path: root/lib/Dancer2/Manual/Deployment.pod

# PODNAME: Dancer2::Manual::Deployment
# ABSTRACT: common ways to put your Dancer app into use

__END__

=pod

=encoding UTF-8

=head1 NAME

Dancer2::Manual::Deployment - common ways to put your Dancer app into use

=head1 VERSION

version 1.1.0

=head1 DESCRIPTION

Dancer has been designed to be flexible, and this flexibility extends to your
choices when deploying your Dancer app.

=head2 Running stand-alone

To start your application, just run plackup:

    $ plackup bin/app.psgi
    HTTP::Server::PSGI: Accepting connections at http://0:5000/

Point your browser at it, and away you go!

This option can be useful for small personal web apps or internal apps, but if
you want to make your app available to the world, it probably won't suit you.

=head3 Auto Reloading the Application

While developing your application, it is often handy to have the server
automatically reload your application when changes are made. There are
two recommended ways of handling this with Dancer: using C< plackup -r >
and L<Plack::Loader::Shotgun>. Both have their advantages and disadvantages
(which will be explained below).

Regardless of the method you use, it is B< not > recommended that you
automatically reload your applications in a production environment, for
reasons of performance, deployment best practices, etc.

For Dancer 1 programmers that used the C< auto_reload > option, please use
one of these alternatives instead:

=head4 Auto reloading with C< plackup -r >

Plack's built-in reloader will reload your application anytime a file in
your application's directory (usually, F< /bin >) changes. You will likely
want to monitor your F< lib/ > directory too, using the C< -R > option:

    $ plackup -r -R lib bin/app.psgi

There is a performance hit associated with this, as Plack will spin off
a separate process that monitors files in the application and other
specified directories. If the timestamp of any files in a watched
directory changes, the application is recompiled and reloaded.

See the L<plackup> docs for more information on the C< -r > and C< -R >
options.

=head4 Auto reloading with plackup and Shotgun

There may be circumstances where Plack's built-in reloader won't work for
you, be it for the way it looks for changes, or because there are many
directories you need to monitor, or you want to reload the application any
time one of the modules in Perl's F< lib/ > path changes.
L<Plack::Loader::Shotgun> makes this easy by recompiling the application
on every request.

To use Shotgun, specify it using the loader argument to C< plackup (-L) >:

    $ plackup -L Shotgun bin/app.psgi

The Shotgun, while effective, can quickly cause you performance issues, even
during the development phase of your application. As the number of plugins
you use in your application grows, as the number of static resources (images,
etc.) grows, the more requests your server process needs to handle. Since
each request recompiles the application, even simple page refreshes can get
unbearably slow over time. Use with caution.

You can bypass Shotgun's auto-reloading of specific modules with the
C< -M > switch:

    $ plackup -L Shotgun -M<MyApp::Foo> -M<MyApp::Bar> bin/app.psgi

On Windows, Shotgun loader is known to cause huge memory leaks in a
fork-emulation layer. If you are aware of this and still want to run the
loader, please use the following command:

    > set PLACK_SHOTGUN_MEMORY_LEAK=1 && plackup -L Shotgun bin\app.psgi
    HTTP::Server::PSGI: Accepting connections at http://0:5000/

B<Please note: > if you are using Dancer 2's asynchronous capabilities, using
Shotgun will kill Twiggy. If you need async processing, consider an
alternative to Shotgun.

=head2 Running under Apache

You can run your Dancer app from Apache using the following examples:

=head3 As a CGI script

In its simplest form, your Dancer app can be run as a simple CGI script
out-of-the-box. You will need to enable the Apache mod_cgi or mod_cgid modules
(C<a2enmod cgi> or C<a2enmod cgid> on Debian-based systems) and mod_rewrite
(C<a2enmod rewrite>). The Perl module L<Plack::Runner> is required.

The following is an example apache configuration. Depending on your Apache
configuration layout, this should be placed in C<httpd.conf> or
C<sites-available/*site*>. The configuration options can also be placed in
C<.htaccess> files if you prefer.

