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 Psr\Log\LoggerAwareInterface;
  25 use Psr\Log\LoggerInterface;
  26
  27 /**
  28  * Job queue runner utility methods
  29  *
  30  * @ingroup JobQueue
  31  * @since 1.24
  32  */
  33 class JobRunner implements LoggerAwareInterface {
  34         /** @var callable|null Debug output handler */
  35         protected $debug;
  36
  37         /**
  38          * @param callable $debug Optional debug output handler
  39          */
  40         public function setDebugHandler( $debug ) {
  41                 $this->debug = $debug;
  42         }
  43
  44         /**
  45          * @var LoggerInterface $logger
  46          */
  47         protected $logger;
  48
  49         /**
  50          * @param LoggerInterface $logger
  51          */
  52         public function setLogger( LoggerInterface $logger ) {
  53                 $this->logger = $logger;
  54         }
  55
  56         /**
  57          * @param LoggerInterface $logger
  58          */
  59         public function __construct( LoggerInterface $logger = null ) {
  60                 if ( $logger === null ) {
  61                         $logger = MWLoggerFactory::getInstance( 'runJobs' );
  62                 }
  63                 $this->setLogger( $logger );
  64         }
  65
  66         /**
  67          * Run jobs of the specified number/type for the specified time
  68          *
  69          * The response map has a 'job' field that lists status of each job, including:
  70          *   - type   : the job type
  71          *   - status : ok/failed
  72          *   - error  : any error message string
  73          *   - time   : the job run time in ms
  74          * The response map also has:
  75          *   - backoffs : the (job type => seconds) map of backoff times
  76          *   - elapsed  : the total time spent running tasks in ms
  77          *   - reached  : the reason the script finished, one of (none-ready, job-limit, time-limit)
  78          *
  79          * This method outputs status information only if a debug handler was set.
  80          * Any exceptions are caught and logged, but are not reported as output.
  81          *
  82          * @param array $options Map of parameters:
  83          *    - type     : the job type (or false for the default types)
  84          *    - maxJobs  : maximum number of jobs to run
  85          *    - maxTime  : maximum time in seconds before stopping
  86          *    - throttle : whether to respect job backoff configuration
  87          * @return array Summary response that can easily be JSON serialized
  88          */
  89         public function run( array $options ) {
  90                 global $wgJobClasses;
  91
  92                 $response = array( 'jobs' => array(), 'reached' => 'none-ready' );
  93
  94                 $type = isset( $options['type'] ) ? $options['type'] : false;
  95                 $maxJobs = isset( $options['maxJobs'] ) ? $options['maxJobs'] : false;
  96                 $maxTime = isset( $options['maxTime'] ) ? $options['maxTime'] : false;
  97                 $noThrottle = isset( $options['throttle'] ) && !$options['throttle'];
  98
  99                 if ( $type !== false && !isset( $wgJobClasses[$type] ) ) {
 100                         $response['reached'] = 'none-possible';
 101                         return $response;
 102                 }
 103
 104                 $group = JobQueueGroup::singleton();
 105                 // Handle any required periodic queue maintenance
 106                 $count = $group->executeReadyPeriodicTasks();
 107                 if ( $count > 0 ) {
 108                         $msg = "Executed $count periodic queue task(s).";
 109                         $this->logger->debug( $msg );
 110                         $this->debugCallback( $msg );
 111                 }
 112
 113                 // Bail out if in read-only mode
 114                 if ( wfReadOnly() ) {
 115                         $response['reached'] = 'read-only';
 116                         return $response;
 117                 }
 118
 119                 // Bail out if there is too much DB lag
 120                 list( , $maxLag ) = wfGetLBFactory()->getMainLB( wfWikiID() )->getMaxLag();
 121                 if ( $maxLag >= 5 ) {
 122                         $response['reached'] = 'slave-lag-limit';
 123                         return $response;
 124                 }
 125
 126                 // Flush any pending DB writes for sanity
 127                 wfGetLBFactory()->commitMasterChanges();
 128
 129                 // Some jobs types should not run until a certain timestamp
 130                 $backoffs = array(); // map of (type => UNIX expiry)
 131                 $backoffDeltas = array(); // map of (type => seconds)
 132                 $wait = 'wait'; // block to read backoffs the first time
 133
 134                 $jobsRun = 0;
 135                 $timeMsTotal = 0;
 136                 $flags = JobQueueGroup::USE_CACHE;
 137                 $checkPeriod = 5.0; // seconds
 138                 $checkPhase = mt_rand( 0, 1000 * $checkPeriod ) / 1000; // avoid stampedes
 139                 $startTime = microtime( true ); // time since jobs started running
