NAME

MasonX::Apache2Handler - experimental (beta) Mason/mod_perl2 interface

SYNOPSIS

   use MasonX::Apache2Handler;

   my $ah = MasonX::Apache2Handler->new (..name/value params..);
   ...
   sub handler {
      my $r = shift;
      $ah->handle_request($r);
   }

DESCRIPTION

MasonX::Apache2Handler is experimental ( beta ) and should only be used in a test environment. The MasonX::Apache2Handler module itself is a simple clone of Mason's HTML::Mason::ApacheHandler, and, as such, seems to be solid. However the components used, mod_perl 2 and libapreq2, are currently under development each give a version of the above warning.

If you are interested in moving to mod_perl 2 when it reaches production status, I encourage you to try these modules in a test environment; your feedback will help getting mod_perl 2 out the door. I have been using this module on two live sites since mid-January 2004, with no problems.

MasonX::Apache2Handler was written to allow Mason to run in a 'pure' mod_perl2/Apache2 environment using the mod_perl2 native request structure as implemented by libapreq2. This is an unofficial release, not supported by the Mason group. If you want to use this module in a testing environment, please address problems, issues, comments, and improvements to me, not Mason.

When deciding to port Mason to mod_perl2 I took the approach to add Module(s) rather than patching core Mason modules. Upon investigation I discovered that all the Apache 'glue' code in Mason was contained in HTML::Mason::ApacheHandler. Therefore, I renamed and modified that module to create MasonX::Apache2Handler.

The actual changes I made can be found in the distribution in diff/ApacheHandler.diff ( made with 'diff -Naru' ... ).

As a result of my approach, you may install the normal Mason ( >= 1.25 ), the modules in "PREREQUISITES" below, and this module. After configuring ( see "CONFIGURATION" below ), you're ready to play.

The Apache2Handler object links Mason to mod_perl2, running components in response to HTTP requests. It is controlled primarily through parameters to the new() constructor.

handle_request() is not a user method, but rather is called from the HTML::Mason::handler() routine in handler.pl.

MasonX::Apache2Handler is a modified copy of the standard HTML::Mason::ApacheHandler. Apache2Handler MUST be used with mod_perl2.

You may, however, run Mason with Apache2/mod_perl2 without Apache2Handler (see my rather dated mini-HOWTO at http://beaucox.com/mason/mason-with-apmp2-mini-HOWTO.htm), but you then use use 'CGI' requests rather than the native 'mod_perl' requests.

The interface is the same as ApacheHandler's, Please refer to HTML::Mason, HTML::Mason::ApacheHandler, and http://masonhq.com/docs/manual/Admin.html.

NAME CHANGE

I have changed module names and released them as of version 0.04 as perl Mason's request:

HTML::Mason::ApacheHandler2 =>
    MasonX::Apache2Handler

MasonX::Request::WithApacheSession2 =>
    MasonX::Request::WithApache2Session

These changes keep my modules out of the 'core' Mason name space. Those of you who have installed my prior modules are encouraged to install the new modules and make the necessary changes to the configuration files and scripts.

'$r' IN 0.05

The '$r' passed to Mason pages has been changed from type 'Apache::Request' to 'Apache::RequestRec' to conform to the parameter passed to mod_perl 2 handlers in version 0.05. This WILL break your Mason pages that currently use '$r'. To get the corresponding 'Apache::Request' in your pages under 0.05, do:

$rr = Apache::Request->new( $r );

$rr' is now an 'Apache::Request', while '$r' is an 'Apache::RequestRec'.

There is a web page in eg/request-examiner.html that displays the type and interface provided by '$r' in your Mason pages. You may want to try this page on your system. Further information may be found on the mod perl site, specifically: http://perl.apache.org/docs/2.0/user/porting/compat.html#Request_Object_Methods and http://perl.apache.org/docs/2.0/api/Apache/RequestRec.html.

PREREQUISITES

You must have the following packages installed:

perl            => 5.8.0
mod_perl        => 1.9910
HTML::Mason     => 1.25
libapreq2       => 2.02-dev

Please refer to the packages' documentation for instructions.

