MediaWiki  1.23.15
RedisLockManager.php
Go to the documentation of this file.
1 <?php
42  protected $lockTypeMap = array(
43  self::LOCK_SH => self::LOCK_SH,
44  self::LOCK_UW => self::LOCK_SH,
45  self::LOCK_EX => self::LOCK_EX
46  );
47 
49  protected $redisPool;
50 
52  protected $lockServers = array();
53 
55  protected $session = '';
56 
67  public function __construct( array $config ) {
68  parent::__construct( $config );
69 
70  $this->lockServers = $config['lockServers'];
71  // Sanitize srvsByBucket config to prevent PHP errors
72  $this->srvsByBucket = array_filter( $config['srvsByBucket'], 'is_array' );
73  $this->srvsByBucket = array_values( $this->srvsByBucket ); // consecutive
74 
75  $config['redisConfig']['serializer'] = 'none';
76  $this->redisPool = RedisConnectionPool::singleton( $config['redisConfig'] );
77 
78  $this->session = wfRandomString( 32 );
79  }
80 
81  protected function getLocksOnServer( $lockSrv, array $pathsByType ) {
82  $status = Status::newGood();
83 
84  $server = $this->lockServers[$lockSrv];
85  $conn = $this->redisPool->getConnection( $server );
86  if ( !$conn ) {
87  foreach ( array_merge( array_values( $pathsByType ) ) as $path ) {
88  $status->fatal( 'lockmanager-fail-acquirelock', $path );
89  }
90 
91  return $status;
92  }
93 
94  $pathsByKey = array(); // (type:hash => path) map
95  foreach ( $pathsByType as $type => $paths ) {
96  $typeString = ( $type == LockManager::LOCK_SH ) ? 'SH' : 'EX';
97  foreach ( $paths as $path ) {
98  $pathsByKey[$this->recordKeyForPath( $path, $typeString )] = $path;
99  }
100  }
101 
102  try {
103  static $script =
104 <<<LUA
105  local failed = {}
106  -- Load input params (e.g. session, ttl, time of request)
107  local rSession, rTTL, rTime = unpack(ARGV)
108  -- Check that all the locks can be acquired
109  for i,requestKey in ipairs(KEYS) do
110  local _, _, rType, resourceKey = string.find(requestKey,"(%w+):(%w+)$")
111  local keyIsFree = true
112  local currentLocks = redis.call('hKeys',resourceKey)
113  for i,lockKey in ipairs(currentLocks) do
114  -- Get the type and session of this lock
115  local _, _, type, session = string.find(lockKey,"(%w+):(%w+)")
116  -- Check any locks that are not owned by this session
117  if session ~= rSession then
118  local lockExpiry = redis.call('hGet',resourceKey,lockKey)
119  if 1*lockExpiry < 1*rTime then
120  -- Lock is stale, so just prune it out
121  redis.call('hDel',resourceKey,lockKey)
122  elseif rType == 'EX' or type == 'EX' then
123  keyIsFree = false
124  break
125  end
126  end
127  end
128  if not keyIsFree then
129  failed[#failed+1] = requestKey
130  end
131  end
132  -- If all locks could be acquired, then do so
133  if #failed == 0 then
134  for i,requestKey in ipairs(KEYS) do
135  local _, _, rType, resourceKey = string.find(requestKey,"(%w+):(%w+)$")
136  redis.call('hSet',resourceKey,rType .. ':' .. rSession,rTime + rTTL)
137  -- In addition to invalidation logic, be sure to garbage collect
138  redis.call('expire',resourceKey,rTTL)
139  end
140  end
141  return failed
142 LUA;
143  $res = $conn->luaEval( $script,
144  array_merge(
145  array_keys( $pathsByKey ), // KEYS[0], KEYS[1],...,KEYS[N]
146  array(
147  $this->session, // ARGV[1]
148  $this->lockTTL, // ARGV[2]
149  time() // ARGV[3]
150  )
151  ),
152  count( $pathsByKey ) # number of first argument(s) that are keys
153  );
154  } catch ( RedisException $e ) {
155  $res = false;
156  $this->redisPool->handleError( $conn, $e );
157  }
158 
159  if ( $res === false ) {
160  foreach ( array_merge( array_values( $pathsByType ) ) as $path ) {
161  $status->fatal( 'lockmanager-fail-acquirelock', $path );
162  }
163  } else {
164  foreach ( $res as $key ) {
165  $status->fatal( 'lockmanager-fail-acquirelock', $pathsByKey[$key] );
166  }
167  }
168 
169  return $status;
170  }
171 
172  protected function freeLocksOnServer( $lockSrv, array $pathsByType ) {
173  $status = Status::newGood();
174 
175  $server = $this->lockServers[$lockSrv];
