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