includes/job/JobQueue.php

   1 <?php
   2 /**
   3  * Job queue base code.
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License as published by
   7  * the Free Software Foundation; either version 2 of the License, or
   8  * (at your option) any later version.
   9  *
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License along
  16  * with this program; if not, write to the Free Software Foundation, Inc.,
  17  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
  18  * http://www.gnu.org/copyleft/gpl.html
  19  *
  20  * @file
  21  * @defgroup JobQueue JobQueue
  22  * @author Aaron Schulz
  23  */
  24
  25 /**
  26  * Class to handle enqueueing and running of background jobs
  27  *
  28  * @ingroup JobQueue
  29  * @since 1.21
  30  */
  31 abstract class JobQueue {
  32         protected $wiki; // string; wiki ID
  33         protected $type; // string; job type
  34         protected $order; // string; job priority for pop()
  35         protected $claimTTL; // integer; seconds
  36
  37         const QoS_Atomic = 1; // integer; "all-or-nothing" job insertions
  38
  39         const MAX_ATTEMPTS = 3; // integer; number of times to try a job
  40
  41         /**
  42          * @param $params array
  43          */
  44         protected function __construct( array $params ) {
  45                 $this->wiki     = $params['wiki'];
  46                 $this->type     = $params['type'];
  47                 $this->order    = isset( $params['order'] ) ? $params['order'] : 'random';
  48                 $this->claimTTL = isset( $params['claimTTL'] ) ? $params['claimTTL'] : 0;
  49         }
  50
  51         /**
  52          * Get a job queue object of the specified type.
  53          * $params includes:
  54          *   - class    : What job class to use (determines job type)
  55          *   - wiki     : wiki ID of the wiki the jobs are for (defaults to current wiki)
  56          *   - type     : The name of the job types this queue handles
  57          *   - order    : Order that pop() selects jobs, one of "fifo", "timestamp" or "random".
  58          *                If "fifo" is used, the queue will effectively be FIFO. Note that
  59          *                job completion will not appear to be exactly FIFO if there are multiple
  60          *                job runners since jobs can take different times to finish once popped.
  61          *                If "timestamp" is used, the queue will at least be loosely ordered
  62          *                by timestamp, allowing for some jobs to be popped off out of order.
  63          *                If "random" is used, pop() will pick jobs in random order. This might be
  64          *                useful for improving concurrency depending on the queue storage medium.
  65          *   - claimTTL : If supported, the queue will recycle jobs that have been popped
  66          *                but not acknowledged as completed after this many seconds. Recycling
  67          *                of jobs simple means re-inserting them into the queue. Jobs can be
  68          *                attempted up to three times before being discarded.
  69          *
  70          * Queue classes should throw an exception if they do not support the options given.
  71          *
  72          * @param $params array
  73          * @return JobQueue
  74          * @throws MWException
  75          */
  76         final public static function factory( array $params ) {
  77                 $class = $params['class'];
  78                 if ( !MWInit::classExists( $class ) ) {
  79                         throw new MWException( "Invalid job queue class '$class'." );
  80                 }
  81                 $obj = new $class( $params );
  82                 if ( !( $obj instanceof self ) ) {
  83                         throw new MWException( "Class '$class' is not a " . __CLASS__ . " class." );
  84                 }
  85                 return $obj;
  86         }
  87
  88         /**
  89          * @return string Wiki ID
  90          */
  91         final public function getWiki() {
  92                 return $this->wiki;
  93         }
  94
  95         /**
  96          * @return string Job type that this queue handles
  97          */
  98         final public function getType() {
  99                 return $this->type;
 100         }
 101
 102         /**
 103          * Quickly check if the queue is empty (has no available jobs).
 104          * Queue classes should use caching if they are any slower without memcached.
