://mojolicious.org')->catch(sub ($err) {
      $c->log->error("Proxy error: $err");
      $c->render(text => 'Could not connect to backend web service!', status => 400);
    });
  };

  # Forward response and customize a few things
  get '/*docs' => sub ($c) {

    # Custom request
    my $tx = $c->ua->build_tx(GET => 'https://docs.mojolicious.org');
    my $docs = $c->param('docs');
    $tx->req->url->path("/$docs");
    $tx->req->headers->user_agent('MojoProxy/1.0');

    # Start non-blocking request
    $c->proxy->start_p($tx)->catch(sub ($err) {
      $c->log->error("Proxy error: $err");
      $c->render(text => 'Could not connect to backend web service!', status => 400);
    });

    # Custom response
    $tx->res->content->once(body => sub ($content) {
      $c->res->headers->server('MojoProxy/1.0');
    });
  };

  app->start;

All proxy helpers return a L<Mojo::Promise> object, which should be used to handle connection errors to backend web
services gracefully. And if you ever need to forward all headers from the client to the backend web service, make sure
to use L<Mojo::Headers/"dehop"> to remove all hop-by-hop headers.

  # Clone and modify request headers
  my $headers = $c->req->headers->clone->dehop;
  $headers->accept('application/json');
  my $tx = $c->ua->build_tx(PUT => 'https://mojolicious.org' => $headers->to_hash);

=head2 Hello World

If every byte matters this is the smallest C<Hello World> application you can write with L<Mojolicious::Lite>.

  use Mojolicious::Lite;
  any {text => 'Hello World!'};
  app->start;

It works because all routes without a pattern default to C</> and automatic rendering kicks in even if no actual code
gets executed by the router. The renderer just picks up the C<text> value from the stash and generates a response.

=head2 Hello World one-liners

The C<Hello World> example above can get even a little bit shorter in an L<ojo> one-liner.

  $ perl -Mojo -E 'a({text => "Hello World!"})->start' daemon

And you can use all the commands from L<Mojolicious::Commands>.

  $ perl -Mojo -E 'a({text => "Hello World!"})->start' get -v /

=head1 MORE

You can continue with L<Mojolicious::Guides> now or take a look at the L<Mojolicious
wiki|https://github.com/mojolicious/mojo/wiki>, which contains a lot more documentation and examples by many different
authors.

=head1 SUPPORT

If you have any questions the documentation might not yet answer, don't hesitate to ask in the
L<Forum|https://forum.mojolicious.org>, on L<IRC|https://web.libera.chat/#mojo>, or
L<Matrix|https://matrix.to/#/#mojo:matrix.org>.

=cut