3 * Job queue runner utility methods
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.
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.
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
24 use MediaWiki\MediaWikiServices
;
25 use MediaWiki\Logger\LoggerFactory
;
26 use Liuggio\StatsdClient\Factory\StatsdDataFactory
;
27 use Psr\Log\LoggerAwareInterface
;
28 use Psr\Log\LoggerInterface
;
29 use Wikimedia\ScopedCallback
;
30 use Wikimedia\Rdbms\LBFactory
;
31 use Wikimedia\Rdbms\DBError
;
32 use Wikimedia\Rdbms\DBReplicationWaitError
;
35 * Job queue runner utility methods
40 class JobRunner
implements LoggerAwareInterface
{
41 /** @var callable|null Debug output handler */
45 * @var LoggerInterface $logger
49 const MAX_ALLOWED_LAG
= 3; // abort if more than this much DB lag is present
50 const LAG_CHECK_PERIOD
= 1.0; // check replica DB lag this many seconds
51 const ERROR_BACKOFF_TTL
= 1; // seconds to back off a queue due to errors
52 const READONLY_BACKOFF_TTL
= 30; // seconds to back off a queue due to read-only errors
55 * @param callable $debug Optional debug output handler
57 public function setDebugHandler( $debug ) {
58 $this->debug
= $debug;
62 * @param LoggerInterface $logger
65 public function setLogger( LoggerInterface
$logger ) {
66 $this->logger
= $logger;
70 * @param LoggerInterface $logger
72 public function __construct( LoggerInterface
$logger = null ) {
73 if ( $logger === null ) {
74 $logger = LoggerFactory
::getInstance( 'runJobs' );
76 $this->setLogger( $logger );
80 * Run jobs of the specified number/type for the specified time
82 * The response map has a 'job' field that lists status of each job, including:
83 * - type : the job type
84 * - status : ok/failed
85 * - error : any error message string
86 * - time : the job run time in ms
87 * The response map also has:
88 * - backoffs : the (job type => seconds) map of backoff times
89 * - elapsed : the total time spent running tasks in ms
90 * - reached : the reason the script finished, one of (none-ready, job-limit, time-limit,
93 * This method outputs status information only if a debug handler was set.
94 * Any exceptions are caught and logged, but are not reported as output.
96 * @param array $options Map of parameters:
97 * - type : the job type (or false for the default types)
98 * - maxJobs : maximum number of jobs to run
99 * - maxTime : maximum time in seconds before stopping
100 * - throttle : whether to respect job backoff configuration
101 * @return array Summary response that can easily be JSON serialized
103 public function run( array $options ) {
104 global $wgJobClasses, $wgTrxProfilerLimits;
106 $response = [ 'jobs' => [], 'reached' => 'none-ready' ];
108 $type = isset( $options['type'] ) ?
$options['type'] : false;
109 $maxJobs = isset( $options['maxJobs'] ) ?
$options['maxJobs'] : false;
110 $maxTime = isset( $options['maxTime'] ) ?
$options['maxTime'] : false;
111 $noThrottle = isset( $options['throttle'] ) && !$options['throttle'];
113 // Bail if job type is invalid
114 if ( $type !== false && !isset( $wgJobClasses[$type] ) ) {
115 $response['reached'] = 'none-possible';
118 // Bail out if DB is in read-only mode
119 if ( wfReadOnly() ) {
120 $response['reached'] = 'read-only';
124 $lbFactory = MediaWikiServices
::getInstance()->getDBLoadBalancerFactory();
125 // Bail out if there is too much DB lag.
126 // This check should not block as we want to try other wiki queues.
