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