includes/jobqueue/JobRunner.php

   1 <?php
   2 /**
   3  * Job queue runner utility methods
   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  * @ingroup JobQueue
  22  */
  23
  24 use MediaWiki\MediaWikiServices;
  25 use MediaWiki\Logger\LoggerFactory;
  26 use Liuggio\StatsdClient\Factory\StatsdDataFactory;
  27 use Psr\Log\LoggerAwareInterface;
  28 use Psr\Log\LoggerInterface;
  29 use Wikimedia\ScopedCallback;
  30 use Wikimedia\Rdbms\LBFactory;
  31 use Wikimedia\Rdbms\DBError;
  32
  33 /**
  34  * Job queue runner utility methods
  35  *
  36  * @ingroup JobQueue
  37  * @since 1.24
  38  */
  39 class JobRunner implements LoggerAwareInterface {
  40         /** @var Config */
  41         protected $config;
  42         /** @var callable|null Debug output handler */
  43         protected $debug;
  44
  45         /**
  46          * @var LoggerInterface $logger
  47          */
  48         protected $logger;
  49
  50         const MAX_ALLOWED_LAG = 3; // abort if more than this much DB lag is present
  51         const LAG_CHECK_PERIOD = 1.0; // check replica DB lag this many seconds
  52         const ERROR_BACKOFF_TTL = 1; // seconds to back off a queue due to errors
  53         const READONLY_BACKOFF_TTL = 30; // seconds to back off a queue due to read-only errors
  54
  55         /**
  56          * @param callable $debug Optional debug output handler
  57          */
  58         public function setDebugHandler( $debug ) {
  59                 $this->debug = $debug;
  60         }
  61
  62         /**
  63          * @param LoggerInterface $logger
  64          * @return void
  65          */
  66         public function setLogger( LoggerInterface $logger ) {
  67                 $this->logger = $logger;
  68         }
  69
  70         /**
  71          * @param LoggerInterface|null $logger
  72          */
  73         public function __construct( LoggerInterface $logger = null ) {
  74                 if ( $logger === null ) {
  75                         $logger = LoggerFactory::getInstance( 'runJobs' );
  76                 }
  77                 $this->setLogger( $logger );
  78                 $this->config = MediaWikiServices::getInstance()->getMainConfig();
  79         }
  80
  81         /**
  82          * Run jobs of the specified number/type for the specified time
  83          *
  84          * The response map has a 'job' field that lists status of each job, including:
  85          *   - type   : the job type
  86          *   - status : ok/failed
  87          *   - error  : any error message string
  88          *   - time   : the job run time in ms
  89          * The response map also has:
  90          *   - backoffs : the (job type => seconds) map of backoff times
  91          *   - elapsed  : the total time spent running tasks in ms
  92          *   - reached  : the reason the script finished, one of (none-ready, job-limit, time-limit,
  93          *  memory-limit)
  94          *
  95          * This method outputs status information only if a debug handler was set.
  96          * Any exceptions are caught and logged, but are not reported as output.
  97          *
  98          * @param array $options Map of parameters:
  99          *    - type     : the job type (or false for the default types)
 100          *    - maxJobs  : maximum number of jobs to run
 101          *    - maxTime  : maximum time in seconds before stopping
 102          *    - throttle : whether to respect job backoff configuration
 103          * @return array Summary response that can easily be JSON serialized
 104          */
 105         public function run( array $options ) {
 106                 $jobClasses = $this->config->get( 'JobClasses' );
 107                 $profilerLimits = $this->config->get( 'TrxProfilerLimits' );
 108
 109                 $response = [ 'jobs' => [], 'reached' => 'none-ready' ];
 110
 111                 $type = $options['type'] ?? false;
 112                 $maxJobs = $options['maxJobs'] ?? false;
 113                 $maxTime = $options['maxTime'] ?? false;
 114                 $noThrottle = isset( $options['throttle'] ) && !$options['throttle'];
 115
 116                 // Bail if job type is invalid
 117                 if ( $type !== false && !isset( $jobClasses[$type] ) ) {
 118                         $response['reached'] = 'none-possible';
 119                         return $response;
 120                 }
 121
 122                 // Bail out if DB is in read-only mode
 123                 if ( wfReadOnly() ) {
 124                         $response['reached'] = 'read-only';
 125                         return $response;
 126                 }
 127
 128                 $lbFactory = MediaWikiServices::getInstance()->getDBLoadBalancerFactory();
 129                 if ( $lbFactory->hasTransactionRound() ) {
 130                         throw new LogicException( __METHOD__ . ' called with an active transaction round.' );
 131                 }
 132                 // Bail out if there is too much DB lag.