WARNING: PERL 5.8.2

If you are using perl 5.8.2 you may get a series of errors in the http/error_log such as:

  Attempt to free unreferenced scalar: SV 0x405e6e78
   at /usr/lib/perl5/site_perl/5.8.2/HTML/Mason/Request.pm line 160.
  ...
  [Fri Jan 30 09:41:58 2004] [error] [client 207.175.219.202]
   Attempt to free unreferenced scalar: SV 0x405e6e78
   at /usr/lib/perl5/site_perl/5.8.2/i686-linux-thread-multi/
   Apache/Cookie.pm line 67.

  Stack:
    [/usr/lib/perl5/site_perl/5.8.2/MasonX/Apache2Handler.pm:892]
    [/usr/lib/perl5/site_perl/5.8.2/MasonX/Apache2Handler.pm:801]
    [/srv/www/perl/MyApache/Mason/Apache2Handler.pm:86]
    [-e:0]

which may return a 500 Internal Server Error to the user.

There was a bug introduced in perl 5.8.2 and fixed in 5.8.3, which affects some XS modules running under a threaded mpm mod_perl 2.0 (or any ithreads perl application). The affected modules can be fixed, to avoid this problem, by using the PERL_NO_GET_CONTEXT macro in the XS code (see the perlguts manpage for more information). So you need to check whether a newer version of the module is available. If not you need to upgrade to perl 5.8.3 or higher and the problem will go away.

When I had these problems, I ended up upgrading to 5.8.3 and recompiling EVERY CPAN module I use and remaking mod_perl2. You may have to do the same thing.

Perl 5.8.0 and 5.8.1 operate correctly, the problem is manifested in 5.8.2.

FLUSH BUFFER AND REDIRECTS

I have 'subclassed' HTML::Mason::Request (as MasonX::Request2) and HTML::Mason::Buffer (as MasonX::Buffer2) to change handling of buffer flushing as per Johnathan Kupferer's suggestion (see http://marc.theaimsgroup.com/?l=mason-devel&m=105337019427296&w=2). Basically the buffer is flushed only if there is really anything to flush, which clears the way for $m->redirect( $url );.

I have tested this change extensively and found it to be solid. You may wish to uncomment the "print STDERR ..." statements in Burrer2 and Request2 if you wish to see the flushing process at work (or you still have redirect problems). The print STDERR's appear in the Apache log.

CONFIGURATION

Configuring your Mason system may be done in many different ways; please refer to the Mason documentation, specifically, the Administrator's manual at http://masonhq.com/docs/manual/Admin.html.

This section gives several sample configurations to get you started and notes the special configuration parameters that are required for this module's operation with the native mod_perl2 interface.

All of the sample configuration files and scripts below can be found in the eg/ subdirectory of this distribution.

All Configuration in the 'httpd.conf' File

The sample httpd.conf file may be found at eg/httpd-mason-simple.conf. You must, of course, change all path names below to suit your particular installation.

First, load the necessary mod_perl2 modules:

LoadModule perl_module /usr/apache2/lib/apache/mod_perl.so
LoadModule apreq_module /usr/apache2/lib/apache/mod_apreq.so
LoadModule cgid_module /usr/apache2/lib/apache/mod_cgid.so

mod_perl2 is loaded as in the past. libapreg2 - the new mod_perl2-native request and cookie interface is loaded next. cgid is the optional CGI daemon module.

Next, load the modules your system requires:

PerlModule Apache2
PerlSwitches -I/usr/local/test/httpd/perl
PerlModule Apache::Request
PerlModule Apache::Cookie
...

'PerlModule' is the configuration file syntax for perl's 'use'; trim this list to suit your system (and conserve memory). The 'PerlSwitches' directive prepends the specified directory to perl's @INC array;

Setup the perl directory for your site:

Alias /perl/ /usr/local/test/httpd/perl/
<Location /perl/>
  SetHandler perl-script
  PerlResponseHandler ModPerl::Registry
  PerlOptions +ParseHeaders
  Options +ExecCGI
</Location>

Now, configure Mason. First add the following perl variables which are required for the operation of Apache2Handler:

PerlSetVar _MasonUser wwwrun
PerlSetVar _MasonGroup nogroup
PerlSetVar _MasonDefaultDocumentRoot "/usr/local/test/httpd/htdocs"

'_MasonUser' and '_MasonGroup' specify the user/group under which Apache runs; they must be the same as specified int the normal 'User' and 'Croup' directives earlier in your configuration file (this change was made necessary because $s->uid and $s->gid are not supported in mod_perl2 - see http://perl.apache.org/docs/2.0/user/porting/compat.html#C__s_E_gt_uid_).

