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