 133                 // This check should not block as we want to try other wiki queues.
 134                 list( , $maxLag ) = $lbFactory->getMainLB( wfWikiID() )->getMaxLag();
 135                 if ( $maxLag >= self::MAX_ALLOWED_LAG ) {
 136                         $response['reached'] = 'replica-lag-limit';
 137                         return $response;
 138                 }
 139
 140                 // Catch huge single updates that lead to replica DB lag
 141                 $trxProfiler = Profiler::instance()->getTransactionProfiler();
 142                 $trxProfiler->setLogger( LoggerFactory::getInstance( 'DBPerformance' ) );
 143                 $trxProfiler->setExpectations( $profilerLimits['JobRunner'], __METHOD__ );
 144
 145                 // Some jobs types should not run until a certain timestamp
 146                 $backoffs = []; // map of (type => UNIX expiry)
 147                 $backoffDeltas = []; // map of (type => seconds)
 148                 $wait = 'wait'; // block to read backoffs the first time
 149
 150                 $group = JobQueueGroup::singleton();
 151                 $stats = MediaWikiServices::getInstance()->getStatsdDataFactory();
 152                 $jobsPopped = 0;
 153                 $timeMsTotal = 0;
 154                 $startTime = microtime( true ); // time since jobs started running
 155                 $lastCheckTime = 1; // timestamp of last replica DB check
 156                 do {
 157                         // Sync the persistent backoffs with concurrent runners
 158                         $backoffs = $this->syncBackoffDeltas( $backoffs, $backoffDeltas, $wait );
 159                         $blacklist = $noThrottle ? [] : array_keys( $backoffs );
 160                         $wait = 'nowait'; // less important now
 161
 162                         if ( $type === false ) {
 163                                 $job = $group->pop(
 164                                         JobQueueGroup::TYPE_DEFAULT,
 165                                         JobQueueGroup::USE_CACHE,
 166                                         $blacklist
 167                                 );
 168                         } elseif ( in_array( $type, $blacklist ) ) {
 169                                 $job = false; // requested queue in backoff state
 170                         } else {
 171                                 $job = $group->pop( $type ); // job from a single queue
 172                         }
 173
 174                         if ( $job ) { // found a job
 175                                 ++$jobsPopped;
 176                                 $popTime = time();
 177                                 $jType = $job->getType();
 178
 179                                 WebRequest::overrideRequestId( $job->getRequestId() );
 180
 181                                 // Back off of certain jobs for a while (for throttling and for errors)
 182                                 $ttw = $this->getBackoffTimeToWait( $job );
 183                                 if ( $ttw > 0 ) {
 184                                         // Always add the delta for other runners in case the time running the
 185                                         // job negated the backoff for each individually but not collectively.
 186                                         $backoffDeltas[$jType] = isset( $backoffDeltas[$jType] )
 187                                                 ? $backoffDeltas[$jType] + $ttw
 188                                                 : $ttw;
 189                                         $backoffs = $this->syncBackoffDeltas( $backoffs, $backoffDeltas, $wait );
 190                                 }
 191
 192                                 $info = $this->executeJob( $job, $lbFactory, $stats, $popTime );
 193                                 if ( $info['status'] !== false || !$job->allowRetries() ) {
 194                                         $group->ack( $job ); // succeeded or job cannot be retried
 195                                 }
 196
 197                                 // Back off of certain jobs for a while (for throttling and for errors)
 198                                 if ( $info['status'] === false && mt_rand( 0, 49 ) == 0 ) {
 199                                         $ttw = max( $ttw, $this->getErrorBackoffTTL( $info['error'] ) );
 200                                         $backoffDeltas[$jType] = isset( $backoffDeltas[$jType] )
 201                                                 ? $backoffDeltas[$jType] + $ttw
 202                                                 : $ttw;
 203                                 }
 204
 205                                 $response['jobs'][] = [
 206                                         'type'   => $jType,
 207                                         'status' => ( $info['status'] === false ) ? 'failed' : 'ok',
 208                                         'error'  => $info['error'],
 209                                         'time'   => $info['timeMs']
 210                                 ];
 211                                 $timeMsTotal += $info['timeMs'];
 212
 213                                 // Break out if we hit the job count or wall time limits...