    <VirtualHost *:80>
        ServerName www.example.com

        # /srv/www.example.com is the root of your
        # dancer application
        DocumentRoot /srv/www.example.com/public

        ServerAdmin you@example.com

        <Directory "/srv/www.example.com/public">
            AllowOverride None
            Options +ExecCGI -MultiViews +SymLinksIfOwnerMatch
            AddHandler cgi-script .cgi
            # Apache 2.2
            Order allow,deny
            Allow from all
            # Apache 2.4
            Require all granted
        </Directory>

        RewriteEngine On
        RewriteCond %{REQUEST_FILENAME} !-f
        RewriteRule ^(.*)$ /dispatch.cgi$1 [QSA,L]

        ErrorLog  /var/log/apache2/www.example.com-error.log
        CustomLog /var/log/apache2/www.example.com-access_log common
    </VirtualHost>

Now you can access your dancer application URLs as if you were using the
embedded web server.

    http://www.example.com/

This option is a no-brainer, easy to setup and low maintenance, but serves
requests slower than all other options, as each time a request is made to your
server, Apache will start your application. This might be suitable for a small,
occasionally-used sites, as the application is not using resources when it is
not being accessed. For anything more, you probably want to use FastCGI instead
(see next section).

To list all currently loaded modules, type C<apachectl -M>
(C<apache2ctl -M> on Debian/Ubuntu).

=head3 As a FastCGI script

This has all the easy-to-setup and low-maintenance advantages of CGI, but is
much faster for each request, as it keeps a copy of the application running all
the time.

You will still need to enable C<mod_rewrite>, but will need to use a FastCGI
module instead of a CGI module. There are 3 available:
L<mod_fcgid|http://httpd.apache.org/mod_fcgid/>,
L<mod_fastcgi|http://www.fastcgi.com/> and
L<mod_proxy_fcgi|https://httpd.apache.org/docs/trunk/mod/mod_proxy_fcgi.html>.
For this example, we will use mod_fastcgi (C<a2enmod fastcgi> in Debian).

The CGI configuration above now changes as follows (differences highlighted
with XXX):

    <VirtualHost *:80>
        ServerName www.example.com

        # /srv/www.example.com is the root of your
        # dancer application
        DocumentRoot /srv/www.example.com/public

        ServerAdmin you@example.com

        # XXX Start a FastCGI server to run in the background
        FastCgiServer /srv/www.example.com/public/dispatch.fcgi

        <Directory "/srv/www.example.com/public">
            AllowOverride None
            Options +ExecCGI -MultiViews +SymLinksIfOwnerMatch
            # XXX Use FastCGI handler instead of CGI
            AddHandler fastcgi-script .fcgi
            # Apache 2.2
            Order allow,deny
            Allow from all
            # Apache 2.4
            Require all granted
        </Directory>

        RewriteEngine On
        RewriteCond %{REQUEST_FILENAME} !-f
        # Run FastCGI dispatcher instead of CGI dispatcher
        RewriteRule ^(.*)$ /dispatch.fcgi$1 [QSA,L]

        ErrorLog  /var/log/apache2/www.example.com-error.log
        CustomLog /var/log/apache2/www.example.com-access_log common
    </VirtualHost>

This is the easiest way to get a production server up and running, as there is
no need to worry about daemonizing your application. Apache manages all that
for you.

=head4 Reloading your application

You can use C<apache2ctl restart> or C<apache2ctl graceful> to reload your
application. The latter will be more friendly to your users in a production
environment. If your application loads relatively quickly, then it should go
unnoticed.

=head4 Configuration

See L<http://www.fastcgi.com/mod_fastcgi/docs/mod_fastcgi.html> for FastCGI
configuration options. An example configuration:

   FastCgiServer /srv/www.example.com/public/dispatch.fcgi -processes 5 -initial-env DANCER_ENVIRONMENT="production"

=head3 With Plack

You can run your app from Apache using PSGI (Plack), with a config like the
following:

    <VirtualHost myapp.example.com>
        ServerName www.myapp.example.com
        ServerAlias myapp.example.com
        DocumentRoot /websites/myapp.example.com

        <Directory /home/myapp/myapp>
            AllowOverride None
            Order allow,deny
            Allow from all
        </Directory>

        <Location />
            SetHandler perl-script
            PerlResponseHandler Plack::Handler::Apache2
            PerlSetVar psgi_app /websites/myapp.example.com/app.psgi
        </Location>

        ErrorLog  /websites/myapp.example.com/logs/error_log
        CustomLog /websites/myapp.example.com/logs/access_log common
    </VirtualHost>

To set the environment you want to use for your application (production or
development), you can set it this way:

    <VirtualHost>
        ...
        SetEnv DANCER_ENVIRONMENT "production"
        ...
    </VirtualHost>

=head3 Running multiple applications under the same virtualhost

If you want to deploy multiple applications under the same C<VirtualHost>
(using one application per directory, for example) you can use the following
example Apache configuration.

This example uses the FastCGI dispatcher that comes with Dancer, but you should
be able to adapt this to use any other way of deployment described in this
guide. The only purpose of this example is to show how to deploy multiple
applications under the same base directory/virtualhost.