 140                 $lastTime = microtime( true ) - $checkPhase; // time since last slave check
 141                 do {
 142                         // Sync the persistent backoffs with concurrent runners
 143                         $backoffs = $this->syncBackoffDeltas( $backoffs, $backoffDeltas, $wait );
 144                         $blacklist = $noThrottle ? array() : array_keys( $backoffs );
 145                         $wait = 'nowait'; // less important now
 146
 147                         if ( $type === false ) {
 148                                 $job = $group->pop( JobQueueGroup::TYPE_DEFAULT, $flags, $blacklist );
 149                         } elseif ( in_array( $type, $blacklist ) ) {
 150                                 $job = false; // requested queue in backoff state
 151                         } else {
 152                                 $job = $group->pop( $type ); // job from a single queue
 153                         }
 154
 155                         if ( $job ) { // found a job
 156                                 $jType = $job->getType();
 157
 158                                 // Back off of certain jobs for a while (for throttling and for errors)
 159                                 $ttw = $this->getBackoffTimeToWait( $job );
 160                                 if ( $ttw > 0 ) {
 161                                         // Always add the delta for other runners in case the time running the
 162                                         // job negated the backoff for each individually but not collectively.
 163                                         $backoffDeltas[$jType] = isset( $backoffDeltas[$jType] )
 164                                                 ? $backoffDeltas[$jType] + $ttw
 165                                                 : $ttw;
 166                                         $backoffs = $this->syncBackoffDeltas( $backoffs, $backoffDeltas, $wait );
 167                                 }
 168
 169                                 $msg = $job->toString() . " STARTING";
 170                                 $this->logger->info( $msg );
 171                                 $this->debugCallback( $msg );
 172
 173                                 // Run the job...
 174                                 $jobStartTime = microtime( true );
 175                                 try {
 176                                         ++$jobsRun;
 177                                         $status = $job->run();
 178                                         $error = $job->getLastError();
 179                                         wfGetLBFactory()->commitMasterChanges();
 180                                 } catch ( Exception $e ) {
 181                                         MWExceptionHandler::rollbackMasterChangesAndLog( $e );
 182                                         $status = false;
 183                                         $error = get_class( $e ) . ': ' . $e->getMessage();
 184                                         MWExceptionHandler::logException( $e );
 185                                 }
 186                                 $timeMs = intval( ( microtime( true ) - $jobStartTime ) * 1000 );
 187                                 $timeMsTotal += $timeMs;
 188
 189                                 // Mark the job as done on success or when the job cannot be retried
 190                                 if ( $status !== false || !$job->allowRetries() ) {
 191                                         $group->ack( $job ); // done
 192                                 }
 193
 194                                 // Back off of certain jobs for a while (for throttling and for errors)
 195                                 if ( $status === false && mt_rand( 0, 49 ) == 0 ) {
 196                                         $ttw = max( $ttw, 30 ); // too many errors
 197                                         $backoffDeltas[$jType] = isset( $backoffDeltas[$jType] )
 198                                                 ? $backoffDeltas[$jType] + $ttw
 199                                                 : $ttw;
 200                                 }
 201
 202                                 if ( $status === false ) {
 203                                         $msg = $job->toString() . " t=$timeMs error={$error}";
 204                                         $this->logger->error( $msg );
 205                                         $this->debugCallback( $msg );
 206                                 } else {
 207                                         $msg = $job->toString() . " t=$timeMs good";
 208                                         $this->logger->info( $msg );
 209                                         $this->debugCallback( $msg );
 210                                 }
 211
 212                                 $response['jobs'][] = array(
 213                                         'type'   => $jType,
 214                                         'status' => ( $status === false ) ? 'failed' : 'ok',
 215                                         'error'  => $error,
 216                                         'time'   => $timeMs
 217                                 );
 218
 219                                 // Break out if we hit the job count or wall time limits...