127 list( , $maxLag ) = $lbFactory->getMainLB( wfWikiID() )->getMaxLag();
128 if ( $maxLag >= self
::MAX_ALLOWED_LAG
) {
129 $response['reached'] = 'replica-lag-limit';
133 // Flush any pending DB writes for sanity
134 $lbFactory->commitAll( __METHOD__
);
136 // Catch huge single updates that lead to replica DB lag
137 $trxProfiler = Profiler
::instance()->getTransactionProfiler();
138 $trxProfiler->setLogger( LoggerFactory
::getInstance( 'DBPerformance' ) );
139 $trxProfiler->setExpectations( $wgTrxProfilerLimits['JobRunner'], __METHOD__
);
141 // Some jobs types should not run until a certain timestamp
142 $backoffs = []; // map of (type => UNIX expiry)
143 $backoffDeltas = []; // map of (type => seconds)
144 $wait = 'wait'; // block to read backoffs the first time
146 $group = JobQueueGroup
::singleton();
147 $stats = MediaWikiServices
::getInstance()->getStatsdDataFactory();
150 $startTime = microtime( true ); // time since jobs started running
151 $lastCheckTime = 1; // timestamp of last replica DB check
153 // Sync the persistent backoffs with concurrent runners
154 $backoffs = $this->syncBackoffDeltas( $backoffs, $backoffDeltas, $wait );
155 $blacklist = $noThrottle ?
[] : array_keys( $backoffs );
156 $wait = 'nowait'; // less important now
158 if ( $type === false ) {
160 JobQueueGroup
::TYPE_DEFAULT
,
161 JobQueueGroup
::USE_CACHE
,
164 } elseif ( in_array( $type, $blacklist ) ) {
165 $job = false; // requested queue in backoff state
167 $job = $group->pop( $type ); // job from a single queue
169 $lbFactory->commitMasterChanges( __METHOD__
); // flush any JobQueueDB writes
171 if ( $job ) { // found a job
174 $jType = $job->getType();
176 WebRequest
::overrideRequestId( $job->getRequestId() );
178 // Back off of certain jobs for a while (for throttling and for errors)
179 $ttw = $this->getBackoffTimeToWait( $job );
181 // Always add the delta for other runners in case the time running the
182 // job negated the backoff for each individually but not collectively.
183 $backoffDeltas[$jType] = isset( $backoffDeltas[$jType] )
184 ?
$backoffDeltas[$jType] +
$ttw
186 $backoffs = $this->syncBackoffDeltas( $backoffs, $backoffDeltas, $wait );
189 $info = $this->executeJob( $job, $lbFactory, $stats, $popTime );
190 if ( $info['status'] !== false ||
!$job->allowRetries() ) {
191 $group->ack( $job ); // succeeded or job cannot be retried
192 $lbFactory->commitMasterChanges( __METHOD__
); // flush any JobQueueDB writes
195 // Back off of certain jobs for a while (for throttling and for errors)
196 if ( $info['status'] === false && mt_rand( 0, 49 ) == 0 ) {
197 $ttw = max( $ttw, $this->getErrorBackoffTTL( $info['error'] ) );
198 $backoffDeltas[$jType] = isset( $backoffDeltas[$jType] )
199 ?
$backoffDeltas[$jType] +
$ttw
203 $response['jobs'][] = [
205 'status' => ( $info['status'] === false ) ?
'failed' : 'ok',
206 'error' => $info['error'],
207 'time' => $info['timeMs']
209 $timeMsTotal +
= $info['timeMs'];
211 // Break out if we hit the job count or wall time limits...
212 if ( $maxJobs && $jobsPopped >= $maxJobs ) {
213 $response['reached'] = 'job-limit';
215 } elseif ( $maxTime && ( microtime( true ) - $startTime ) > $maxTime ) {
216 $response['reached'] = 'time-limit';
220 // Don't let any of the main DB replica DBs get backed up.
221 // This only waits for so long before exiting and letting
222 // other wikis in the farm (on different masters) get a chance.