    <VirtualHost *:80>
        ServerName localhost
        DocumentRoot "/path/to/rootdir"
        RewriteEngine On
        RewriteCond %{REQUEST_FILENAME} !-f

        <Directory "/path/to/rootdir">
            AllowOverride None
            Options +ExecCGI -MultiViews +SymLinksIfOwnerMatch
            Order allow,deny
            Allow from all
            AddHandler fastcgi-script .fcgi
        </Directory>

        RewriteRule /App1(.*)$ /App1/public/dispatch.fcgi$1 [QSA,L]
        RewriteRule /App2(.*)$ /App2/public/dispatch.fcgi$1 [QSA,L]
        ...
        RewriteRule /AppN(.*)$ /AppN/public/dispatch.fcgi$1 [QSA,L]
    </VirtualHost>

Of course, if your Apache configuration allows that, you can put the
RewriteRules in a .htaccess file directly within the application's directory,
which lets you add a new application without changing the Apache configuration.

=head2 Running on PSGI-based Perl webservers

A number of Perl web servers supporting PSGI are available on cpan:

=over 4

=item L<Starman>

C<Starman> is a high performance web server, with support for preforking,
signals, multiple interfaces, graceful restarts and dynamic worker pool
configuration.

=item L<Twiggy>

C<Twiggy> is an C<AnyEvent> web server, it's light and fast.

=item L<Corona>

C<Corona> is a C<Coro> based web server.

=back

Similar to running standalone, use plackup to start your application
(see L<Plack> and specific servers above for all available options):

    $ plackup bin/app.psgi
    $ plackup -E deployment -s Starman --workers=10 -p 5001 -a bin/app.psgi

As you can see, the scaffolded Perl script for your app can be used as a PSGI
startup file.

=head3 Enabling content compression

Content compression (gzip, deflate) can be easily enabled via a Plack
middleware (see L<Plack/Plack::Middleware>): L<Plack::Middleware::Deflater>.
It's a middleware to encode the response body in gzip or deflate, based on
the C<Accept-Encoding> HTTP request header.

Enable it as you would enable any Plack middleware. First you need to
install L<Plack::Middleware::Deflater>, then in the handler (usually
F<app.psgi>) edit it to use L<Plack::Builder>, as described above:

    use Dancer2;
    use MyWebApp;
    use Plack::Builder;

    builder {
        enable 'Deflater';
        dance;
    };

To test if content compression works, trace the HTTP request and response
before and after enabling this middleware. Among other things, you should
notice that the response is gzip or deflate encoded, and contains a header
C<Content-Encoding> set to C<gzip> or C<deflate>.

=head3 Creating a service

You can turn your app into proper service running in background using one of
the following examples:

=head4 Using Ubic

L<Ubic> is an extensible perlish service manager. You can use it to start
and stop any services, automatically start them on reboots or daemon
failures, and implement custom status checks.

A basic PSGI service description (usually in C</etc/ubic/service/application>):

    use parent qw(Ubic::Service::Plack);

    # if your application is not installed in @INC path:
    sub start {
        my $self = shift;
        $ENV{PERL5LIB} = '/path/to/your/application/lib';
        $self->SUPER::start(@_);
    }

    __PACKAGE__->new(
        server => 'Starman',
        app => '/path/to/your/application/app.psgi',
        port => 5000,
        user => 'www-data',
    );

Run C<ubic start application> to start the service.

=head4 Using daemontools

daemontools is a collection of tools for managing UNIX services. You can use
it to easily start/restart/stop services.

A basic script to start an application: (in C</service/application/run>)

    #!/bin/sh

    # if your application is not installed in @INC path:
    export PERL5LIB='/path/to/your/application/lib'

    exec 2>&1 \
    /usr/local/bin/plackup -s Starman -a /path/to/your/application/app.psgi -p 5000

=head2 Running stand-alone behind a proxy / load balancer

Another option would be to run your app stand-alone as described above, but then
use a proxy or load balancer to accept incoming requests (on the standard port
80, say) and feed them to your Dancer app. Also, in this case you might want
to look at the C<behind_proxy> configuration option, to make sure that all the
URLs are constructed properly.

    behind_proxy: 1

This setup can be achieved using various software; examples would include:

=head3 Using Apache's mod_proxy

You could set up a C<VirtualHost> for your web app, and proxy all requests through
to it:

    <VirtualHost mywebapp.example.com:80>
    ProxyPass / http://localhost:3000/
    ProxyPassReverse / http://localhost:3000/
    </VirtualHost>

Or, if you want your webapp to share an existing VirtualHost, you could have
it under a specified dir:

    ProxyPass /mywebapp/ http://localhost:3000/
    ProxyPassReverse /mywebapp/ http://localhost:3000/

It is important for you to note that the Apache2 modules C<mod_proxy> and
C<mod_proxy_http> must be enabled:

    $ a2enmod proxy
    $ a2enmod proxy_http

It is also important to set permissions for proxying for security purposes,
below is an example.

