Make MessageCache use the immutable text cache during cache rebuilds

[mediawiki.git] / docs / distributors.txt

This document is intended to provide useful advice for parties seeking to
redistribute MediaWiki to end users. It's targeted particularly at maintainers
for Linux distributions, since it's been observed that distribution packages of
MediaWiki often break. We've consistently had to recommend that users seeking
support use official tarballs instead of their distribution's packages, and
this often solves whatever problem the user is having. It would be nice if
this could change.

== Background: why web applications are different ==

MediaWiki is intended to be usable on any web host that provides support for
PHP and a database. Many users of low-end shared hosting have very limited
access to their machine: often only FTP access to some subdirectory of the web
root. Support for these users entails several restrictions, such as:

  1) We cannot require installation of any files outside the web root. Few of
  our users have access to directories like /usr or /etc.
  2) We cannot require the ability to run any utility on the command line.
  Many shared hosts have exec() and similar PHP functions disabled.
  3) We cannot assume that the software has write access anywhere useful. The
  user account that MediaWiki (including its installer) runs under is often
  different from the account the user used to upload the files, and we might be
  restricted by PHP settings such as safe mode or open_basedir.
  4) We cannot assume that the software even has read access anywhere useful.
  Many shared hosts run all users' web applications under the same user, so
  they can't rely on Unix permissions, and must forbid reads to even standard
  directories like /tmp lest users read each others' files.
  5) We cannot assume that the user has the ability to install or run any
  programs not written as web-accessible PHP scripts.

Since anything that works on cheap shared hosting will work if you have shell
or root access too, MediaWiki's design is based around catering to the lowest
common denominator. Although we support higher-end setups as well (like
Wikipedia!), the way many things work by default is tailored toward shared
hosting. These defaults are unconventional from the point of view of normal
(non-web) applications -- they might conflict with distributors' policies, and
they certainly aren't ideal for someone who's installing MediaWiki as root.

== Directory structure ==

Because of constraint (1) above, MediaWiki does not conform to normal
Unix filesystem layout. Hopefully we'll offer direct support for standard
layouts in the future, but for now *any change to the location of files is
unsupported*. Moving things and leaving symlinks will *probably* not break
anything, but it is *strongly* advised not to try any more intrusive changes to
get MediaWiki to conform more closely to your filesystem hierarchy. Any such
attempt will almost certainly result in unnecessary bugs.

The standard recommended location to install MediaWiki, relative to the web
root, is /w (so, e.g., /var/www/w). Rewrite rules can then be used to enable
"pretty URLs" like /wiki/Article instead of /w/index.php?title=Article. (This
is the convention Wikipedia uses.)  In theory, it should be possible to enable
the appropriate rewrite rules by default, if you can reconfigure the web
server, but you'd need to alter LocalSettings.php too. See
<https://www.mediawiki.org/wiki/Manual:Short_URL> for details on short URLs.

If you really must mess around with the directory structure, note that the
following files *must* all be web-accessible for MediaWiki to function
correctly:

  * api.php, img_auth.php, index.php, load.php, opensearch_desc.php, thumb.php,
  profileinfo.php. These are the entry points for normal usage. This list may be
  incomplete and is subject to change.
  * mw-config/index.php: Used for web-based installation (sets up the database,
  prompts for the name of the wiki, etc.).
  * images/: Used for uploaded files. This could be somewhere else if
  $wgUploadDirectory and $wgUploadPath are changed appropriately.
  * skins/*/: Subdirectories of skins/ contain CSS and JavaScript files that
  must be accessible to web browsers. The PHP files and Skin.sample in skins/
  don't need to be accessible. This could be somewhere else if
  $wgStyleDirectory and $wgStylePath are changed appropriately.
  * extensions/: Many extensions include CSS and JavaScript files in their
  extensions directory, and will break if they aren't web-accessible. Some
  extensions might theoretically provide additional entry points as well, at
  least in principle.

But all files should keep their position relative to the web-visible
installation directory no matter what. If you must move includes/ somewhere in
/usr/share, provide a symlink from /var/www/w. If you don't, you *will* break
something. You have been warned.

== Configuration ==

MediaWiki is configured using LocalSettings.php. This is a PHP file that's
generated when the user visits mw-config/index.php to install the software, and
which the user can edit by hand thereafter. It's just a plain old PHP file,
and can contain any PHP statements. It usually sets global variables that are
used for configuration, and includes files used by any extensions.

Distributors can easily change the installer behavior, including LocalSettings
generated, by placing their overrides into mw-config/overrides directory. Doing
that is highly preferred to modifying MediaWiki code directly. See
mw-config/overrides/README for more details and examples.

There's a new maintenance/install.php script which could be used for performing
an install through the command line.

Some configuration options that distributors might be in a position to set
intelligently:

  * $wgEmergencyContact: An e-mail address that can be used to contact the wiki
  administrator. By default, "wikiadmin@ServerName".
  * $wgPasswordSender: The e-mail address to use when sending password e-mails.
  By default, "MediaWiki Mail <apache@ServerName>".
        (with ServerName guessed from the http request)
  * $wgSMTP: Can be configured to use SMTP for mail sending instead of PHP
  mail().