 220                                 if ( $maxJobs && $jobsRun >= $maxJobs ) {
 221                                         $response['reached'] = 'job-limit';
 222                                         break;
 223                                 } elseif ( $maxTime && ( microtime( true ) - $startTime ) > $maxTime ) {
 224                                         $response['reached'] = 'time-limit';
 225                                         break;
 226                                 }
 227
 228                                 // Don't let any of the main DB slaves get backed up.
 229                                 // This only waits for so long before exiting and letting
 230                                 // other wikis in the farm (on different masters) get a chance.
 231                                 $timePassed = microtime( true ) - $lastTime;
 232                                 if ( $timePassed >= 5 || $timePassed < 0 ) {
 233                                         if ( !wfWaitForSlaves( $lastTime, false, '*', 5 ) ) {
 234                                                 $response['reached'] = 'slave-lag-limit';
 235                                                 break;
 236                                         }
 237                                         $lastTime = microtime( true );
 238                                 }
 239                                 // Don't let any queue slaves/backups fall behind
 240                                 if ( $jobsRun > 0 && ( $jobsRun % 100 ) == 0 ) {
 241                                         $group->waitForBackups();
 242                                 }
 243
 244                                 // Bail if near-OOM instead of in a job
 245                                 $this->assertMemoryOK();
 246                         }
 247                 } while ( $job ); // stop when there are no jobs
 248
 249                 // Sync the persistent backoffs for the next runJobs.php pass
 250                 if ( $backoffDeltas ) {
 251                         $this->syncBackoffDeltas( $backoffs, $backoffDeltas, 'wait' );
 252                 }
 253
 254                 $response['backoffs'] = $backoffs;
 255                 $response['elapsed'] = $timeMsTotal;
 256
 257                 return $response;
 258         }
 259
 260         /**
 261          * @param Job $job
 262          * @return int Seconds for this runner to avoid doing more jobs of this type
 263          * @see $wgJobBackoffThrottling
 264          */
 265         private function getBackoffTimeToWait( Job $job ) {
 266                 global $wgJobBackoffThrottling;
 267
 268                 if ( !isset( $wgJobBackoffThrottling[$job->getType()] ) ||
 269                         $job instanceof DuplicateJob // no work was done
 270                 ) {
 271                         return 0; // not throttled
 272                 }
 273
 274                 $itemsPerSecond = $wgJobBackoffThrottling[$job->getType()];
 275                 if ( $itemsPerSecond <= 0 ) {
 276                         return 0; // not throttled
 277                 }
 278
 279                 $seconds = 0;
 280                 if ( $job->workItemCount() > 0 ) {
 281                         $exactSeconds = $job->workItemCount() / $itemsPerSecond;
 282                         // use randomized rounding
 283                         $seconds = floor( $exactSeconds );
 284                         $remainder = $exactSeconds - $seconds;
 285                         $seconds += ( mt_rand() / mt_getrandmax() < $remainder ) ? 1 : 0;
 286                 }
 287
 288                 return (int)$seconds;
 289         }
 290
 291         /**
 292          * Get the previous backoff expiries from persistent storage
 293          * On I/O or lock acquisition failure this returns the original $backoffs.
 294          *
 295          * @param array $backoffs Map of (job type => UNIX timestamp)
 296          * @param string $mode Lock wait mode - "wait" or "nowait"
 297          * @return array Map of (job type => backoff expiry timestamp)
 298          */
 299         private function loadBackoffs( array $backoffs, $mode = 'wait' ) {
 300
 301                 $file = wfTempDir() . '/mw-runJobs-backoffs.json';
 302                 if ( is_file( $file ) ) {
 303                         $noblock = ( $mode === 'nowait' ) ? LOCK_NB : 0;
 304                         $handle = fopen( $file, 'rb' );
 305                         if ( !flock( $handle, LOCK_SH | $noblock ) ) {
 306                                 fclose( $handle );
 307                                 return $backoffs; // don't wait on lock
 308                         }
 309                         $content = stream_get_contents( $handle );
 310                         flock( $handle, LOCK_UN );
 311                         fclose( $handle );
 312                         $ctime = microtime( true );
 313                         $cBackoffs = json_decode( $content, true ) ?: array();
 314                         foreach ( $cBackoffs as $type => $timestamp ) {
 315                                 if ( $timestamp < $ctime ) {
 316                                         unset( $cBackoffs[$type] );
 317                                 }
 318                         }
 319                 } else {
 320                         $cBackoffs = array();
 321                 }
 322
 323                 return $cBackoffs;
 324         }
 325
 326         /**
 327          * Merge the current backoff expiries from persistent storage
 328          *
 329          * The $deltas map is set to an empty array on success.