 214                                 if ( $maxJobs && $jobsPopped >= $maxJobs ) {
 215                                         $response['reached'] = 'job-limit';
 216                                         break;
 217                                 } elseif ( $maxTime && ( microtime( true ) - $startTime ) > $maxTime ) {
 218                                         $response['reached'] = 'time-limit';
 219                                         break;
 220                                 }
 221
 222                                 // Don't let any of the main DB replica DBs get backed up.
 223                                 // This only waits for so long before exiting and letting
 224                                 // other wikis in the farm (on different masters) get a chance.
 225                                 $timePassed = microtime( true ) - $lastCheckTime;
 226                                 if ( $timePassed >= self::LAG_CHECK_PERIOD || $timePassed < 0 ) {
 227                                         $success = $lbFactory->waitForReplication( [
 228                                                 'ifWritesSince' => $lastCheckTime,
 229                                                 'timeout' => self::MAX_ALLOWED_LAG,
 230                                         ] );
 231                                         if ( !$success ) {
 232                                                 $response['reached'] = 'replica-lag-limit';
 233                                                 break;
 234                                         }
 235                                         $lastCheckTime = microtime( true );
 236                                 }
 237
 238                                 // Bail if near-OOM instead of in a job
 239                                 if ( !$this->checkMemoryOK() ) {
 240                                         $response['reached'] = 'memory-limit';
 241                                         break;
 242                                 }
 243                         }
 244                 } while ( $job ); // stop when there are no jobs
 245
 246                 // Sync the persistent backoffs for the next runJobs.php pass
 247                 if ( $backoffDeltas ) {
 248                         $this->syncBackoffDeltas( $backoffs, $backoffDeltas, 'wait' );
 249                 }
 250
 251                 $response['backoffs'] = $backoffs;
 252                 $response['elapsed'] = $timeMsTotal;
 253
 254                 return $response;
 255         }
 256
 257         /**
 258          * @param string $error
 259          * @return int TTL in seconds
 260          */
 261         private function getErrorBackoffTTL( $error ) {
 262                 return strpos( $error, 'DBReadOnlyError' ) !== false
 263                         ? self::READONLY_BACKOFF_TTL
 264                         : self::ERROR_BACKOFF_TTL;
 265         }
 266
 267         /**
 268          * @param Job $job
 269          * @param LBFactory $lbFactory
 270          * @param StatsdDataFactory $stats
 271          * @param float $popTime
 272          * @return array Map of status/error/timeMs
 273          */
 274         private function executeJob( Job $job, LBFactory $lbFactory, $stats, $popTime ) {
 275                 $jType = $job->getType();
 276                 $msg = $job->toString() . " STARTING";
 277                 $this->logger->debug( $msg, [
 278                         'job_type' => $job->getType(),
 279                 ] );
 280                 $this->debugCallback( $msg );
 281
 282                 // Run the job...
 283                 $rssStart = $this->getMaxRssKb();
 284                 $jobStartTime = microtime( true );
 285                 try {
 286                         $fnameTrxOwner = get_class( $job ) . '::run'; // give run() outer scope
 287                         if ( !$job->hasExecutionFlag( $job::JOB_NO_EXPLICIT_TRX_ROUND ) ) {
 288                                 $lbFactory->beginMasterChanges( $fnameTrxOwner );
 289                         }
 290                         $status = $job->run();
 291                         $error = $job->getLastError();
 292                         $this->commitMasterChanges( $lbFactory, $job, $fnameTrxOwner );
 293                         // Important: this must be the last deferred update added (T100085, T154425)
 294                         DeferredUpdates::addCallableUpdate( [ JobQueueGroup::class, 'pushLazyJobs' ] );
 295                         // Run any deferred update tasks; doUpdates() manages transactions itself
 296                         DeferredUpdates::doUpdates();
 297                 } catch ( Exception $e ) {
 298                         MWExceptionHandler::rollbackMasterChangesAndLog( $e );
 299                         $status = false;
 300                         $error = get_class( $e ) . ': ' . $e->getMessage();
 301                 }
 302                 // Always attempt to call teardown() even if Job throws exception.
 303                 try {
 304                         $job->teardown( $status );
 305                 } catch ( Exception $e ) {
 306                         MWExceptionHandler::logException( $e );
 307                 }
 308
 309                 // Commit all outstanding connections that are in a transaction
 310                 // to get a fresh repeatable read snapshot on every connection.