 105          *
 106          * @return bool
 107          * @throws MWException
 108          */
 109         final public function isEmpty() {
 110                 wfProfileIn( __METHOD__ );
 111                 $res = $this->doIsEmpty();
 112                 wfProfileOut( __METHOD__ );
 113                 return $res;
 114         }
 115
 116         /**
 117          * @see JobQueue::isEmpty()
 118          * @return bool
 119          */
 120         abstract protected function doIsEmpty();
 121
 122         /**
 123          * Get the number of available jobs in the queue.
 124          * Queue classes should use caching if they are any slower without memcached.
 125          *
 126          * @return integer
 127          * @throws MWException
 128          */
 129         final public function getSize() {
 130                 wfProfileIn( __METHOD__ );
 131                 $res = $this->doGetSize();
 132                 wfProfileOut( __METHOD__ );
 133                 return $res;
 134         }
 135
 136         /**
 137          * @see JobQueue::getSize()
 138          * @return integer
 139          */
 140         abstract protected function doGetSize();
 141
 142         /**
 143          * Get the number of acquired jobs (these are temporarily out of the queue).
 144          * Queue classes should use caching if they are any slower without memcached.
 145          *
 146          * @return integer
 147          * @throws MWException
 148          */
 149         final public function getAcquiredCount() {
 150                 wfProfileIn( __METHOD__ );
 151                 $res = $this->doGetAcquiredCount();
 152                 wfProfileOut( __METHOD__ );
 153                 return $res;
 154         }
 155
 156         /**
 157          * @see JobQueue::getAcquiredCount()
 158          * @return integer
 159          */
 160         abstract protected function doGetAcquiredCount();
 161
 162         /**
 163          * Push a single jobs into the queue.
 164          * This does not require $wgJobClasses to be set for the given job type.
 165          *
 166          * @param $jobs Job|Array
 167          * @param $flags integer Bitfield (supports JobQueue::QoS_Atomic)
 168          * @return bool Returns false on failure
 169          * @throws MWException
 170          */
 171         final public function push( $jobs, $flags = 0 ) {
 172                 $jobs = is_array( $jobs ) ? $jobs : array( $jobs );
 173
 174                 return $this->batchPush( $jobs, $flags );
 175         }
 176
 177         /**
 178          * Push a batch of jobs into the queue.
 179          * This does not require $wgJobClasses to be set for the given job type.
 180          *
 181          * @param $jobs array List of Jobs
 182          * @param $flags integer Bitfield (supports JobQueue::QoS_Atomic)
 183          * @return bool Returns false on failure
 184          * @throws MWException
 185          */
 186         final public function batchPush( array $jobs, $flags = 0 ) {
 187                 foreach ( $jobs as $job ) {
 188                         if ( $job->getType() !== $this->type ) {
 189                                 throw new MWException( "Got '{$job->getType()}' job; expected '{$this->type}'." );
 190                         }
 191                 }
 192                 wfProfileIn( __METHOD__ );
 193                 $ok = $this->doBatchPush( $jobs, $flags );
 194                 wfProfileOut( __METHOD__ );
 195                 return $ok;
 196         }
 197
 198         /**
 199          * @see JobQueue::batchPush()
 200          * @return bool
 201          */
 202         abstract protected function doBatchPush( array $jobs, $flags );
 203
 204         /**
 205          * Pop a job off of the queue.
 206          * This requires $wgJobClasses to be set for the given job type.
 207          *
 208          * @return Job|bool Returns false on failure
 209          * @throws MWException
 210          */
 211         final public function pop() {
 212                 global $wgJobClasses;
 213
 214                 if ( $this->wiki !== wfWikiID() ) {
 215                         throw new MWException( "Cannot pop '{$this->type}' job off foreign wiki queue." );
 216                 } elseif ( !isset( $wgJobClasses[$this->type] ) ) {
 217                         // Do not pop jobs if there is no class for the queue type
 218                         throw new MWException( "Unrecognized job type '{$this->type}'." );
 219                 }
 220
 221                 wfProfileIn( __METHOD__ );
 222                 $job = $this->doPop();
 223                 wfProfileOut( __METHOD__ );
 224                 return $job;
 225         }
 226
 227         /**
 228          * @see JobQueue::pop()
 229          * @return Job
 230          */
 231         abstract protected function doPop();
 232
 233         /**
 234          * Acknowledge that a job was completed.
