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