 311                 // Note that jobs are still responsible for handling replica DB lag.
 312                 $lbFactory->flushReplicaSnapshots( __METHOD__ );
 313                 // Clear out title cache data from prior snapshots
 314                 MediaWikiServices::getInstance()->getLinkCache()->clear();
 315                 $timeMs = intval( ( microtime( true ) - $jobStartTime ) * 1000 );
 316                 $rssEnd = $this->getMaxRssKb();
 317
 318                 // Record how long jobs wait before getting popped
 319                 $readyTs = $job->getReadyTimestamp();
 320                 if ( $readyTs ) {
 321                         $pickupDelay = max( 0, $popTime - $readyTs );
 322                         $stats->timing( 'jobqueue.pickup_delay.all', 1000 * $pickupDelay );
 323                         $stats->timing( "jobqueue.pickup_delay.$jType", 1000 * $pickupDelay );
 324                 }
 325                 // Record root job age for jobs being run
 326                 $rootTimestamp = $job->getRootJobParams()['rootJobTimestamp'];
 327                 if ( $rootTimestamp ) {
 328                         $age = max( 0, $popTime - wfTimestamp( TS_UNIX, $rootTimestamp ) );
 329                         $stats->timing( "jobqueue.pickup_root_age.$jType", 1000 * $age );
 330                 }
 331                 // Track the execution time for jobs
 332                 $stats->timing( "jobqueue.run.$jType", $timeMs );
 333                 // Track RSS increases for jobs (in case of memory leaks)
 334                 if ( $rssStart && $rssEnd ) {
 335                         $stats->updateCount( "jobqueue.rss_delta.$jType", $rssEnd - $rssStart );
 336                 }
 337
 338                 if ( $status === false ) {
 339                         $msg = $job->toString() . " t={job_duration} error={job_error}";
 340                         $this->logger->error( $msg, [
 341                                 'job_type' => $job->getType(),
 342                                 'job_duration' => $timeMs,
 343                                 'job_error' => $error,
 344                         ] );
 345
 346                         $msg = $job->toString() . " t=$timeMs error={$error}";
 347                         $this->debugCallback( $msg );
 348                 } else {
 349                         $msg = $job->toString() . " t={job_duration} good";
 350                         $this->logger->info( $msg, [
 351                                 'job_type' => $job->getType(),
 352                                 'job_duration' => $timeMs,
 353                         ] );
 354
 355                         $msg = $job->toString() . " t=$timeMs good";
 356                         $this->debugCallback( $msg );
 357                 }
 358
 359                 return [ 'status' => $status, 'error' => $error, 'timeMs' => $timeMs ];
 360         }
 361
 362         /**
 363          * @return int|null Max memory RSS in kilobytes
 364          */
 365         private function getMaxRssKb() {
 366                 $info = wfGetRusage() ?: [];
 367                 // see https://linux.die.net/man/2/getrusage
 368                 return isset( $info['ru_maxrss'] ) ? (int)$info['ru_maxrss'] : null;
 369         }
 370
 371         /**
 372          * @param Job $job
 373          * @return int Seconds for this runner to avoid doing more jobs of this type
 374          * @see $wgJobBackoffThrottling
 375          */
 376         private function getBackoffTimeToWait( Job $job ) {
 377                 $throttling = $this->config->get( 'JobBackoffThrottling' );
 378
 379                 if ( !isset( $throttling[$job->getType()] ) || $job instanceof DuplicateJob ) {
 380                         return 0; // not throttled
 381                 }
 382
 383                 $itemsPerSecond = $throttling[$job->getType()];
 384                 if ( $itemsPerSecond <= 0 ) {
 385                         return 0; // not throttled
 386                 }
 387
 388                 $seconds = 0;
 389                 if ( $job->workItemCount() > 0 ) {
 390                         $exactSeconds = $job->workItemCount() / $itemsPerSecond;
 391                         // use randomized rounding
 392                         $seconds = floor( $exactSeconds );
 393                         $remainder = $exactSeconds - $seconds;
 394                         $seconds += ( mt_rand() / mt_getrandmax() < $remainder ) ? 1 : 0;
 395                 }
 396
 397                 return (int)$seconds;
 398         }
 399
 400         /**
 401          * Get the previous backoff expiries from persistent storage
 402          * On I/O or lock acquisition failure this returns the original $backoffs.
