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                 $jobsPopped = 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                                 $popTime = time();
 161                                 $jType = $job->getType();
 162
 163                                 // Back off of certain jobs for a while (for throttling and for errors)
 164                                 $ttw = $this->getBackoffTimeToWait( $job );
 165                                 if ( $ttw > 0 ) {
 166                                         // Always add the delta for other runners in case the time running the
 167                                         // job negated the backoff for each individually but not collectively.
 168                                         $backoffDeltas[$jType] = isset( $backoffDeltas[$jType] )
 169                                                 ? $backoffDeltas[$jType] + $ttw
 170                                                 : $ttw;
 171                                         $backoffs = $this->syncBackoffDeltas( $backoffs, $backoffDeltas, $wait );
 172                                 }
 173
 174                                 $msg = $job->toString() . " STARTING";
 175                                 $this->logger->debug( $msg );
 176                                 $this->debugCallback( $msg );
 177
 178                                 // Run the job...
 179                                 $psection = $profiler->scopedProfileIn( __METHOD__ . '-' . $jType );
 180                                 $jobStartTime = microtime( true );
 181                                 try {
 182                                         ++$jobsPopped;
 183                                         $status = $job->run();
 184                                         $error = $job->getLastError();
 185                                         $this->commitMasterChanges( $job );
 186
 187                                         DeferredUpdates::doUpdates();
 188                                         $this->commitMasterChanges( $job );
 189                                 } catch ( Exception $e ) {
 190                                         MWExceptionHandler::rollbackMasterChangesAndLog( $e );
 191                                         $status = false;
 192                                         $error = get_class( $e ) . ': ' . $e->getMessage();
 193                                         MWExceptionHandler::logException( $e );
 194                                 }
 195                                 // Commit all outstanding connections that are in a transaction
 196                                 // to get a fresh repeatable read snapshot on every connection.
 197                                 wfGetLBFactory()->commitAll();
 198                                 $timeMs = intval( ( microtime( true ) - $jobStartTime ) * 1000 );
 199                                 $timeMsTotal += $timeMs;
 200                                 $profiler->scopedProfileOut( $psection );
 201
 202                                 if ( $job->getQueuedTimestamp() ) {
 203                                         // Record time to run for the job type
 204                                         $stats->timing( "job-pickuptime-$jType",
 205                                                 $popTime - $job->getQueuedTimestamp() );
 206                                 }
 207
 208                                 // Mark the job as done on success or when the job cannot be retried
 209                                 if ( $status !== false || !$job->allowRetries() ) {
 210                                         $group->ack( $job ); // done
 211                                 }
 212
 213                                 // Back off of certain jobs for a while (for throttling and for errors)
 214                                 if ( $status === false && mt_rand( 0, 49 ) == 0 ) {
 215                                         $ttw = max( $ttw, 30 ); // too many errors
 216                                         $backoffDeltas[$jType] = isset( $backoffDeltas[$jType] )
 217                                                 ? $backoffDeltas[$jType] + $ttw
 218                                                 : $ttw;
 219                                 }
 220
 221                                 if ( $status === false ) {
 222                                         $msg = $job->toString() . " t=$timeMs error={$error}";
 223                                         $this->logger->error( $msg );
 224                                         $this->debugCallback( $msg );
 225                                 } else {
 226                                         $msg = $job->toString() . " t=$timeMs good";
 227                                         $this->logger->info( $msg );
 228                                         $this->debugCallback( $msg );
 229                                 }
 230
 231                                 $response['jobs'][] = array(
 232                                         'type'   => $jType,
 233                                         'status' => ( $status === false ) ? 'failed' : 'ok',
 234                                         'error'  => $error,
 235                                         'time'   => $timeMs
 236                                 );
 237
 238                                 // Break out if we hit the job count or wall time limits...
 239                                 if ( $maxJobs && $jobsPopped >= $maxJobs ) {
 240                                         $response['reached'] = 'job-limit';
 241                                         break;
 242                                 } elseif ( $maxTime && ( microtime( true ) - $startTime ) > $maxTime ) {
 243                                         $response['reached'] = 'time-limit';
 244                                         break;
 245                                 }
 246
 247                                 // Don't let any of the main DB slaves get backed up.
 248                                 // This only waits for so long before exiting and letting
 249                                 // other wikis in the farm (on different masters) get a chance.
