includes/job/JobQueueFederated.php

   1 <?php
   2 /**
   3  * Job queue code for federated queues.
   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  * @author Aaron Schulz
  22  */
  23
  24 /**
  25  * Class to handle enqueueing and running of background jobs for federated queues
  26  *
  27  * This class allows for queues to be partitioned into smaller queues.
  28  * A partition is defined by the configuration for a JobQueue instance.
  29  * For example, one can set $wgJobTypeConf['refreshLinks'] to point to a
  30  * JobQueueFederated instance, which itself would consist of three JobQueueRedis
  31  * instances, each using their own redis server. This would allow for the jobs
  32  * to be split (evenly or based on weights) accross multiple servers if a single
  33  * server becomes impractical or expensive. Different JobQueue classes can be mixed.
  34  *
  35  * The basic queue configuration (e.g. "order", "claimTTL") of a federated queue
  36  * is inherited by the partition queues. Additional configuration defines what
  37  * section each wiki is in, what partition queues each section uses (and their weight),
  38  * and the JobQueue configuration for each partition. Some sections might only need a
  39  * single queue partition, like the sections for groups of small wikis.
  40  *
  41  * If used for performance, then $wgMainCacheType should be set to memcached/redis.
  42  * Note that "fifo" cannot be used for the ordering, since the data is distributed.
  43  * One can still use "timestamp" instead, as in "roughly timestamp ordered". Also,
  44  * queue classes used by this should ignore down servers (with TTL) to avoid slowness.
  45  *
  46  * @ingroup JobQueue
  47  * @since 1.22
  48  */
  49 class JobQueueFederated extends JobQueue {
  50         /** @var array (partition name => weight) reverse sorted by weight */
  51         protected $partitionMap = array();
  52
  53         /** @var array (partition name => JobQueue) reverse sorted by weight */
  54         protected $partitionQueues = array();
  55
  56         /** @var HashRing */
  57         protected $partitionPushRing;
  58
  59         /** @var BagOStuff */
  60         protected $cache;
  61
  62         /** @var int Maximum number of partitions to try */
  63         protected $maxPartitionsTry;
  64
  65         const CACHE_TTL_SHORT = 30; // integer; seconds to cache info without re-validating
  66         const CACHE_TTL_LONG = 300; // integer; seconds to cache info that is kept up to date
  67
  68         /**
  69          * @params include:
  70          *  - sectionsByWiki      : A map of wiki IDs to section names.
  71          *                          Wikis will default to using the section "default".
  72          *  - partitionsBySection : Map of section names to maps of (partition name => weight).
  73          *                          A section called 'default' must be defined if not all wikis
  74          *                          have explicitly defined sections.
  75          *  - configByPartition   : Map of queue partition names to configuration arrays.
  76          *                          These configuration arrays are passed to JobQueue::factory().
  77          *                          The options set here are overriden by those passed to this
  78          *                          the federated queue itself (e.g. 'order' and 'claimTTL').
  79          *  - partitionsNoPush    : List of partition names that can handle pop() but not push().
  80          *                          This can be used to migrate away from a certain partition.
  81          *  - maxPartitionsTry    : Maximum number of times to attempt job insertion using
  82          *                          different partition queues. This improves availability
  83          *                          during failure, at the cost of added latency and somewhat
  84          *                          less reliable job de-duplication mechanisms.