'_MasonDefaultDocumentRoot' is required because the configuration directives are not available during server startup in mod_perl2. It should specify the 'DocumentRoot' directory specified earlier in your configuration file.

You may (optionally) pre-load any modules your Mason modules may require here - see http://masonhq.com/docs/manual/Admin.html#external_modules. The '<Perl>...</Perl>' directives are used because this modules must be loaded within the HTML::Mason::Commands name space.

<Perl>
{
    package HTML::Mason::Commands;
    use Apache::Const -compile => ':common';
    ...
}
</Perl>

Finally, here is an example of a virtual host:

   Listen 12984
   NameVirtualHost localhost:12984

   # for general testing - default - on localhost

   # site: bctest
   <VirtualHost localhost:12984>
     ServerName bctest.beaucox.com
     DocumentRoot "/usr/local/test/httpd/htdocs/bctest"
     PerlSetVar MasonCompRoot /usr/local/test/httpd/htdocs/bctest
     PerlSetVar MasonDataDir  /usr/local/test/httpd/mason/bctest
     PerlSetVar MasonRequestClass          MasonX::Request::WithApache2Session
     PerlSetVar MasonSessionAllowInvalidId yes
     PerlSetVar MasonSessionCookieName     beaucox-bctest-cookie
     PerlSetVar MasonSessionCookieDomain   beaucox.com
     PerlSetVar MasonSessionCookieExpires  +7d
     PerlSetVar MasonSessionClass          Apache::Session::MySQL
     PerlSetVar MasonSessionDataSource     dbi:mysql:bctest_sessions
     PerlSetVar MasonSessionUserName       mysql
     PerlSetVar MasonSessionPassword       mysql
     PerlSetVar MasonSessionLockDataSource dbi:mysql:bctest_sessions
     PerlSetVar MasonSessionLockUserName   mysql
     PerlSetVar MasonSessionLockPassword   mysql
     PerlSetVar MasonSessionUseCookie      yes
     <FilesMatch "^_">
       SetHandler perl-script
       PerlResponseHandler MasonX::Apache2Handler
    </FilesMatch>
    <Directory "/usr/local/test/httpd/htdocs/bctest">
       <FilesMatch "\.html$|\.htm$">
         SetHandler perl-script
         PerlResponseHandler MasonX::Apache2Handler
       </FilesMatch>
     </Directory>
   </VirtualHost>

In this example, the virtual host is using MasonX::Request::WithApache2Session (note the '2') - hence the 'MasonSession...' directives. Either modify them for your own use ( see MasonX::Request::WithApache2Session and MasonX::Request::WithApacheSession or omit them if you are not using the session module. For mod_perl2 compatibility, 'PerlResponseHandler' is used instead of 'PerlHandler' - see http://perl.apache.org/docs/2.0/user/porting/compat.html#C_PerlHandler_.

Configuration with Scripts

In this example, you must manage the http.conf file and two perl scripts; you gain superior flexibility in exchange for a little elbow grease.

httpd.conf

Here are the relevant sections of the httpd.conf file (the complete configuration file may be found at eg/httpd-mason.conf):

LoadModule perl_module /usr/apache2/lib/apache/mod_perl.so
LoadModule apreq_module /usr/apache2/lib/apache/mod_apreq.so
LoadModule cgid_module /usr/apache2/lib/apache/mod_cgid.so

Nothing new here, same as the configuration-file-only example above;

PerlSetEnv MOD_PERL_INC	"/usr/local/test/httpd/perl"
PerlRequire "/usr/local/test/httpd/conf/startup2.pl"

The 'PerlSetEnv' directive is used so the startup scripts can be written without any 'hard' path dependences. The 'PerlRequire' runs the actual startup script whose name and location are of your choosing.