 250                                 $timePassed = microtime( true ) - $lastCheckTime;
 251                                 if ( $timePassed >= $checkLagPeriod || $timePassed < 0 ) {
 252                                         if ( !wfWaitForSlaves( $lastCheckTime, false, '*', $maxAllowedLag ) ) {
 253                                                 $response['reached'] = 'slave-lag-limit';
 254                                                 break;
 255                                         }
 256                                         $lastCheckTime = microtime( true );
 257                                 }
 258                                 // Don't let any queue slaves/backups fall behind
 259                                 if ( $jobsPopped > 0 && ( $jobsPopped % 100 ) == 0 ) {
 260                                         $group->waitForBackups();
 261                                 }
 262
 263                                 // Bail if near-OOM instead of in a job
 264                                 $this->assertMemoryOK();
 265                         }
 266                 } while ( $job ); // stop when there are no jobs
 267
 268                 // Sync the persistent backoffs for the next runJobs.php pass
 269                 if ( $backoffDeltas ) {
 270                         $this->syncBackoffDeltas( $backoffs, $backoffDeltas, 'wait' );
 271                 }
 272
 273                 $response['backoffs'] = $backoffs;
 274                 $response['elapsed'] = $timeMsTotal;
 275
 276                 return $response;
 277         }
 278
 279         /**
 280          * @param Job $job
 281          * @return int Seconds for this runner to avoid doing more jobs of this type
 282          * @see $wgJobBackoffThrottling
 283          */
 284         private function getBackoffTimeToWait( Job $job ) {
 285                 global $wgJobBackoffThrottling;
 286
 287                 if ( !isset( $wgJobBackoffThrottling[$job->getType()] ) ||
 288                         $job instanceof DuplicateJob // no work was done
 289                 ) {
 290                         return 0; // not throttled
 291                 }
 292
 293                 $itemsPerSecond = $wgJobBackoffThrottling[$job->getType()];
 294                 if ( $itemsPerSecond <= 0 ) {
 295                         return 0; // not throttled
 296                 }
 297
 298                 $seconds = 0;
 299                 if ( $job->workItemCount() > 0 ) {
 300                         $exactSeconds = $job->workItemCount() / $itemsPerSecond;
 301                         // use randomized rounding
 302                         $seconds = floor( $exactSeconds );
 303                         $remainder = $exactSeconds - $seconds;
 304                         $seconds += ( mt_rand() / mt_getrandmax() < $remainder ) ? 1 : 0;
 305                 }
 306
 307                 return (int)$seconds;
 308         }
 309
 310         /**
 311          * Get the previous backoff expiries from persistent storage
 312          * On I/O or lock acquisition failure this returns the original $backoffs.
 313          *
 314          * @param array $backoffs Map of (job type => UNIX timestamp)
 315          * @param string $mode Lock wait mode - "wait" or "nowait"
 316          * @return array Map of (job type => backoff expiry timestamp)
 317          */
 318         private function loadBackoffs( array $backoffs, $mode = 'wait' ) {
 319                 $file = wfTempDir() . '/mw-runJobs-backoffs.json';
 320                 if ( is_file( $file ) ) {
 321                         $noblock = ( $mode === 'nowait' ) ? LOCK_NB : 0;
 322                         $handle = fopen( $file, 'rb' );
 323                         if ( !flock( $handle, LOCK_SH | $noblock ) ) {
 324                                 fclose( $handle );
 325                                 return $backoffs; // don't wait on lock
 326                         }
 327                         $content = stream_get_contents( $handle );
 328                         flock( $handle, LOCK_UN );
 329                         fclose( $handle );
 330                         $ctime = microtime( true );
 331                         $cBackoffs = json_decode( $content, true ) ?: array();
 332                         foreach ( $cBackoffs as $type => $timestamp ) {
 333                                 if ( $timestamp < $ctime ) {
 334                                         unset( $cBackoffs[$type] );
 335                                 }
 336                         }
 337                 } else {
 338                         $cBackoffs = array();
 339                 }
 340
 341                 return $cBackoffs;
 342         }
 343
 344         /**
 345          * Merge the current backoff expiries from persistent storage
 346          *
 347          * The $deltas map is set to an empty array on success.
 348          * On I/O or lock acquisition failure this returns the original $backoffs.
 349          *
 350          * @param array $backoffs Map of (job type => UNIX timestamp)
 351          * @param array $deltas Map of (job type => seconds)
 352          * @param string $mode Lock wait mode - "wait" or "nowait"
 353          * @return array The new backoffs account for $backoffs and the latest file data
 354          */
 355         private function syncBackoffDeltas( array $backoffs, array &$deltas, $mode = 'wait' ) {
 356                 if ( !$deltas ) {
 357                         return $this->loadBackoffs( $backoffs, $mode );
 358                 }
 359
 360                 $noblock = ( $mode === 'nowait' ) ? LOCK_NB : 0;
 361                 $file = wfTempDir() . '/mw-runJobs-backoffs.json';
 362                 $handle = fopen( $file, 'wb+' );
 363                 if ( !flock( $handle, LOCK_EX | $noblock ) ) {
 364                         fclose( $handle );
 365                         return $backoffs; // don't wait on lock
 366                 }
 367                 $ctime = microtime( true );
 368                 $content = stream_get_contents( $handle );
 369                 $cBackoffs = json_decode( $content, true ) ?: array();
 370                 foreach ( $deltas as $type => $seconds ) {
 371                         $cBackoffs[$type] = isset( $cBackoffs[$type] ) && $cBackoffs[$type] >= $ctime
 372                                 ? $cBackoffs[$type] + $seconds
 373                                 : $ctime + $seconds;
 374                 }
 375                 foreach ( $cBackoffs as $type => $timestamp ) {
 376                         if ( $timestamp < $ctime ) {
 377                                 unset( $cBackoffs[$type] );
 378                         }
 379                 }
 380                 ftruncate( $handle, 0 );
 381                 fwrite( $handle, json_encode( $cBackoffs ) );
 382                 flock( $handle, LOCK_UN );
 383                 fclose( $handle );
 384
 385                 $deltas = array();
 386
 387                 return $cBackoffs;
 388         }
 389
 390         /**
 391          * Make sure that this script is not too close to the memory usage limit.