  85          * @param array $params
  86          * @throws MWException
  87          */
  88         protected function __construct( array $params ) {
  89                 parent::__construct( $params );
  90                 $section = isset( $params['sectionsByWiki'][$this->wiki] )
  91                         ? $params['sectionsByWiki'][$this->wiki]
  92                         : 'default';
  93                 if ( !isset( $params['partitionsBySection'][$section] ) ) {
  94                         throw new MWException( "No configuration for section '$section'." );
  95                 }
  96                 $this->maxPartitionsTry = isset( $params['maxPartitionsTry'] )
  97                         ? $params['maxPartitionsTry']
  98                         : 2;
  99                 // Get the full partition map
 100                 $this->partitionMap = $params['partitionsBySection'][$section];
 101                 arsort( $this->partitionMap, SORT_NUMERIC );
 102                 // Get the partitions jobs can actually be pushed to
 103                 $partitionPushMap = $this->partitionMap;
 104                 if ( isset( $params['partitionsNoPush'] ) ) {
 105                         foreach ( $params['partitionsNoPush'] as $partition ) {
 106                                 unset( $partitionPushMap[$partition] );
 107                         }
 108                 }
 109                 // Get the config to pass to merge into each partition queue config
 110                 $baseConfig = $params;
 111                 foreach ( array( 'class', 'sectionsByWiki', 'maxPartitionsTry',
 112                         'partitionsBySection', 'configByPartition', 'partitionsNoPush' ) as $o
 113                 ) {
 114                         unset( $baseConfig[$o] ); // partition queue doesn't care about this
 115                 }
 116                 // Get the partition queue objects
 117                 foreach ( $this->partitionMap as $partition => $w ) {
 118                         if ( !isset( $params['configByPartition'][$partition] ) ) {
 119                                 throw new MWException( "No configuration for partition '$partition'." );
 120                         }
 121                         $this->partitionQueues[$partition] = JobQueue::factory(
 122                                 $baseConfig + $params['configByPartition'][$partition] );
 123                 }
 124                 // Get the ring of partitions to push jobs into
 125                 $this->partitionPushRing = new HashRing( $partitionPushMap );
 126                 // Aggregate cache some per-queue values if there are multiple partition queues
 127                 $this->cache = count( $this->partitionMap ) > 1 ? wfGetMainCache() : new EmptyBagOStuff();
 128         }
 129
 130         protected function supportedOrders() {
 131                 // No FIFO due to partitioning, though "rough timestamp order" is supported
 132                 return array( 'undefined', 'random', 'timestamp' );
 133         }
 134
 135         protected function optimalOrder() {
 136                 return 'undefined'; // defer to the partitions
 137         }
 138
 139         protected function supportsDelayedJobs() {
 140                 return true; // defer checks to the partitions
 141         }
 142
 143         protected function doIsEmpty() {
 144                 $key = $this->getCacheKey( 'empty' );
 145
 146                 $isEmpty = $this->cache->get( $key );
 147                 if ( $isEmpty === 'true' ) {
 148                         return true;
 149                 } elseif ( $isEmpty === 'false' ) {
 150                         return false;
 151                 }
 152
 153                 foreach ( $this->partitionQueues as $queue ) {
 154                         try {
 155                                 if ( !$queue->doIsEmpty() ) {
 156                                         $this->cache->add( $key, 'false', self::CACHE_TTL_LONG );
 157
 158                                         return false;
 159                                 }
 160                         } catch ( JobQueueError $e ) {
 161                                 MWExceptionHandler::logException( $e );
 162                         }
 163                 }
 164
 165                 $this->cache->add( $key, 'true', self::CACHE_TTL_LONG );
 166
 167                 return true;
 168         }
 169
 170         protected function doGetSize() {
 171                 return $this->getCrossPartitionSum( 'size', 'doGetSize' );
 172         }
 173
 174         protected function doGetAcquiredCount() {
 175                 return $this->getCrossPartitionSum( 'acquiredcount', 'doGetAcquiredCount' );
 176         }
 177
 178         protected function doGetDelayedCount() {
 179                 return $this->getCrossPartitionSum( 'delayedcount', 'doGetDelayedCount' );
 180         }
 181
 182         protected function doGetAbandonedCount() {
 183                 return $this->getCrossPartitionSum( 'abandonedcount', 'doGetAbandonedCount' );
 184         }
 185
 186         /**
 187          * @param string $type
 188          * @param string $method
 189          * @return int
 190          */
 191         protected function getCrossPartitionSum( $type, $method ) {
 192                 $key = $this->getCacheKey( $type );
 193
 194                 $count = $this->cache->get( $key );
 195                 if ( is_int( $count ) ) {
 196                         return $count;
 197                 }
 198
 199                 $count = 0;
 200                 foreach ( $this->partitionQueues as $queue ) {
 201                         try {
 202                                 $count += $queue->$method();
 203                         } catch ( JobQueueError $e ) {
 204                                 MWExceptionHandler::logException( $e );
 205                         }
 206                 }
 207
 208                 $this->cache->set( $key, $count, self::CACHE_TTL_SHORT );
 209
 210                 return $count;
 211         }
 212
 213         protected function doBatchPush( array $jobs, $flags ) {
 214                 // Local ring variable that may be changed to point to a new ring on failure
 215                 $partitionRing = $this->partitionPushRing;
 216                 // Try to insert the jobs and update $partitionsTry on any failures.
