Apache2::Reload - Reload Perl Modules when Changed on Disk

  # Monitor and reload all modules in %INC:
  # httpd.conf:
  PerlModule Apache2::Reload
  PerlInitHandler Apache2::Reload

  # when working with protocols and connection filters
  # PerlPreConnectionHandler Apache2::Reload

  # Reload groups of modules:
  # httpd.conf:
  PerlModule Apache2::Reload
  PerlInitHandler Apache2::Reload
  PerlSetVar ReloadAll Off
  PerlSetVar ReloadModules "ModPerl::* Apache2::*"
  #PerlSetVar ReloadDebug On

  # Reload a single module from within itself:
  package My::Apache2::Module;
  use Apache2::Reload;
  sub handler { ... }
  1;

CWApache2::Reload reloads modules that change on the disk.

When Perl pulls a file via CWrequire, it stores the filename in the global hash CW%INC. The next time Perl tries to CWrequire the same file, it sees the file in CW%INC and does not reload from disk. This module's handler can be configured to iterate over the modules in CW%INC and reload those that have changed on disk or only specific modules that have registered themselves with CWApache2::Reload. It can also do the check for modified modules, when a special touch-file has been modified.

Note that CWApache2::Reload operates on the current context of CW@INC. Which means, when called as a CWPerl*Handler it will not see CW@INC paths added or removed by CWModPerl::Registry scripts, as the value of CW@INC is saved on server startup and restored to that value after each request. In other words, if you want CWApache2::Reload to work with modules that live in custom CW@INC paths, you should modify CW@INC when the server is started. Besides, CW'use lib' in the startup script, you can also set the CWPERL5LIB variable in the httpd's environment to include any non-standard 'lib' directories that you choose. For example, to accomplish that you can include a line:

  PERL5LIB=/home/httpd/perl/extra; export PERL5LIB

in the script that starts Apache. Alternatively, you can set this environment variable in httpd.conf:

  PerlSetEnv PERL5LIB /home/httpd/perl/extra

  PerlModule Apache2::Reload
  PerlInitHandler Apache2::Reload

When working with connection filters and protocol modules CWApache2::Reload should be invoked in the pre_connection stage:

  PerlPreConnectionHandler Apache2::Reload

See also the discussion on CWPerlPreConnectionHandler.

To only reload modules that have registered with CWApache2::Reload, add the following to the httpd.conf:

  PerlModule Apache2::Reload
  PerlInitHandler Apache2::Reload
  PerlSetVar ReloadAll Off
  # ReloadAll defaults to On

Then any modules with the line:

  use Apache2::Reload;

Will be reloaded when they change.

You can also register modules explicitly in your httpd.conf file that you want to be reloaded on change:

  PerlModule Apache2::Reload
  PerlInitHandler Apache2::Reload
  PerlSetVar ReloadAll Off
  PerlSetVar ReloadModules "My::Foo My::Bar Foo::Bar::Test"

Note that these are split on whitespace, but the module list must be in quotes, otherwise Apache tries to parse the parameter list.

The CW* wild character can be used to register groups of files under the same namespace. For example the setting:

  PerlSetVar ReloadModules "ModPerl::* Apache2::*"

will monitor all modules under the namespaces CWModPerl:: and CWApache2::.

To reload modules only in certain directories (and their subdirectories) add the following to the httpd.conf:

  PerlModule Apache2::Reload
  PerlInitHandler Apache2::Reload
  PerlSetVar ReloadDirectories "/tmp/project1 /tmp/project2"

You can further narrow the list of modules to be reloaded from the chosen directories with CWReloadModules as in:

  PerlModule Apache2::Reload
  PerlInitHandler Apache2::Reload
  PerlSetVar ReloadDirectories "/tmp/project1 /tmp/project2"
  PerlSetVar ReloadAll Off
  PerlSetVar ReloadModules "MyApache2::*"

In this configuration example only modules from the namespace CWMyApache2:: found in the directories /tmp/project1/ and /tmp/project2/ (and their subdirectories) will be reloaded. You can also declare a file, which when gets CWtouch(1)ed, causes the reloads to be performed. For example if you set:

  PerlSetVar ReloadTouchFile /tmp/reload_modules

and don't CWtouch(1) the file /tmp/reload_modules, the reloads won't happen until you go to the command line and type:

  % touch /tmp/reload_modules

When you do that, the modules that have been changed, will be magically reloaded on the next request. This option works with any mode described before.

In some cases, it might be necessary to explicitely stop reloading a module.

  Apache2::Reload->unregister_module('Some::Module');

But be carefull, since unregistering a module in this way will only do so for the current interpreter. This feature should be used with care.

This modules is perfectly suited for a development environment. Though it's possible that you would like to use it in a production environment, since with CWApache2::Reload you don't have to restart the server in order to reload changed modules during software updates. Though this convenience comes at a price:

•

If the touch file feature is used, CWApache2::Reload has to stat(2) the touch file on each request, which adds a slight but most likely insignificant overhead to response times. Otherwise CWApache2::Reload will stat(2) each registered module or even worseall modules in CW%INC, which will significantly slow everything down.