 330          * On I/O or lock acquisition failure this returns the original $backoffs.
 331          *
 332          * @param array $backoffs Map of (job type => UNIX timestamp)
 333          * @param array $deltas Map of (job type => seconds)
 334          * @param string $mode Lock wait mode - "wait" or "nowait"
 335          * @return array The new backoffs account for $backoffs and the latest file data
 336          */
 337         private function syncBackoffDeltas( array $backoffs, array &$deltas, $mode = 'wait' ) {
 338
 339                 if ( !$deltas ) {
 340                         return $this->loadBackoffs( $backoffs, $mode );
 341                 }
 342
 343                 $noblock = ( $mode === 'nowait' ) ? LOCK_NB : 0;
 344                 $file = wfTempDir() . '/mw-runJobs-backoffs.json';
 345                 $handle = fopen( $file, 'wb+' );
 346                 if ( !flock( $handle, LOCK_EX | $noblock ) ) {
 347                         fclose( $handle );
 348                         return $backoffs; // don't wait on lock
 349                 }
 350                 $ctime = microtime( true );
 351                 $content = stream_get_contents( $handle );
 352                 $cBackoffs = json_decode( $content, true ) ?: array();
 353                 foreach ( $deltas as $type => $seconds ) {
 354                         $cBackoffs[$type] = isset( $cBackoffs[$type] ) && $cBackoffs[$type] >= $ctime
 355                                 ? $cBackoffs[$type] + $seconds
 356                                 : $ctime + $seconds;
 357                 }
 358                 foreach ( $cBackoffs as $type => $timestamp ) {
 359                         if ( $timestamp < $ctime ) {
 360                                 unset( $cBackoffs[$type] );
 361                         }
 362                 }
 363                 ftruncate( $handle, 0 );
 364                 fwrite( $handle, json_encode( $cBackoffs ) );
 365                 flock( $handle, LOCK_UN );
 366                 fclose( $handle );
 367
 368                 $deltas = array();
 369
 370                 return $cBackoffs;
 371         }
 372
 373         /**
 374          * Make sure that this script is not too close to the memory usage limit.
 375          * It is better to die in between jobs than OOM right in the middle of one.
 376          * @throws MWException
 377          */
 378         private function assertMemoryOK() {
 379                 static $maxBytes = null;
 380                 if ( $maxBytes === null ) {
 381                         $m = array();
 382                         if ( preg_match( '!^(\d+)(k|m|g|)$!i', ini_get( 'memory_limit' ), $m ) ) {
 383                                 list( , $num, $unit ) = $m;
 384                                 $conv = array( 'g' => 1073741824, 'm' => 1048576, 'k' => 1024, '' => 1 );
 385                                 $maxBytes = $num * $conv[strtolower( $unit )];
 386                         } else {
 387                                 $maxBytes = 0;
 388                         }
 389                 }
 390                 $usedBytes = memory_get_usage();
 391                 if ( $maxBytes && $usedBytes >= 0.95 * $maxBytes ) {
 392                         throw new MWException( "Detected excessive memory usage ($usedBytes/$maxBytes)." );
 393                 }
 394         }
 395
 396         /**
 397          * Log the job message
 398          * @param string $msg The message to log
 399          */
 400         private function debugCallback( $msg ) {
 401                 if ( $this->debug ) {
 402                         call_user_func_array( $this->debug, array( wfTimestamp( TS_DB ) . " $msg\n" ) );
 403                 }
 404         }
 405 }