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