•

Once the child process reloads the modules, the memory used by these modules is not shared with the parent process anymore. Therefore the memory consumption may grow significantly.

Therefore doing a full server stop and restart is probably a better solution.

If you aren't sure whether the modules that are supposed to be reloaded, are actually getting reloaded, turn the debug mode on:

  PerlSetVar ReloadDebug On

If you modify modules, which don't declare their CWpackage, and rely on CWApache2::Reload to reload them, you may encounter problems: i.e., it'll appear as if the module wasn't reloaded when in fact it was. This happens because when CWApache2::Reload CWrequire()s such a module all the global symbols end up in the CWApache2::Reload namespace! So the module does get reloaded and you see the compile time errors if there are any, but the symbols don't get imported to the right namespace. Therefore the old version of the code is running.

CWApache2::Reload uses CW%INC to find the files on the filesystem. If an entry for a certain filepath in CW%INC is relative, CWApache2::Reload will use CW@INC to try to resolve that relative path. Now remember that mod_perl freezes the value of CW@INC at the server startup, and you can modify it only for the duration of one request when you need to load some module which is not in on of the CW@INC directories. So a module gets loaded, and registered in CW%INC with a relative path. Now when CWApache2::Reload tries to find that module to check whether it has been modified, it can't find since its directory is not in CW@INC. So CWApache2::Reload will silently skip that module.

You can enable the CWDebug|/Debug mode to see what CWApache2::Reload does behind the scenes.

The following problem is relevant only to registry handlers that cache the compiled script. For example it concerns CWModPerl::Registry but not CWModPerl::PerlRun.

The Problem

Let's say that there is a module CWMy::Utils:

  #file:My/Utils.pm
  #----------------
  package My::Utils;
  BEGIN { warn __PACKAGE__ , " was reloaded\n" }
  use base qw(Exporter);
  @EXPORT = qw(colour);
  sub colour { "white" }
  1;

And a registry script test.pl:

  #file:test.pl
  #------------
  use My::Utils;
  print "Content-type: text/plain\n\n";
  print "the color is " . colour();

Assuming that the server is running in a single mode, we request the script for the first time and we get the response:

  the color is white

Now we change My/Utils.pm:

  -  sub colour { "white" }
  +  sub colour { "red" }

And issue the request again. CWApache2::Reload does its job and we can see that CWMy::Utils was reloaded (look in the error_log file). However the script still returns:

  the color is white

The Explanation

Even though My/Utils.pm was reloaded, CWModPerl::Registry's cached code won't run 'CWuse My::Utils;' again (since it happens only once, i.e. during the compile time). Therefore the script doesn't know that the subroutine reference has been changed.

This is easy to verify. Let's change the script to be:

  #file:test.pl
  #------------
  use My::Utils;
  print "Content-type: text/plain\n\n";
  my $sub_int = \&colour;
  my $sub_ext = \&My::Utils::colour;
  print "int $sub_int\n";
  print "ext $sub_ext\n";

Issue a request, you will see something similar to:

  int CODE(0)
  ext CODE(0)

As you can see both point to the same CODE reference (meaning that it's the same symbol). After modifying My/Utils.pm again:

  -  sub colour { "red" }
  +  sub colour { "blue" }

and calling the script on the secondnd time, we get:

  int CODE(0)
  ext CODE(0)

You can see that the internal CODE reference is not the same as the external one.

The Solution

There are two solutions to this problem:

Solution 1: replace CWuse() with an explicit CWrequire() + CWimport().

 - use My::Utils;
 + require My::Utils; My::Utils->import();

now the changed functions will be reimported on every request.

Solution 2: remember to touch the script itself every time you change the module that it requires.

If you use CWApache2::Reload with a threaded MPM and multiple Perl interpreters, the modules will be reloaded by each interpreter as they are used, not every interpreters at once. Similar to mod_perl 1.0 where each child has its own Perl interpreter, the modules are reloaded as each child is hit with a request.

If a module is loaded at startup, the syntax tree of each subroutine is shared between interpreters (big win), but each subroutine has its own padlist (where lexical my variables are stored). Once CWApache2::Reload reloads a module, this sharing goes away and each Perl interpreter will have its own copy of the syntax tree for the reloaded subroutines.

The short summary of this is: Don't use pseudo-hashes. They are deprecated since Perl 5.8 and are removed in 5.9.

Use an array with constant indexes. Its faster in the general case, its more guaranteed, and generally, it works.

The long summary is that some work has been done to get this module working with modules that use pseudo-hashes, but it's still broken in the case of a single module that contains multiple packages that all use pseudo-hashes.

So don't do that.

mod_perl 2.0 and its core modules are copyrighted under The Apache Software License, Version 2.0.

Matt Sergeant, matt@sergeant.org

Stas Bekman (porting to mod_perl 2.0)

A few concepts borrowed from CWStonehenge::Reload by Randal Schwartz and CWApache::StatINC (mod_perl 1.x) by Doug MacEachern and Ask Bjoern Hansen.