 217                 // Retry to insert any remaning jobs again, ignoring the bad partitions.
 218                 $jobsLeft = $jobs;
 219                 for ( $i = $this->maxPartitionsTry; $i > 0 && count( $jobsLeft ); --$i ) {
 220                         $jobsLeft = $this->tryJobInsertions( $jobsLeft, $partitionRing, $flags );
 221                 }
 222                 if ( count( $jobsLeft ) ) {
 223                         throw new JobQueueError(
 224                                 "Could not insert job(s), {$this->maxPartitionsTry} partitions tried." );
 225                 }
 226
 227                 return true;
 228         }
 229
 230         /**
 231          * @param array $jobs
 232          * @param HashRing $partitionRing
 233          * @param int $flags
 234          * @throws JobQueueError
 235          * @return array List of Job object that could not be inserted
 236          */
 237         protected function tryJobInsertions( array $jobs, HashRing &$partitionRing, $flags ) {
 238                 $jobsLeft = array();
 239
 240                 // Because jobs are spread across partitions, per-job de-duplication needs
 241                 // to use a consistent hash to avoid allowing duplicate jobs per partition.
 242                 // When inserting a batch of de-duplicated jobs, QOS_ATOMIC is disregarded.
 243                 $uJobsByPartition = array(); // (partition name => job list)
 244                 /** @var Job $job */
 245                 foreach ( $jobs as $key => $job ) {
 246                         if ( $job->ignoreDuplicates() ) {
 247                                 $sha1 = sha1( serialize( $job->getDeduplicationInfo() ) );
 248                                 $uJobsByPartition[$partitionRing->getLocation( $sha1 )][] = $job;
 249                                 unset( $jobs[$key] );
 250                         }
 251                 }
 252                 // Get the batches of jobs that are not de-duplicated
 253                 if ( $flags & self::QOS_ATOMIC ) {
 254                         $nuJobBatches = array( $jobs ); // all or nothing
 255                 } else {
 256                         // Split the jobs into batches and spread them out over servers if there
 257                         // are many jobs. This helps keep the partitions even. Otherwise, send all
 258                         // the jobs to a single partition queue to avoids the extra connections.
 259                         $nuJobBatches = array_chunk( $jobs, 300 );
 260                 }
 261
 262                 // Insert the de-duplicated jobs into the queues...
 263                 foreach ( $uJobsByPartition as $partition => $jobBatch ) {
 264                         /** @var JobQueue $queue */
 265                         $queue = $this->partitionQueues[$partition];
 266                         try {
 267                                 $ok = $queue->doBatchPush( $jobBatch, $flags | self::QOS_ATOMIC );
 268                         } catch ( JobQueueError $e ) {
 269                                 $ok = false;
 270                                 MWExceptionHandler::logException( $e );
 271                         }
 272                         if ( $ok ) {
 273                                 $key = $this->getCacheKey( 'empty' );
 274                                 $this->cache->set( $key, 'false', JobQueueDB::CACHE_TTL_LONG );
 275                         } else {
 276                                 $partitionRing = $partitionRing->newWithoutLocation( $partition ); // blacklist
 277                                 if ( !$partitionRing ) {
 278                                         throw new JobQueueError( "Could not insert job(s), all partitions are down." );
 279                                 }
 280                                 $jobsLeft = array_merge( $jobsLeft, $jobBatch ); // not inserted
 281                         }
 282                 }
 283
 284                 // Insert the jobs that are not de-duplicated into the queues...