223 $timePassed = microtime( true ) - $lastCheckTime;
224 if ( $timePassed >= self
::LAG_CHECK_PERIOD ||
$timePassed < 0 ) {
226 $lbFactory->waitForReplication( [
227 'ifWritesSince' => $lastCheckTime,
228 'timeout' => self
::MAX_ALLOWED_LAG
230 } catch ( DBReplicationWaitError
$e ) {
231 $response['reached'] = 'replica-lag-limit';
234 $lastCheckTime = microtime( true );
236 // Don't let any queue replica DBs/backups fall behind
237 if ( $jobsPopped > 0 && ( $jobsPopped %
100 ) == 0 ) {
238 $group->waitForBackups();
241 // Bail if near-OOM instead of in a job
242 if ( !$this->checkMemoryOK() ) {
243 $response['reached'] = 'memory-limit';
247 } while ( $job ); // stop when there are no jobs
249 // Sync the persistent backoffs for the next runJobs.php pass
250 if ( $backoffDeltas ) {
251 $this->syncBackoffDeltas( $backoffs, $backoffDeltas, 'wait' );
254 $response['backoffs'] = $backoffs;
255 $response['elapsed'] = $timeMsTotal;
261 * @param string $error
262 * @return int TTL in seconds
264 private function getErrorBackoffTTL( $error ) {
265 return strpos( $error, 'DBReadOnlyError' ) !== false
266 ? self
::READONLY_BACKOFF_TTL
267 : self
::ERROR_BACKOFF_TTL
;
272 * @param LBFactory $lbFactory
273 * @param StatsdDataFactory $stats
274 * @param float $popTime
275 * @return array Map of status/error/timeMs
277 private function executeJob( Job
$job, LBFactory
$lbFactory, $stats, $popTime ) {
278 $jType = $job->getType();
279 $msg = $job->toString() . " STARTING";
280 $this->logger
->debug( $msg );
281 $this->debugCallback( $msg );
284 $rssStart = $this->getMaxRssKb();
285 $jobStartTime = microtime( true );
287 $fnameTrxOwner = get_class( $job ) . '::run'; // give run() outer scope
288 $lbFactory->beginMasterChanges( $fnameTrxOwner );
289 $status = $job->run();
290 $error = $job->getLastError();
291 $this->commitMasterChanges( $lbFactory, $job, $fnameTrxOwner );
292 // Run any deferred update tasks; doUpdates() manages transactions itself
293 DeferredUpdates
::doUpdates();
294 // Push lazy jobs added by the job or its deferred udpates
295 JobQueueGroup
::pushLazyJobs();
296 } catch ( Exception
$e ) {
297 MWExceptionHandler
::rollbackMasterChangesAndLog( $e );
299 $error = get_class( $e ) . ': ' . $e->getMessage();
301 // Always attempt to call teardown() even if Job throws exception.
303 $job->teardown( $status );
304 } catch ( Exception
$e ) {
305 MWExceptionHandler
::logException( $e );
308 // Commit all outstanding connections that are in a transaction
309 // to get a fresh repeatable read snapshot on every connection.
310 // Note that jobs are still responsible for handling replica DB lag.
311 $lbFactory->flushReplicaSnapshots( __METHOD__
);
312 // Clear out title cache data from prior snapshots
313 MediaWikiServices
::getInstance()->getLinkCache()->clear();
314 $timeMs = intval( ( microtime( true ) - $jobStartTime ) * 1000 );
315 $rssEnd = $this->getMaxRssKb();
317 // Record how long jobs wait before getting popped
318 $readyTs = $job->getReadyTimestamp();
320 $pickupDelay = max( 0, $popTime - $readyTs );
321 $stats->timing( 'jobqueue.pickup_delay.all', 1000 * $pickupDelay );
322 $stats->timing( "jobqueue.pickup_delay.$jType", 1000 * $pickupDelay );
324 // Record root job age for jobs being run
325 $rootTimestamp = $job->getRootJobParams()['rootJobTimestamp'];
326 if ( $rootTimestamp ) {
327 $age = max( 0, $popTime - wfTimestamp( TS_UNIX
, $rootTimestamp ) );
328 $stats->timing( "jobqueue.pickup_root_age.$jType", 1000 * $age );
330 // Track the execution time for jobs
331 $stats->timing( "jobqueue.run.$jType", $timeMs );
332 // Track RSS increases for jobs (in case of memory leaks)
333 if ( $rssStart && $rssEnd ) {
334 $stats->updateCount( "jobqueue.rss_delta.$jType", $rssEnd - $rssStart );
337 if ( $status === false ) {
338 $msg = $job->toString() . " t=$timeMs error={$error}";
339 $this->logger
->error( $msg );
340 $this->debugCallback( $msg );
342 $msg = $job->toString() . " t=$timeMs good";
343 $this->logger
->info( $msg );
344 $this->debugCallback( $msg );
347 return [ 'status' => $status, 'error' => $error, 'timeMs' => $timeMs ];
351 * @return int|null Max memory RSS in kilobytes
353 private function getMaxRssKb() {
354 $info = wfGetRusage() ?