Alias /perl/ /usr/local/test/httpd/perl/
<Location /perl/>
  SetHandler perl-script
  PerlResponseHandler ModPerl::Registry
  PerlOptions +ParseHeaders
  Options +ExecCGI
</Location>

Setup you perl directory as shown in the previous section.

Now to Mason:

PerlSetVar _MasonUser wwwrun
PerlSetVar _MasonGroup nogroup
PerlSetVar _MasonDefaultDocumentRoot "/usr/local/test/httpd/htdocs"

The same as described in the previous section.

PerlSetEnv MASON_COMP_ROOT	"/usr/local/test/httpd/htdocs"
PerlSetEnv MASON_DATA_ROOT	"/usr/local/test/httpd/mason"
PerlSetEnv MASON_SITES		"bctest:masontest"

We will see below how thews environment variables are used int the scripts below.

Finally, here is a sample virtual host:

   Listen 12984
   NameVirtualHost localhost:12984

   # for general testing - default - on localhost

   <VirtualHost localhost:12984>
     ServerName bctest.beaucox.com
     DocumentRoot "/usr/local/test/httpd/htdocs/bctest"
     PerlSetVar mason_site 'bctest'
     <FilesMatch "^_">
       SetHandler perl-script
       PerlResponseHandler MyApache::Apache2Handler
     </FilesMatch>
     <Directory "/usr/local/test/httpd/htdocs/bctest">
       <FilesMatch "\.html$|\.htm$">
         SetHandler perl-script
         PerlResponseHandler MyApache::Apache2Handler
       </FilesMatch>
     </Directory>
   </VirtualHost>

Much like the virtual host described in the previous section, but much of the 'guts' are now filled in by the handler script below. Remember to specify 'PerlResponseHandler'.

startup2.pl

Here is the sample 'startup2.pl' script (found at eq/startup2.pl):

   use Apache2 ();
   use lib ( $ENV{MOD_PERL_INC} );
   
   use Apache::Request ();
   use Apache::Cookie ();
   use CGI ();
   use CGI::Cookie ();
   ...
   use MyApache::Apache2Handler ();

   1;

Again, the modules you require are pre-loaded ('use'), and the the perl @INC array is adjusted. See how the use of the environment variable 'MOD_PERL_INC' - set in the httpd.conf - allows this script to be path-independent.

If you execute ('use') your Handler script here, the Apache2Handler request objects are pre-loaded; otherwise they are loaded 'on the fly'. Refer the the discussion at http://masonhq.com/docs/manual/Admin.html#wrappers_with_virtual_hosts.

Apache2Handler.pm

Here is the sample 'MyApache::Apache2Handler.pm' script. The full sample script may be found at eq/Apache2Handler.pm. It should be installed under your-perl-directory/Apache2Handler on your system; i.e:, in my case at: /usr/local/test/httpd/perl/Apache2Handler.pm.

   #!/usr/bin/perl

   package MyApache::Apache2Handler;

   use strict;
   use warnings;

Pretty standard perl startup stuff.

   use Apache2 ();
   use lib ( $ENV{MOD_PERL_INC} );

   use Apache::Request ();
   use Apache::Cookie ();
   use CGI ();
   use CGI::Cookie ();

Includes you may need;

our %ah = ();

This is a global hash that will hold, one for each site, Apache2Handler's.

   #	Mason w/Apache support
   use MasonX::Apache2Handler;

   #	Modules my components will use
   {
       package HTML::Mason::Commands;
	
       use Apache::Const -compile => ':common';
       ...
   }

Any includes you may want to pre-load for your Mason components.

setup_sites();

This line, if present, will pre-load all the Apache2Handler's (one for each site) at server startup time.

Now to handle the request:

   #  actual request handler
   sub handler
   {
     my ($r) = @_;

     # DON'T allow internal components (starting with '_')
     my $fn = $r->filename;
     if ($fn =~ m{.*/(.*)} && $1 && $1 =~ /^_/) {
       my $rip = $r->connection->remote_ip;
       $r->log_error ("attempt to access internal component: $fn remote ip: $rip\n");
       return Apache::NOT_FOUND;
     }

A check to prevent outside direct access to internal Mason components - in my system, components that start with '_'.