176  $conn = $this->redisPool->getConnection( $server );
177  if ( !$conn ) {
178  foreach ( array_merge( array_values( $pathsByType ) ) as $path ) {
179  $status->fatal( 'lockmanager-fail-releaselock', $path );
180  }
181 
182  return $status;
183  }
184 
185  $pathsByKey = array(); // (type:hash => path) map
186  foreach ( $pathsByType as $type => $paths ) {
187  $typeString = ( $type == LockManager::LOCK_SH ) ? 'SH' : 'EX';
188  foreach ( $paths as $path ) {
189  $pathsByKey[$this->recordKeyForPath( $path, $typeString )] = $path;
190  }
191  }
192 
193  try {
194  static $script =
195 <<<LUA
196  local failed = {}
197  -- Load input params (e.g. session)
198  local rSession = unpack(ARGV)
199  for i,requestKey in ipairs(KEYS) do
200  local _, _, rType, resourceKey = string.find(requestKey,"(%w+):(%w+)$")
201  local released = redis.call('hDel',resourceKey,rType .. ':' .. rSession)
202  if released > 0 then
203  -- Remove the whole structure if it is now empty
204  if redis.call('hLen',resourceKey) == 0 then
205  redis.call('del',resourceKey)
206  end
207  else
208  failed[#failed+1] = requestKey
209  end
210  end
211  return failed
212 LUA;
213  $res = $conn->luaEval( $script,
214  array_merge(
215  array_keys( $pathsByKey ), // KEYS[0], KEYS[1],...,KEYS[N]
216  array(
217  $this->session, // ARGV[1]
218  )
219  ),
220  count( $pathsByKey ) # number of first argument(s) that are keys
221  );
222  } catch ( RedisException $e ) {
223  $res = false;
224  $this->redisPool->handleError( $conn, $e );
225  }
226 
227  if ( $res === false ) {
228  foreach ( array_merge( array_values( $pathsByType ) ) as $path ) {
229  $status->fatal( 'lockmanager-fail-releaselock', $path );
230  }
231  } else {
232  foreach ( $res as $key ) {
233  $status->fatal( 'lockmanager-fail-releaselock', $pathsByKey[$key] );
234  }
235  }
236 
237  return $status;
238  }
239 
240  protected function releaseAllLocks() {
241  return Status::newGood(); // not supported
242  }
243 
244  protected function isServerUp( $lockSrv ) {
245  return (bool)$this->redisPool->getConnection( $this->lockServers[$lockSrv] );
246  }
247 
253  protected function recordKeyForPath( $path, $type ) {
254  return implode( ':',
255  array( __CLASS__, 'locks', "$type:" . $this->sha1Base36Absolute( $path ) ) );
256  }
257 
261  function __destruct() {
262  while ( count( $this->locksHeld ) ) {
263  $pathsByType = array();
264  foreach ( $this->locksHeld as $path => $locks ) {
265  foreach ( $locks as $type => $count ) {
266  $pathsByType[$type][] = $path;
267  }
268  }
269  $this->unlockByType( $pathsByType );
270  }
271  }
272 }
RedisLockManager\$lockTypeMap
array $lockTypeMap
Mapping of lock types to the type actually used *.
Definition: RedisLockManager.php:41
RedisLockManager\$lockServers
$lockServers
Definition: RedisLockManager.php:50
of
globals txt Globals are evil The original MediaWiki code relied on globals for processing context far too often MediaWiki development since then has been a story of slowly moving context out of global variables and into objects Storing processing context in object member variables allows those objects to be reused in a much more flexible way Consider the elegance of
Definition: globals.txt:10
RedisConnectionPool\singleton
static singleton(array $options)
Definition: RedisConnectionPool.php:115
php
skin txt MediaWiki includes four core it has been set as the default in MediaWiki since the replacing Monobook it had been been the default skin since before being replaced by Vector largely rewritten in while keeping its appearance Several legacy skins were removed in the as the burden of supporting them became too heavy to bear Those in etc for skin dependent CSS etc for skin dependent JavaScript These can also be customised on a per user by etc This feature has led to a wide variety of user styles becoming that gallery is a good place to ending in php
Definition: skin.txt:62
LockManager\LOCK_SH
const LOCK_SH
Lock types; stronger locks have higher values.
