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