 403          *
 404          * @param array $backoffs Map of (job type => UNIX timestamp)
 405          * @param string $mode Lock wait mode - "wait" or "nowait"
 406          * @return array Map of (job type => backoff expiry timestamp)
 407          */
 408         private function loadBackoffs( array $backoffs, $mode = 'wait' ) {
 409                 $file = wfTempDir() . '/mw-runJobs-backoffs.json';
 410                 if ( is_file( $file ) ) {
 411                         $noblock = ( $mode === 'nowait' ) ? LOCK_NB : 0;
 412                         $handle = fopen( $file, 'rb' );
 413                         if ( !flock( $handle, LOCK_SH | $noblock ) ) {
 414                                 fclose( $handle );
 415                                 return $backoffs; // don't wait on lock
 416                         }
 417                         $content = stream_get_contents( $handle );
 418                         flock( $handle, LOCK_UN );
 419                         fclose( $handle );
 420                         $ctime = microtime( true );
 421                         $cBackoffs = json_decode( $content, true ) ?: [];
 422                         foreach ( $cBackoffs as $type => $timestamp ) {
 423                                 if ( $timestamp < $ctime ) {
 424                                         unset( $cBackoffs[$type] );
 425                                 }
 426                         }
 427                 } else {
 428                         $cBackoffs = [];
 429                 }
 430
 431                 return $cBackoffs;
 432         }
 433
 434         /**
 435          * Merge the current backoff expiries from persistent storage
 436          *
 437          * The $deltas map is set to an empty array on success.
 438          * On I/O or lock acquisition failure this returns the original $backoffs.
 439          *
 440          * @param array $backoffs Map of (job type => UNIX timestamp)
 441          * @param array $deltas Map of (job type => seconds)
 442          * @param string $mode Lock wait mode - "wait" or "nowait"
 443          * @return array The new backoffs account for $backoffs and the latest file data
 444          */
 445         private function syncBackoffDeltas( array $backoffs, array &$deltas, $mode = 'wait' ) {
 446                 if ( !$deltas ) {
 447                         return $this->loadBackoffs( $backoffs, $mode );
 448                 }
 449
 450                 $noblock = ( $mode === 'nowait' ) ? LOCK_NB : 0;
 451                 $file = wfTempDir() . '/mw-runJobs-backoffs.json';
 452                 $handle = fopen( $file, 'wb+' );
 453                 if ( !flock( $handle, LOCK_EX | $noblock ) ) {
 454                         fclose( $handle );
 455                         return $backoffs; // don't wait on lock
 456                 }
 457                 $ctime = microtime( true );
 458                 $content = stream_get_contents( $handle );
 459                 $cBackoffs = json_decode( $content, true ) ?: [];
 460                 foreach ( $deltas as $type => $seconds ) {
 461                         $cBackoffs[$type] = isset( $cBackoffs[$type] ) && $cBackoffs[$type] >= $ctime
 462                                 ? $cBackoffs[$type] + $seconds
 463                                 : $ctime + $seconds;
 464                 }
 465                 foreach ( $cBackoffs as $type => $timestamp ) {
 466                         if ( $timestamp < $ctime ) {
 467                                 unset( $cBackoffs[$type] );
 468                         }
 469                 }
 470                 ftruncate( $handle, 0 );
 471                 fwrite( $handle, json_encode( $cBackoffs ) );
 472                 flock( $handle, LOCK_UN );
 473                 fclose( $handle );
 474
 475                 $deltas = [];
 476
 477                 return $cBackoffs;
 478         }
 479
 480         /**
 481          * Make sure that this script is not too close to the memory usage limit.
 482          * It is better to die in between jobs than OOM right in the middle of one.