Definition: LockManager.php:58
is
We use the convention $dbr for read and $dbw for write to help you keep track of whether the database object is a the world will explode Or to be a subsequent write query which succeeded on the master may fail when replicated to the slave due to a unique key collision Replication on the slave will stop and it may take hours to repair the database and get it back online Setting read_only in my cnf on the slave will avoid this but given the dire we prefer to have as many checks as possible We provide a but the wrapper functions like please read the documentation for except in special pages derived from QueryPage It s a common pitfall for new developers to submit code containing SQL queries which examine huge numbers of rows Remember that COUNT * is(N), counting rows in atable is like counting beans in a bucket.------------------------------------------------------------------------ Replication------------------------------------------------------------------------The largest installation of MediaWiki, Wikimedia, uses a large set ofslave MySQL servers replicating writes made to a master MySQL server. Itis important to understand the issues associated with this setup if youwant to write code destined for Wikipedia.It 's often the case that the best algorithm to use for a given taskdepends on whether or not replication is in use. Due to our unabashedWikipedia-centrism, we often just use the replication-friendly version, but if you like, you can use wfGetLB() ->getServerCount() > 1 tocheck to see if replication is in use.===Lag===Lag primarily occurs when large write queries are sent to the master.Writes on the master are executed in parallel, but they are executed inserial when they are replicated to the slaves. The master writes thequery to the binlog when the transaction is committed. The slaves pollthe binlog and start executing the query as soon as it appears. They canservice reads while they are performing a write query, but will not readanything more from the binlog and thus will perform no more writes. Thismeans that if the write query runs for a long time, the slaves will lagbehind the master for the time it takes for the write query to complete.Lag can be exacerbated by high read load. MediaWiki 's load balancer willstop sending reads to a slave when it is lagged by more than 30 seconds.If the load ratios are set incorrectly, or if there is too much loadgenerally, this may lead to a slave permanently hovering around 30seconds lag.If all slaves are lagged by more than 30 seconds, MediaWiki will stopwriting to the database. All edits and other write operations will berefused, with an error returned to the user. This gives the slaves achance to catch up. Before we had this mechanism, the slaves wouldregularly lag by several minutes, making review of recent editsdifficult.In addition to this, MediaWiki attempts to ensure that the user seesevents occurring on the wiki in chronological order. A few seconds of lagcan be tolerated, as long as the user sees a consistent picture fromsubsequent requests. This is done by saving the master binlog positionin the session, and then at the start of each request, waiting for theslave to catch up to that position before doing any reads from it. Ifthis wait times out, reads are allowed anyway, but the request isconsidered to be in "lagged slave mode". Lagged slave mode can bechecked by calling wfGetLB() ->getLaggedSlaveMode(). The onlypractical consequence at present is a warning displayed in the pagefooter.===Lag avoidance===To avoid excessive lag, queries which write large numbers of rows shouldbe split up, generally to write one row at a time. Multi-row INSERT ...SELECT queries are the worst offenders should be avoided altogether.Instead do the select first and then the insert.===Working with lag===Despite our best efforts, it 's not practical to guarantee a low-lagenvironment. Lag will usually be less than one second, but mayoccasionally be up to 30 seconds. For scalability, it 's very importantto keep load on the master low, so simply sending all your queries tothe master is not the answer. So when you have a genuine need forup-to-date data, the following approach is advised:1) Do a quick query to the master for a sequence number or timestamp 2) Run the full query on the slave and check if it matches the data you gotfrom the master 3) If it doesn 't, run the full query on the masterTo avoid swamping the master every time the slaves lag, use of thisapproach should be kept to a minimum. In most cases you should just readfrom the slave and let the user deal with the delay.------------------------------------------------------------------------ Lock contention------------------------------------------------------------------------Due to the high write rate on Wikipedia(and some other wikis), MediaWiki developers need to be very careful to structure their writesto avoid long-lasting locks. By default, MediaWiki opens a transactionat the first query, and commits it before the output is sent. Locks willbe held from the time when the query is done until the commit. So youcan reduce lock time by doing as much processing as possible before youdo your write queries.Often this approach is not good enough, and it becomes necessary toenclose small groups of queries in their own transaction. Use thefollowing syntax:$dbw=wfGetDB(DB_MASTER
RedisLockManager\isServerUp
isServerUp( $lockSrv)
Check if a lock server is up.
