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