    <Proxy *>
      Order allow,deny
      Allow from all
    </Proxy>

=head3 Using perlbal

C<Perlbal> is a single-threaded event-based server written in Perl supporting
HTTP load balancing, web serving, and a mix of the two, available from
L<http://www.danga.com/perlbal/>

It processes hundreds of millions of requests a day just for LiveJournal, Vox
and TypePad and dozens of other "Web 2.0" applications.

It can also provide a management interface to let you see various information on
requests handled etc.

It could easily be used to handle requests for your Dancer apps, too.

It can be easily installed from CPAN:

    perl -MCPAN -e 'install Perlbal'

Once installed, you'll need to write a configuration file.  See the examples
provided with perlbal, but you'll probably want something like:

    CREATE POOL my_dancers
    POOL my_dancers ADD 10.0.0.10:3030
    POOL my_dancers ADD 10.0.0.11:3030
    POOL my_dancers ADD 10.0.0.12:3030
    POOL my_dancers ADD 10.0.0.13:3030

    CREATE SERVICE my_webapp
    SET listen          = 0.0.0.0:80
    SET role            = reverse_proxy
    SET pool            = my_dancers
    SET persist_client  = on
    SET persist_backend = on
    SET verify_backend  = on
    ENABLE my_webapp

=head3 Using balance

C<balance> is a simple load-balancer from Inlab Software, available from
L<http://www.inlab.de/balance.html>.

It could be used simply to hand requests to a standalone Dancer app. You could
even run several instances of your Dancer app, on the same machine or on several
machines, and use a machine running C<balance> to distribute the requests between
them, for some serious heavy traffic handling!

To listen on port 80, and send requests to a Dancer app on port 3000:

    balance http localhost:3000

To listen on a specified IP only on port 80, and distribute requests between
multiple Dancer apps on multiple other machines:

    balance -b 10.0.0.1 80 10.0.0.2:3000 10.0.0.3:3000 10.0.0.4:3000

=head3 Using Lighttpd

You can use Lighttpd's C<mod_proxy>:

    $HTTP["url"] =~ "/application" {
        proxy.server = (
            "/" => (
                "application" => ( "host" => "127.0.0.1", "port" => 3000 )
            )
        )
    }

This configuration will proxy all request to the B</application> path to the
path B</> on localhost:3000.

=head3 Using Nginx

with Nginx:

    upstream backendurl {
        server unix:THE_PATH_OF_YOUR_PLACKUP_SOCKET_HERE.sock;
    }

    server {
      listen       80;
      server_name YOUR_HOST_HERE;

      access_log /var/log/YOUR_ACCESS_LOG_HERE.log;
      error_log  /var/log/YOUR_ERROR_LOG_HERE.log info;

      root YOUR_ROOT_PROJECT/public;
      location / {
        try_files $uri @proxy;
        access_log off;
        expires max;
      }

      location @proxy {
            proxy_set_header Host $http_host;
            proxy_set_header X-Forwarded-Host $host;
            proxy_set_header X-Real-IP $remote_addr;
            proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
            proxy_pass       http://backendurl;
      }

    }

You will need plackup to start a worker listening on a socket :

    cd YOUR_PROJECT_PATH
    sudo -u www plackup -E production -s Starman --workers=2 -l THE_PATH_OF_YOUR_PLACKUP_SOCKET_HERE.sock -a bin/app.pl

A good way to start this is to use C<daemontools> and place this line with
all environments variables in the "run" file.

=head3 Using HAProxy

C<HAProxy> is a reliable high-performance TCP/HTTP load balancer written in C available from
L<https://www.haproxy.org/>.

Suppose we want to run an application at C<app.example.com:80> and would to use two
backends listen on hosts C<app-be1.example.com:3000> and C<app-be2.example.com:3000>.