Definition: RedisLockManager.php:241
RedisLockManager
Manage locks using redis servers.
Definition: RedisLockManager.php:40
RedisLockManager\releaseAllLocks
releaseAllLocks()
Release all locks that this session is holding.
Definition: RedisLockManager.php:237
Status\newGood
static newGood( $value=null)
Factory function for good results.
Definition: Status.php:77
RedisLockManager\recordKeyForPath
recordKeyForPath( $path, $type)
Definition: RedisLockManager.php:250
e
in this case you re responsible for computing and outputting the entire conflict i e
Definition: hooks.txt:1038
request
this hook is for auditing only etc create2 Corresponds to logging log_action database field and which is displayed in the UI similar to $comment this hook should only be used to add variables that depend on the current page request
Definition: hooks.txt:1644
QuorumLockManager
Version of LockManager that uses a quorum from peer servers for locks.
Definition: QuorumLockManager.php:31
LockManager\sha1Base36Absolute
sha1Base36Absolute( $path)
Get the base 36 SHA-1 of a string, padded to 31 digits.
Definition: LockManager.php:157
so
I won t presume to tell you how to I m just describing the methods I chose to use for myself If you do choose to follow these it will probably be easier for you to collaborate with others on the but if you want to contribute without by all means do so(and don 't be surprised if I reformat your code). - I have the code indented with tabs to save file size and so that users can set their tab stops to any depth they like. I use 4-space tab stops
RedisLockManager\freeLocksOnServer
freeLocksOnServer( $lockSrv, array $pathsByType)
Get a connection to a lock server and release locks on $paths.
Definition: RedisLockManager.php:169
RedisLockManager\__destruct
__destruct()
Make sure remaining locks get cleared for sanity.
Definition: RedisLockManager.php:258
array
the array() calling protocol came about after MediaWiki 1.4rc1.
List of Api Query prop modules.
LockManager\lock
lock(array $paths, $type=self::LOCK_EX, $timeout=0)
Lock the resources at the given abstract paths.
Definition: LockManager.php:90
LockManager\unlockByType
unlockByType(array $pathsByType)
Unlock the resources at the given abstract paths.
Definition: LockManager.php:140
RedisConnectionPool
Helper class to manage Redis connections.
Definition: RedisConnectionPool.php:38
RedisLockManager\$redisPool
RedisConnectionPool $redisPool
Definition: RedisLockManager.php:47
addition
We ve cleaned up the code here by removing clumps of infrequently used code and moving them off somewhere else It s much easier for someone working with this code to see what s _really_ going and make changes or fix bugs In addition
Definition: hooks.txt:86
$count
$count
Definition: UtfNormalTest2.php:96
it
=Architecture==Two class hierarchies are used to provide the functionality associated with the different content models:*Content interface(and AbstractContent base class) define functionality that acts on the concrete content of a page, and *ContentHandler base class provides functionality specific to a content model, but not acting on concrete content. The most important function of ContentHandler is to act as a factory for the appropriate implementation of Content. These Content objects are to be used by MediaWiki everywhere, instead of passing page content around as text. All manipulation and analysis of page content must be done via the appropriate methods of the Content object. For each content model, a subclass of ContentHandler has to be registered with $wgContentHandlers. The ContentHandler object for a given content model can be obtained using ContentHandler::getForModelID($id). Also Title, WikiPage and Revision now have getContentHandler() methods for convenience. ContentHandler objects are singletons that provide functionality specific to the content type, but not directly acting on the content of some page. ContentHandler::makeEmptyContent() and ContentHandler::unserializeContent() can be used to create a Content object of the appropriate type. However, it is recommended to instead use WikiPage::getContent() resp. Revision::getContent() to get a page 's content as a Content object. These two methods should be the ONLY way in which page content is accessed. Another important function of ContentHandler objects is to define custom action handlers for a content model, see ContentHandler::getActionOverrides(). This is similar to what WikiPage::getActionOverrides() was already doing.