 483          * @return bool
 484          */
 485         private function checkMemoryOK() {
 486                 static $maxBytes = null;
 487                 if ( $maxBytes === null ) {
 488                         $m = [];
 489                         if ( preg_match( '!^(\d+)(k|m|g|)$!i', ini_get( 'memory_limit' ), $m ) ) {
 490                                 list( , $num, $unit ) = $m;
 491                                 $conv = [ 'g' => 1073741824, 'm' => 1048576, 'k' => 1024, '' => 1 ];
 492                                 $maxBytes = $num * $conv[strtolower( $unit )];
 493                         } else {
 494                                 $maxBytes = 0;
 495                         }
 496                 }
 497                 $usedBytes = memory_get_usage();
 498                 if ( $maxBytes && $usedBytes >= 0.95 * $maxBytes ) {
 499                         $msg = "Detected excessive memory usage ({used_bytes}/{max_bytes}).";
 500                         $this->logger->error( $msg, [
 501                                 'used_bytes' => $usedBytes,
 502                                 'max_bytes' => $maxBytes,
 503                         ] );
 504
 505                         $msg = "Detected excessive memory usage ($usedBytes/$maxBytes).";
 506                         $this->debugCallback( $msg );
 507
 508                         return false;
 509                 }
 510
 511                 return true;
 512         }
 513
 514         /**
 515          * Log the job message
 516          * @param string $msg The message to log
 517          */
 518         private function debugCallback( $msg ) {
 519                 if ( $this->debug ) {
 520                         call_user_func_array( $this->debug, [ wfTimestamp( TS_DB ) . " $msg\n" ] );
 521                 }
 522         }
 523
 524         /**
 525          * Issue a commit on all masters who are currently in a transaction and have
 526          * made changes to the database. It also supports sometimes waiting for the
 527          * local wiki's replica DBs to catch up. See the documentation for
 528          * $wgJobSerialCommitThreshold for more.
 529          *
 530          * @param LBFactory $lbFactory
 531          * @param Job $job
 532          * @param string $fnameTrxOwner
 533          * @throws DBError
 534          */
 535         private function commitMasterChanges( LBFactory $lbFactory, Job $job, $fnameTrxOwner ) {
 536                 $syncThreshold = $this->config->get( 'JobSerialCommitThreshold' );
 537
 538                 $time = false;
 539                 $lb = $lbFactory->getMainLB( wfWikiID() );
 540                 if ( $syncThreshold !== false && $lb->getServerCount() > 1 ) {
 541                         // Generally, there is one master connection to the local DB
 542                         $dbwSerial = $lb->getAnyOpenConnection( $lb->getWriterIndex() );
 543                         // We need natively blocking fast locks
 544                         if ( $dbwSerial && $dbwSerial->namedLocksEnqueue() ) {
 545                                 $time = $dbwSerial->pendingWriteQueryDuration( $dbwSerial::ESTIMATE_DB_APPLY );
 546                                 if ( $time < $syncThreshold ) {
 547                                         $dbwSerial = false;
 548                                 }
 549                         } else {
 550                                 $dbwSerial = false;
 551                         }
 552                 } else {
 553                         // There are no replica DBs or writes are all to foreign DB (we don't handle that)
 554                         $dbwSerial = false;
 555                 }
 556
 557                 if ( !$dbwSerial ) {
 558                         $lbFactory->commitMasterChanges(
 559                                 $fnameTrxOwner,
 560                                 // Abort if any transaction was too big
 561                                 [ 'maxWriteDuration' => $this->config->get( 'MaxJobDBWriteDuration' ) ]
 562                         );
 563
 564                         return;
 565                 }
 566
 567                 $ms = intval( 1000 * $time );
 568
 569                 $msg = $job->toString() . " COMMIT ENQUEUED [{job_commit_write_ms}ms of writes]";
 570                 $this->logger->info( $msg, [
 571                         'job_type' => $job->getType(),
 572                         'job_commit_write_ms' => $ms,
 573                 ] );
 574
 575                 $msg = $job->toString() . " COMMIT ENQUEUED [{$ms}ms of writes]";
 576                 $this->debugCallback( $msg );
 577
 578                 // Wait for an exclusive lock to commit
 579                 if ( !$dbwSerial->lock( 'jobrunner-serial-commit', __METHOD__, 30 ) ) {
 580                         // This will trigger a rollback in the main loop
 581                         throw new DBError( $dbwSerial, "Timed out waiting on commit queue." );
 582                 }
 583                 $unlocker = new ScopedCallback( function () use ( $dbwSerial ) {
 584                         $dbwSerial->unlock( 'jobrunner-serial-commit', __METHOD__ );
 585                 } );
 586
 587                 // Wait for the replica DBs to catch up
 588                 $pos = $lb->getMasterPos();
 589                 if ( $pos ) {
 590                         $lb->waitForAll( $pos );
 591                 }
 592
 593                 // Actually commit the DB master changes
 594                 $lbFactory->commitMasterChanges(
 595                         $fnameTrxOwner,
 596                         // Abort if any transaction was too big
 597                         [ 'maxWriteDuration' => $this->config->get( 'MaxJobDBWriteDuration' ) ]
 598                 );
 599                 ScopedCallback::consume( $unlocker );
 600         }
 601 }