 285                 foreach ( $nuJobBatches as $jobBatch ) {
 286                         $partition = ArrayUtils::pickRandom( $partitionRing->getLocationWeights() );
 287                         $queue = $this->partitionQueues[$partition];
 288                         try {
 289                                 $ok = $queue->doBatchPush( $jobBatch, $flags | self::QOS_ATOMIC );
 290                         } catch ( JobQueueError $e ) {
 291                                 $ok = false;
 292                                 MWExceptionHandler::logException( $e );
 293                         }
 294                         if ( $ok ) {
 295                                 $key = $this->getCacheKey( 'empty' );
 296                                 $this->cache->set( $key, 'false', JobQueueDB::CACHE_TTL_LONG );
 297                         } else {
 298                                 $partitionRing = $partitionRing->newWithoutLocation( $partition ); // blacklist
 299                                 if ( !$partitionRing ) {
 300                                         throw new JobQueueError( "Could not insert job(s), all partitions are down." );
 301                                 }
 302                                 $jobsLeft = array_merge( $jobsLeft, $jobBatch ); // not inserted
 303                         }
 304                 }
 305
 306                 return $jobsLeft;
 307         }
 308
 309         protected function doPop() {
 310                 $key = $this->getCacheKey( 'empty' );
 311
 312                 $isEmpty = $this->cache->get( $key );
 313                 if ( $isEmpty === 'true' ) {
 314                         return false;
 315                 }
 316
 317                 $partitionsTry = $this->partitionMap; // (partition => weight)
 318
 319                 while ( count( $partitionsTry ) ) {
 320                         $partition = ArrayUtils::pickRandom( $partitionsTry );
 321                         if ( $partition === false ) {
 322                                 break; // all partitions at 0 weight
 323                         }
 324
 325                         /** @var JobQueue $queue */
 326                         $queue = $this->partitionQueues[$partition];
 327                         try {
 328                                 $job = $queue->pop();
 329                         } catch ( JobQueueError $e ) {
 330                                 $job = false;
 331                                 MWExceptionHandler::logException( $e );
 332                         }
 333                         if ( $job ) {
 334                                 $job->metadata['QueuePartition'] = $partition;
 335
 336                                 return $job;
 337                         } else {
 338                                 unset( $partitionsTry[$partition] ); // blacklist partition
 339                         }
 340                 }
 341
 342                 $this->cache->set( $key, 'true', JobQueueDB::CACHE_TTL_LONG );
 343
 344                 return false;
 345         }
 346
 347         protected function doAck( Job $job ) {
 348                 if ( !isset( $job->metadata['QueuePartition'] ) ) {
 349                         throw new MWException( "The given job has no defined partition name." );
 350                 }
 351
 352                 return $this->partitionQueues[$job->metadata['QueuePartition']]->ack( $job );
 353         }
 354
 355         protected function doIsRootJobOldDuplicate( Job $job ) {
 356                 $params = $job->getRootJobParams();
 357                 $partitions = $this->partitionPushRing->getLocations( $params['rootJobSignature'], 2 );
 358                 try {
 359                         return $this->partitionQueues[$partitions[0]]->doIsRootJobOldDuplicate( $job );
 360                 } catch ( JobQueueError $e ) {
 361                         if ( isset( $partitions[1] ) ) { // check fallback partition
 362                                 return $this->partitionQueues[$partitions[1]]->doIsRootJobOldDuplicate( $job );
 363                         }
 364                 }
 365
 366                 return false;
 367         }
 368
 369         protected function doDeduplicateRootJob( Job $job ) {
 370                 $params = $job->getRootJobParams();
 371                 $partitions = $this->partitionPushRing->getLocations( $params['rootJobSignature'], 2 );
 372                 try {
 373                         return $this->partitionQueues[$partitions[0]]->doDeduplicateRootJob( $job );
 374                 } catch ( JobQueueError $e ) {
 375                         if ( isset( $partitions[1] ) ) { // check fallback partition
 376                                 return $this->partitionQueues[$partitions[1]]->doDeduplicateRootJob( $job );
 377                         }
 378                 }
 379
 380                 return false;
 381         }
 382
 383         protected function doDelete() {
 384                 /** @var JobQueue $queue */
 385                 foreach ( $this->partitionQueues as $queue ) {
 386                         try {
 387                                 $queue->doDelete();
 388                         } catch ( JobQueueError $e ) {
 389                                 MWExceptionHandler::logException( $e );
 390                         }
 391                 }
 392         }
 393
 394         protected function doWaitForBackups() {
 395                 /** @var JobQueue $queue */