==Serialization==With the ContentHandler facility, page content no longer has to be text based. Objects implementing the Content interface are used to represent and handle the content internally. For storage and data exchange, each content model supports at least one serialization format via ContentHandler::serializeContent($content). The list of supported formats for a given content model can be accessed using ContentHandler::getSupportedFormats(). Content serialization formats are identified using MIME type like strings. The following formats are built in:*text/x-wiki - wikitext *text/javascript - for js pages *text/css - for css pages *text/plain - for future use, e.g. with plain text messages. *text/html - for future use, e.g. with plain html messages. *application/vnd.php.serialized - for future use with the api and for extensions *application/json - for future use with the api, and for use by extensions *application/xml - for future use with the api, and for use by extensions In PHP, use the corresponding CONTENT_FORMAT_XXX constant. Note that when using the API to access page content, especially action=edit, action=parse and action=query &prop=revisions, the model and format of the content should always be handled explicitly. Without that information, interpretation of the provided content is not reliable. The same applies to XML dumps generated via maintenance/dumpBackup.php or Special:Export. Also note that the API will provide encapsulated, serialized content - so if the API was called with format=json, and contentformat is also json(or rather, application/json), the page content is represented as a string containing an escaped json structure. Extensions that use JSON to serialize some types of page content may provide specialized API modules that allow access to that content in a more natural form.==Compatibility==The ContentHandler facility is introduced in a way that should allow all existing code to keep functioning at least for pages that contain wikitext or other text based content. However, a number of functions and hooks have been deprecated in favor of new versions that are aware of the page 's content model, and will now generate warnings when used. Most importantly, the following functions have been deprecated:*Revisions::getText() and Revisions::getRawText() is deprecated in favor Revisions::getContent() *WikiPage::getText() is deprecated in favor WikiPage::getContent() Also, the old Article::getContent()(which returns text) is superceded by Article::getContentObject(). However, both methods should be avoided since they do not provide clean access to the page 's actual content. For instance, they may return a system message for non-existing pages. Use WikiPage::getContent() instead. Code that relies on a textual representation of the page content should eventually be rewritten. However, ContentHandler::getContentText() provides a stop-gap that can be used to get text for a page. Its behavior is controlled by $wgContentHandlerTextFallback it
Definition: contenthandler.txt:107
type
This document describes the state of Postgres support in and is fairly well maintained The main code is very well while extensions are very hit and miss it is probably the most supported database after MySQL Much of the work in making MediaWiki database agnostic came about through the work of creating Postgres as and are nearing end of but without copying over all the usage comments General notes on the but these can almost always be programmed around *Although Postgres has a true BOOLEAN type
Definition: postgres.txt:22
in
Prior to maintenance scripts were a hodgepodge of code that had no cohesion or formal method of action Beginning in
Definition: maintenance.txt:1
are
The ContentHandler facility adds support for arbitrary content types on wiki instead of relying on wikitext for everything It was introduced in MediaWiki Each kind of and so on Built in content types are
Definition: contenthandler.txt:5
then
skin txt MediaWiki includes four core it has been set as the default in MediaWiki since the replacing Monobook it had been been the default skin since then
Definition: skin.txt:10
$path
$path
Definition: NoLocalSettings.php:35
as
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 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 and this often solves whatever problem the user is having It would be nice if this could such as
Definition: distributors.txt:9
that
deferred txt A few of the database updates required by various functions here can be deferred until after the result page is displayed to the user For updating the view updating the linked to tables after a etc PHP does not yet have any way to tell the server to actually return and disconnect while still running these but it might have such a feature in the future We handle these by creating a deferred update object and putting those objects on a global then executing the whole list after the page is displayed We don t do anything smart like collating updates to the same table or such because the list is almost always going to have just one item on if that
Definition: deferred.txt:11
$res
$res
Definition: database.txt:21
RedisLockManager\__construct
__construct(array $config)
Construct a new instance from configuration.
Definition: RedisLockManager.php:64
RedisLockManager\getLocksOnServer
getLocksOnServer( $lockSrv, array $pathsByType)
Get a connection to a lock server and acquire locks.
Definition: RedisLockManager.php:78
RedisLockManager\$session
string $session
random UUID *
Definition: RedisLockManager.php:52
$e
div flags Integer display flags(NO_ACTION_LINK, NO_EXTRA_USER_LINKS) 'LogException' returning false will NOT prevent logging $e
Definition: hooks.txt:1632
wfRandomString
wfRandomString( $length=32)
Get a random string containing a number of pseudo-random hex characters.
Definition: GlobalFunctions.php:347
$type
$type
Definition: testCompression.php:46