# allow only text/xxx content type
  return -1 if $r->content_type && $r->content_type !~ m|^text/|i;

Skip Mason processing for non-text items (images, binary downloads, etc.)

     # find site and handler: dispatch request
     my $site = $r->dir_config ('mason_site');

     unless( $site ) {
        $r->log_error ("no 'mason_site' specified\n");
        return Apache::NOT_FOUND;
     }

If there is no site configured with 'PserSetVar mason_site xxx', you have boo-boo-ed and the request is logged and rejected. You could force a more noticeable alert, i.e. an email, if you really want to know when this happens, but you really should be able to prevent these error with adequate testing.

unless( $ah{$site} ) {
  setup_sites( $r, $site );
  unless( $ah{$site} ) {
    $r->log_error ("no 'ah' found for 'mason_site' $site\n");
    return Apache::NOT_FOUND;
  }
}

Here we check the the Apache2Handler is loaded, and load it if not; Of that does not work, you've got problems.

my $status = $ah{$site}->handle_request ($r);

Finally! The request is sent on it's way.

# special error handling here (email, etc...)

You could check the status here and do extra fancy error reporting here...

  $status;
}

Return the status and exit.

Now, here is where the Apache2Handler requests are loaded, either at startup time or on the fly.

   # set up an Apache2Handler for each site
   sub setup_sites
   {
     my ( $r, $site ) = shift;
     my @asites = ();
     if( $site ) {
       push @asites, $site;
     } else {
       my $sites = $ENV{MASON_SITES};
       return unless $sites;
       @asites = split /:/, $sites;
     }
     for my $site( @asites ) {
       next if $ah{$site};
       my @args =
       (
         args_method   => "mod_perl",
         comp_root     => $ENV{MASON_COMP_ROOT}."/$site",
         data_dir      => $ENV{MASON_DATA_ROOT}."/$site",
         error_mode    => 'output',
         request_class =>'MasonX::Request::WithApache2Session',
         session_allow_invalid_id => 'yes',
         ...
       );
       push @args, $r if $r;
       $ah{$site} = new MasonX::Apache2Handler( @args );
     }
   }

   1;

If your sites Apache2Handlers are being setup 'on-the-fly', this method is called as 'setup_sites( $r, $site );'. Only that site is loaded.

On the other hand, if the sites are all loaded at server start as follows:

The 'MASON_SITES' environment variable, set in the httpd.conf file, consists of a list of site names separated by ':'s. This trick is used so the sites served may be changed in one place, the httpd.conf file, without having to update this script too

Note the use of the native mod_perl2 args_method: 'mod_perl'. Again, the environment variables set int the httpd.conf file are used her to keep this script path-independent.

This example is using the session subclass 'MasonX::With::Apache2Session'; modify or omit these statements.

STRESS TESTING

To see if your server works under load, you must do some stress testing. There are several Apache Test modules on CPAN, but if you are lazy ( like me ), you may try my simple test scripts.

stress.pl

A stress script is in the MasonX::Apache2Handler distribution at scripts/stress.pl. This simple perl script, which requires LWP::UserAgent in libwww, repeatedly gets a uri on you server and checks the result. Usage:

perl stress.pl <uri-to-a-page-on-your-test-server> [repeat-count]

If the repeat count is missing, the test is endless ( stop it with ^C ).

httpd-mem,pl

To check for memory leaks, try scripts/httpd-mem.pl. This script finds all the processes running for your server and totals the memory usage using the '/proc/<pid>/status' pseudo-files. This script will only work on systems with the GNU-Linux /proc file system.

Usage:

perl httpd-mem.pl [id-for-ps]

Where [id-for-ps] is a string to select your test server pids from the ps aux command. The default is 'httpd -k'.

Every two seconds a line is printed to the terminal and httpd-mem.log:

 VmData      VmExe      VmLck      VmLib      VmRSS     VmSize      VmStk
4297576      32248          0     524944    1479132    4904200       2784
4297576      32248          0     524944    1479132    4904200       2784
4297576      32248          0     524944    1479132    4904200       2784
...