Here is HAProxy configuration file (haproxy.conf):

    global
        nbproc  1
        maxconn 4096
        user    nobody
        group   nobody
        # haproxy logs will be collected by syslog
        # syslog: unix socket path or tcp pair (ipaddress:port)
        log     /var/run/log local0
        daemon
        # enable compression (haproxy v1.5-dev13 and above required)
        tune.comp.maxlevel  5

    defaults
        log     global
        option  httpclose
        option  httplog
        option  dontlognull
        option  forwardfor
        option  abortonclose
        mode    http
        balance roundrobin
        retries 3
        timeout connect         5s
        timeout server          30s
        timeout client          30s
        timeout http-keep-alive 200m
        # enable compression (haproxy v1.5-dev13 and above required)
        compression algo gzip
        compression type text/html application/javascript text/css application/x-javascript text/javascript

    # application frontend (available at http://app.example.com)
    frontend app.example.com
        bind                  :80
        # modify request headers
        reqadd                X-Forwarded-Proto:\ http
        reqadd                X-Forwarded-Port:\ 80
        # modify response headers
        rspdel                ^Server:.*
        rspdel                ^X-Powered-By:.*
        rspadd                Server:\ Dethklok\ (Unix/0.2.3)
        rate-limit sessions   1024
        acl is-haproxy-stats  path_beg /stats
        # uncomment if you'd like to get haproxy usage statistics
        # use_backend haproxy   if is-haproxy-stats
        default_backend       dynamic

    # haproxy statistics (available at http://app.example.com/stats)
    backend haproxy
        stats uri             /stats
        stats refresh         180s
        stats realm           app.example.com\ haproxy\ statistics
        # change credentials
        stats auth            admin1:password1
        stats auth            admin2:password2
        stats hide-version
        stats show-legends

    # application backends
    backend dynamic
        # change path_info to check and value of the Host header sent to application server
        option httpchk HEAD / HTTP/1.1\r\nHost:\ app.example.com
        server app1 app-be1.example.com:3000 check inter 30s
        server app2 app-be2.example.com:3000 check inter 30s

We will need to start the workers on each backend of our application. This can be done by starman utility:

    # on app-be1.example.com
    $ starman --workers=2 --listen :3000 /path/to/app.pl
    # on app-be2.example.com
    $ starman --workers=2 --listen :3000 /path/to/app.pl

Then start the haproxy itself:

    # check the configuration..
    $ sudo haproxy -c -f haproxy.conf
    # now really start it..
    $ sudo haproxy -f haproxy.conf

=head2 Running on lighttpd

=head3 Running on lighttpd (CGI)

To run as a CGI app on lighttpd, just create a soft link to the C<dispatch.cgi>
script (created when you run C<dancer -a MyApp>) inside your system's C<cgi-bin>
folder. Make sure C<mod_cgi> is enabled.

    ln -s /path/to/MyApp/public/dispatch.cgi /usr/lib/cgi-bin/mycoolapp.cgi

=head3 Running on lighttpd (FastCGI)

Make sure C<mod_fcgi> is enabled. You also must have L<FCGI> installed.

This example configuration uses TCP/IP:

    $HTTP["url"] == "^/app" {
        fastcgi.server += (
            "/app" => (
                "" => (
                    "host" => "127.0.0.1",
                    "port" => "5000",
                    "check-local" => "disable",
                )
            )
        )
    }

Launch your application:

    plackup -s FCGI --port 5000 bin/app.psgi

This example configuration uses a socket:

    $HTTP["url"] =~ "^/app" {
        fastcgi.server += (
            "/app" => (
                "" => (
                    "socket" => "/tmp/fcgi.sock",
                    "check-local" => "disable",
                )
            )
        )
    }

Launch your application:

    plackup -s FCGI --listen /tmp/fcgi.sock bin/app.psgi

=head2 Performance Improvements

The following modules can be used to speed up an app in Dancer2:

=over 4

=item * L<CGI::Deurl::XS>

=item * L<Class::XSAccessor>

=item * L<Cpanel::JSON::XS>

=item * L<Crypt::URandom>

=item * L<HTTP::XSCookies>

=item * L<HTTP::XSHeaders>

=item * L<Math::Random::ISAAC::XS>

=item * L<MooX::TypeTiny>

=item * L<Type::Tiny::XS>

=item * L<URL::Encode::XS>

=item * L<YAML::XS>

=back

If you generated your application with C<dancer2 gen>, you can easily install
these with the following command:

    cpanm --installdeps . --with-feature=accelerate

To build them, you will need access to a C compiler, and using these modules
will prevent you from fatpacking your application.

These modules are installed by default when building a Docker container 
containing your application.

=head1 AUTHOR

Dancer Core Developers

=head1 COPYRIGHT AND LICENSE

This software is copyright (c) 2023 by Alexis Sukrieh.

This is free software; you can redistribute it and/or modify it under
the same terms as the Perl 5 programming language system itself.

=cut