 392          * It is better to die in between jobs than OOM right in the middle of one.
 393          * @throws MWException
 394          */
 395         private function assertMemoryOK() {
 396                 static $maxBytes = null;
 397                 if ( $maxBytes === null ) {
 398                         $m = array();
 399                         if ( preg_match( '!^(\d+)(k|m|g|)$!i', ini_get( 'memory_limit' ), $m ) ) {
 400                                 list( , $num, $unit ) = $m;
 401                                 $conv = array( 'g' => 1073741824, 'm' => 1048576, 'k' => 1024, '' => 1 );
 402                                 $maxBytes = $num * $conv[strtolower( $unit )];
 403                         } else {
 404                                 $maxBytes = 0;
 405                         }
 406                 }
 407                 $usedBytes = memory_get_usage();
 408                 if ( $maxBytes && $usedBytes >= 0.95 * $maxBytes ) {
 409                         throw new MWException( "Detected excessive memory usage ($usedBytes/$maxBytes)." );
 410                 }
 411         }
 412
 413         /**
 414          * Log the job message
 415          * @param string $msg The message to log
 416          */
 417         private function debugCallback( $msg ) {
 418                 if ( $this->debug ) {
 419                         call_user_func_array( $this->debug, array( wfTimestamp( TS_DB ) . " $msg\n" ) );
 420                 }
 421         }
 422
 423         /**
 424          * Issue a commit on all masters who are currently in a transaction and have
 425          * made changes to the database. It also supports sometimes waiting for the
 426          * local wiki's slaves to catch up. See the documentation for
 427          * $wgJobSerialCommitThreshold for more.
 428          *
 429          * @param Job $job
 430          * @throws DBError
 431          */
 432         private function commitMasterChanges( Job $job ) {
 433                 global $wgJobSerialCommitThreshold;
 434
 435                 $lb = wfGetLB( wfWikiID() );
 436                 if ( $wgJobSerialCommitThreshold !== false ) {
 437                         // Generally, there is one master connection to the local DB
 438                         $dbwSerial = $lb->getAnyOpenConnection( $lb->getWriterIndex() );
 439                 } else {
 440                         $dbwSerial = false;
 441                 }
 442
 443                 if ( !$dbwSerial
 444                         || !$dbwSerial->namedLocksEnqueue()
 445                         || $dbwSerial->pendingWriteQueryDuration() < $wgJobSerialCommitThreshold
 446                 ) {
 447                         // Writes are all to foreign DBs, named locks don't form queues,
 448                         // or $wgJobSerialCommitThreshold is not reached; commit changes now
 449                         wfGetLBFactory()->commitMasterChanges();
 450                         return;
 451                 }
 452
 453                 $ms = intval( 1000 * $dbwSerial->pendingWriteQueryDuration() );
 454                 $msg = $job->toString() . " COMMIT ENQUEUED [{$ms}ms of writes]";
 455                 $this->logger->info( $msg );
 456                 $this->debugCallback( $msg );
 457
 458                 // Wait for an exclusive lock to commit
 459                 if ( !$dbwSerial->lock( 'jobrunner-serial-commit', __METHOD__, 30 ) ) {
 460                         // This will trigger a rollback in the main loop
 461                         throw new DBError( $dbwSerial, "Timed out waiting on commit queue." );
 462                 }
 463                 // Wait for the generic slave to catch up
 464                 $pos = $lb->getMasterPos();
 465                 if ( $pos ) {
 466                         $lb->waitForOne( $pos );
 467                 }
 468
 469                 $fname = __METHOD__;
 470                 // Re-ping all masters with transactions. This throws DBError if some
 471                 // connection died while waiting on locks/slaves, triggering a rollback.
 472                 wfGetLBFactory()->forEachLB( function( LoadBalancer $lb ) use ( $fname ) {
 473                         $lb->forEachOpenConnection( function( DatabaseBase $conn ) use ( $fname ) {
 474                                 if ( $conn->writesOrCallbacksPending() ) {
 475                                         $conn->query( "SELECT 1", $fname );
 476                                 }
 477                         } );
 478                 } );
 479
 480                 // Actually commit the DB master changes
 481                 wfGetLBFactory()->commitMasterChanges();
 482
 483                 // Release the lock
 484                 $dbwSerial->unlock( 'jobrunner-serial-commit', __METHOD__ );
 485         }
 486 }