 235          *
 236          * This does nothing for certain queue classes or if "claimTTL" is not set.
 237          *
 238          * @param $job Job
 239          * @return bool
 240          * @throws MWException
 241          */
 242         final public function ack( Job $job ) {
 243                 if ( $job->getType() !== $this->type ) {
 244                         throw new MWException( "Got '{$job->getType()}' job; expected '{$this->type}'." );
 245                 }
 246                 wfProfileIn( __METHOD__ );
 247                 $ok = $this->doAck( $job );
 248                 wfProfileOut( __METHOD__ );
 249                 return $ok;
 250         }
 251
 252         /**
 253          * @see JobQueue::ack()
 254          * @return bool
 255          */
 256         abstract protected function doAck( Job $job );
 257
 258         /**
 259          * Register the "root job" of a given job into the queue for de-duplication.
 260          * This should only be called right *after* all the new jobs have been inserted.
 261          * This is used to turn older, duplicate, job entries into no-ops. The root job
 262          * information will remain in the registry until it simply falls out of cache.
 263          *
 264          * This requires that $job has two special fields in the "params" array:
 265          *   - rootJobSignature : hash (e.g. SHA1) that identifies the task
 266          *   - rootJobTimestamp : TS_MW timestamp of this instance of the task
 267          *
 268          * A "root job" is a conceptual job that consist of potentially many smaller jobs
 269          * that are actually inserted into the queue. For example, "refreshLinks" jobs are
 270          * spawned when a template is edited. One can think of the task as "update links
 271          * of pages that use template X" and an instance of that task as a "root job".
 272          * However, what actually goes into the queue are potentially many refreshLinks2 jobs.
 273          * Since these jobs include things like page ID ranges and DB master positions, and morph
 274          * into smaller refreshLinks2 jobs recursively, simple duplicate detection (like job_sha1)
 275          * for individual jobs being identical is not useful.
 276          *
 277          * In the case of "refreshLinks", if these jobs are still in the queue when the template
 278          * is edited again, we want all of these old refreshLinks jobs for that template to become
 279          * no-ops. This can greatly reduce server load, since refreshLinks jobs involves parsing.
 280          * Essentially, the new batch of jobs belong to a new "root job" and the older ones to a
 281          * previous "root job" for the same task of "update links of pages that use template X".
 282          *
 283          * This does nothing for certain queue classes.
 284          *
 285          * @param $job Job
 286          * @return bool
 287          * @throws MWException
 288          */
 289         final public function deduplicateRootJob( Job $job ) {
 290                 if ( $job->getType() !== $this->type ) {
 291                         throw new MWException( "Got '{$job->getType()}' job; expected '{$this->type}'." );
 292                 }
 293                 wfProfileIn( __METHOD__ );
 294                 $ok = $this->doDeduplicateRootJob( $job );
 295                 wfProfileOut( __METHOD__ );
 296                 return $ok;
 297         }
 298
 299         /**
 300          * @see JobQueue::deduplicateRootJob()
 301          * @param $job Job
 302          * @return bool
 303          */
 304         protected function doDeduplicateRootJob( Job $job ) {
 305                 return true;
 306         }
 307
 308         /**
 309          * Wait for any slaves or backup servers to catch up.
 310          *
 311          * This does nothing for certain queue classes.
 312          *
 313          * @return void
 314          * @throws MWException
 315          */
 316         final public function waitForBackups() {
 317                 wfProfileIn( __METHOD__ );
 318                 $this->doWaitForBackups();
 319                 wfProfileOut( __METHOD__ );
 320         }
 321
 322         /**
 323          * @see JobQueue::waitForBackups()
 324          * @return void
 325          */
 326         protected function doWaitForBackups() {}
 327 }