 396                 foreach ( $this->partitionQueues as $queue ) {
 397                         try {
 398                                 $queue->waitForBackups();
 399                         } catch ( JobQueueError $e ) {
 400                                 MWExceptionHandler::logException( $e );
 401                         }
 402                 }
 403         }
 404
 405         protected function doGetPeriodicTasks() {
 406                 $tasks = array();
 407                 /** @var JobQueue $queue */
 408                 foreach ( $this->partitionQueues as $partition => $queue ) {
 409                         foreach ( $queue->getPeriodicTasks() as $task => $def ) {
 410                                 $tasks["{$partition}:{$task}"] = $def;
 411                         }
 412                 }
 413
 414                 return $tasks;
 415         }
 416
 417         protected function doFlushCaches() {
 418                 static $types = array(
 419                         'empty',
 420                         'size',
 421                         'acquiredcount',
 422                         'delayedcount',
 423                         'abandonedcount'
 424                 );
 425
 426                 foreach ( $types as $type ) {
 427                         $this->cache->delete( $this->getCacheKey( $type ) );
 428                 }
 429
 430                 /** @var JobQueue $queue */
 431                 foreach ( $this->partitionQueues as $queue ) {
 432                         $queue->doFlushCaches();
 433                 }
 434         }
 435
 436         public function getAllQueuedJobs() {
 437                 $iterator = new AppendIterator();
 438
 439                 /** @var JobQueue $queue */
 440                 foreach ( $this->partitionQueues as $queue ) {
 441                         $iterator->append( $queue->getAllQueuedJobs() );
 442                 }
 443
 444                 return $iterator;
 445         }
 446
 447         public function getAllDelayedJobs() {
 448                 $iterator = new AppendIterator();
 449
 450                 /** @var JobQueue $queue */
 451                 foreach ( $this->partitionQueues as $queue ) {
 452                         $iterator->append( $queue->getAllDelayedJobs() );
 453                 }
 454
 455                 return $iterator;
 456         }
 457
 458         public function getCoalesceLocationInternal() {
 459                 return "JobQueueFederated:wiki:{$this->wiki}" .
 460                         sha1( serialize( array_keys( $this->partitionMap ) ) );
 461         }
 462
 463         protected function doGetSiblingQueuesWithJobs( array $types ) {
 464                 $result = array();
 465
 466                 /** @var JobQueue $queue */
 467                 foreach ( $this->partitionQueues as $queue ) {
 468                         try {
 469                                 $nonEmpty = $queue->doGetSiblingQueuesWithJobs( $types );
 470                                 if ( is_array( $nonEmpty ) ) {
 471                                         $result = array_unique( array_merge( $result, $nonEmpty ) );
 472                                 } else {
 473                                         return null; // not supported on all partitions; bail
 474                                 }
 475                                 if ( count( $result ) == count( $types ) ) {
 476                                         break; // short-circuit
 477                                 }
 478                         } catch ( JobQueueError $e ) {
 479                                 MWExceptionHandler::logException( $e );
 480                         }
 481                 }
 482
 483                 return array_values( $result );
 484         }
 485
 486         protected function doGetSiblingQueueSizes( array $types ) {
 487                 $result = array();
 488
 489                 /** @var JobQueue $queue */
 490                 foreach ( $this->partitionQueues as $queue ) {
 491                         try {
 492                                 $sizes = $queue->doGetSiblingQueueSizes( $types );
 493                                 if ( is_array( $sizes ) ) {
 494                                         foreach ( $sizes as $type => $size ) {
 495                                                 $result[$type] = isset( $result[$type] ) ? $result[$type] + $size : $size;
 496                                         }
 497                                 } else {
 498                                         return null; // not supported on all partitions; bail
 499                                 }
 500                         } catch ( JobQueueError $e ) {
 501                                 MWExceptionHandler::logException( $e );
 502                         }
 503                 }
 504
 505                 return $result;
 506         }
 507
 508         public function setTestingPrefix( $key ) {
 509                 /** @var JobQueue $queue */
 510                 foreach ( $this->partitionQueues as $queue ) {
 511                         $queue->setTestingPrefix( $key );
 512                 }
 513         }
 514
 515         /**
 516          * @param $property
 517          * @return string
 518          */
 519         private function getCacheKey( $property ) {
 520                 list( $db, $prefix ) = wfSplitWikiID( $this->wiki );
 521
 522                 return wfForeignMemcKey( $db, $prefix, 'jobqueue', $this->type, $property );
 523         }
 524 }