Check this output when running the stress test above to see if anything ( especially VmSize ) is growing; that _may_ indicate a memory leak.

mod_perl2 ALL THE WAY

If you want to take the next step, making a pure mod_perl2 site, you should:

remake and install mod_perl

Disable global mod_perl backward compatibility by adding the 'MP_COMPAT_1X=0' flag to 'Makefile.PL':

perl Makefile.PL MP_APXS=/where/ever MP_COMPAT_1X=0

This flag is ON by default.

grep your site for Apache::compat

Remove 'use Apache::compat' from all of your mod_perl modules; you may have to rework them to bring them up to speed. Stas and the guys at mod_perl have several excellent 1.x => 2.x porting documents, my personal favorite being http://perl.apache.org/docs/2.0/user/porting/compat.html.

update your http.conf file

Once you have removed 1.x backward compatibility, you must bring your http.conf directives up to mod_perl2 standards as shown in http://perl.apache.org/docs/2.0/user/porting/compat.html#Configuration_Files_Porting.

Some of the changes you will have to make are:

   PerlHandler        => PerlResponseHandler.
 
   PerlSendHeader On  => PerlOptions +ParseHeaders
   PerlSendHeader Off => PerlOptions -ParseHeaders

   PerlSetupEnv On    => PerlOptions +SetupEnv
   PerlSetupEnv Off   => PerlOptions -SetupEnv

   PerlTaintCheck     => PerlSwitches -T
   PerlWarn           => PerlSwitches -w

   PerlFreshRestart   => is a mod_perl 1.0 legacy => see docs.

I found this to be a snap, but then I started coding with mod_perl2; I suppose the port could be a bear if you have a mature site with lots of 1.x modules.

DON'T

Mix and Match normal MasonX::... modules with MasonX::Apache2Handler

Any MasonX... modules that use ( subclass ) HTML::Mason::ApacheHandler will NOT work in your pure mod_perl2 environment. Let me know ( or change them yourself ) when you want to use one I have not changed.

Bother the Mason developers with questions, etc.

MasonX::Apache2Handler is unofficial and was written and is supported by me, not the Mason developers. Talk to me ( <mason@beaucox.com> ).

DO

Try it!

Setup a test server and see if the pure mod_perl2 Mason works for you.

Tell me what you think

Let me know your reaction to this effort. I welcome comments, suggestions, bug reports, and, yes, even mild flames.

TODO

Build tests

Currently, there are no 'real' tests defined in 'make test'. I plan to design and build some. Until then, the testing is left to you. Sorry :)

Continue to monitor my web site for problems

I am running this module at my web site ( http://beaucox.com ). I am continually monitoring the site logs and memory usage to catch and correct any bugs I find.

Investigate Other MasonX:: modules that may have to be converted

The MasonX:: modules that currently tie to ApacheHandler must be reworked to operate with Apache2Handler; I will attack those on demand. Please let me know.

A COMPLETELY OT COMMENT ON 'MasonSessionCookieDomain'

As an aside, I have found that some browsers, mainly Mozilla, cannot cope with the leading 'dot' in the '-domain' parameter sent in the 'Set-Cookie' header (you specify the '-domain' in the 'MasonSessionCookieDomain' configuration directive or the 'session_cookie_domain' parameter in your handler). I suggest you leave it out, i.e. change:

.beaucox.com

to:

beaucox.com

This makes Mozilla happy and works on the other browsers I have tested (IE, Konqueror).

BUGS

Too early to tell; they are bound to come in as people give it a try.

AUTHOR

Beau E. Cox <mason@beaucox.com> http://beaucox.com.

The real authors (I just made mod_perl2 changes) are the Mason crew, including: Jonathan Swartz <swartz@pobox.com>, Dave Rolsky <autarch@urth.org>, Ken Williams <ken@mathforum.org>.

Version 0.05 as of April 13, 2004.

SEE ALSO

My documents, including: MasonX::Apache2Handler, MasonX::Request::WithApache2Session, MasonX::Request::WithMulti2Session,

Original Mason documents, including: HTML::Mason::ApacheHandler, MasonX::Request::WithApacheSession, MasonX::Request::WithMultiSession.

Also see the Mason documentation at http://masonhq.com/docs/manual/.