== Updates ==

The correct way for updating a wiki is to update the files and then run from 
command line the maintenance/update.php script (with appropriate parameters if
files were moved). It will perform all the needed steps to update the database 
schema and contents to the version from whatever old one it has.
Any package manager which replaces the files but doesn't update the db is leaving
an inconsistent wiki that may produce blank pages (php errors) when new features 
using the changed schema would be used.

Since MediaWiki 1.17 it is possible to upgrade using the web installer by providing
an arbitrary secret value stored as $wgUpgradeKey in LocalSettings (older versions 
needed to rename LocalSettings.php in order to upgrade using the installer).

== Documentation ==

MediaWiki's official documentation is split between two places: the source
code, and <https://www.mediawiki.org/>. The source code documentation is written
exclusively by developers, and so is likely to be reliable (at worst,
outdated). However, it can be pretty sparse. mediawiki.org documentation is
often much more thorough, but it's maintained by a wiki that's open to
anonymous edits, so its quality is sometimes sketchy -- don't assume that
anything there is officially endorsed!

== Upstream ==

MediaWiki is a project hosted and led by the Wikimedia Foundation, the
not-for-profit charity that operates Wikipedia. Wikimedia employs the lead
developer and several other paid developers, but commit access is given out
liberally and there are multiple very active volunteer developers as well. A
list of developers can be found at <https://www.mediawiki.org/wiki/Developers>.

MediaWiki's bug tracker is at <https://phabricator.wikimedia.org>. However, you
might find that the best place to post if you want to get developers' attention
is the wikitech-l mailing list
<https://lists.wikimedia.org/mailman/listinfo/wikitech-l>. Posts to wikitech-l
will inevitably be read by multiple experienced MediaWiki developers. There's
also an active IRC chat at <irc://irc.freenode.net/mediawiki>, where there are
usually several developers at reasonably busy times of day.

Our Git repositories are hosted at <https://gerrit.wikimedia.org>, see
<https://www.mediawiki.org/wiki/Gerrit> for more information. Patches should
be submitted there. If you know which developers are best suited to review your
patch, add them to it, otherwise ask on IRC to get better review time.

All redistributors of MediaWiki should be subscribed to mediawiki-announce
<https://lists.wikimedia.org/mailman/listinfo/mediawiki-announce>. It's
extremely low-traffic, with an average of less than one post per month. All
new releases are announced here, including critical security updates.

== Useful software to install ==

There are several other pieces of software that MediaWiki can make good use of.
Distributors might choose to install these automatically with MediaWiki and
perhaps configure it to use them (see Configuration section of this document):

  * APC (Alternative PHP Cache), XCache, or similar: Will greatly speed up the
  execution of MediaWiki, and all other PHP applications, at some cost in
  memory usage. Will be used automatically for the most part.
  * clamav: Can be used for virus scanning of uploaded files. Enable with
  "$wgAntivirus = 'clamav';".
  * DjVuLibre: Allows processing of DjVu files. To enable this, set
  "$wgDjvuDump = 'djvudump'; $wgDjvuRenderer = 'ddjvu'; $wgDjvuTxt = 'djvutxt';".
  * HTML Tidy: Fixes errors in HTML at runtime. Can be enabled with 
        "$wgUseTidy = true;".
  * ImageMagick: For resizing images. "$wgUseImageMagick = true;" will enable
  it. PHP's GD can also be used, but ImageMagick is preferable.
  * HTTP cache such as Varnish or Squid: can provide a drastic speedup and a
  major cut in resource consumption, but enabling it may interfere with other
  applications. It might be suitable for a separate package. For setup details, see:
  - <https://www.mediawiki.org/wiki/Manual:Varnish_caching>
  - <https://www.mediawiki.org/wiki/Manual:Squid_caching>
  * rsvg or other SVG rasterizer: ImageMagick can be used for SVG support, but
  is not ideal. Wikipedia (as of the time of this writing) uses rsvg. To
  enable, set "$wgSVGConverter = 'rsvg';" (or other as appropriate).

MediaWiki uses some standard GNU utilities as well, such as diff and diff3. If
these are present in /usr/bin or some other reasonable location, they will be
configured automatically on install.

MediaWiki also has a "job queue" that handles background processing. Because
shared hosts often don't provide access to cron, the job queue is run on every
page view by default. This means the background tasks aren't really done in
the background. Busy wikis can set $wgJobRunRate to 0 and run
maintenance/runJobs.php periodically out of cron. Distributors probably
shouldn't set this up as a default, however, since the extra cron job is
unnecessary overhead for a little-used wiki.

== Web server configuration ==

MediaWiki includes several .htaccess files to restrict access to some
directories. If the web server is not configured to support these files, and
the relevant directories haven't been moved someplace inaccessible anyway (e.g.
symlinked in /usr/share with the web server configured to not follow symlinks),
then it might be useful to deny web access to those directories in the web
server's configuration.