: [];
355 // see https://linux.die.net/man/2/getrusage
356 return isset( $info['ru_maxrss'] ) ?
(int)$info['ru_maxrss'] : null;
361 * @return int Seconds for this runner to avoid doing more jobs of this type
362 * @see $wgJobBackoffThrottling
364 private function getBackoffTimeToWait( Job
$job ) {
365 global $wgJobBackoffThrottling;
367 if ( !isset( $wgJobBackoffThrottling[$job->getType()] ) ||
368 $job instanceof DuplicateJob
// no work was done
370 return 0; // not throttled
373 $itemsPerSecond = $wgJobBackoffThrottling[$job->getType()];
374 if ( $itemsPerSecond <= 0 ) {
375 return 0; // not throttled
379 if ( $job->workItemCount() > 0 ) {
380 $exactSeconds = $job->workItemCount() / $itemsPerSecond;
381 // use randomized rounding
382 $seconds = floor( $exactSeconds );
383 $remainder = $exactSeconds - $seconds;
384 $seconds +
= ( mt_rand() / mt_getrandmax() < $remainder ) ?
1 : 0;
387 return (int)$seconds;
391 * Get the previous backoff expiries from persistent storage
392 * On I/O or lock acquisition failure this returns the original $backoffs.
394 * @param array $backoffs Map of (job type => UNIX timestamp)
395 * @param string $mode Lock wait mode - "wait" or "nowait"
396 * @return array Map of (job type => backoff expiry timestamp)
398 private function loadBackoffs( array $backoffs, $mode = 'wait' ) {
399 $file = wfTempDir() . '/mw-runJobs-backoffs.json';
400 if ( is_file( $file ) ) {
401 $noblock = ( $mode === 'nowait' ) ? LOCK_NB
: 0;
402 $handle = fopen( $file, 'rb' );
403 if ( !flock( $handle, LOCK_SH |
$noblock ) ) {
405 return $backoffs; // don't wait on lock
407 $content = stream_get_contents( $handle );
408 flock( $handle, LOCK_UN
);
410 $ctime = microtime( true );
411 $cBackoffs = json_decode( $content, true ) ?
: [];
412 foreach ( $cBackoffs as $type => $timestamp ) {
413 if ( $timestamp < $ctime ) {
414 unset( $cBackoffs[$type] );
425 * Merge the current backoff expiries from persistent storage
427 * The $deltas map is set to an empty array on success.
428 * On I/O or lock acquisition failure this returns the original $backoffs.
430 * @param array $backoffs Map of (job type => UNIX timestamp)
431 * @param array $deltas Map of (job type => seconds)
432 * @param string $mode Lock wait mode - "wait" or "nowait"
433 * @return array The new backoffs account for $backoffs and the latest file data
435 private function syncBackoffDeltas( array $backoffs, array &$deltas, $mode = 'wait' ) {
437 return $this->loadBackoffs( $backoffs, $mode );
440 $noblock = ( $mode === 'nowait' ) ? LOCK_NB
: 0;
441 $file = wfTempDir() . '/mw-runJobs-backoffs.json';
442 $handle = fopen( $file, 'wb+' );
443 if ( !flock( $handle, LOCK_EX |
$noblock ) ) {
445 return $backoffs; // don't wait on lock
447 $ctime = microtime( true );
448 $content = stream_get_contents( $handle );
449 $cBackoffs = json_decode( $content, true ) ?
: [];
450 foreach ( $deltas as $type => $seconds ) {
451 $cBackoffs[$type] = isset( $cBackoffs[$type] ) && $cBackoffs[$type] >= $ctime
452 ?
$cBackoffs[$type] +
$seconds
455 foreach ( $cBackoffs as $type => $timestamp ) {
456 if ( $timestamp < $ctime ) {
457 unset( $cBackoffs[$type] );
460 ftruncate( $handle, 0 );
461 fwrite( $handle, json_encode( $cBackoffs ) );
462 flock( $handle, LOCK_UN
);
471 * Make sure that this script is not too close to the memory usage limit.
472 * It is better to die in between jobs than OOM right in the middle of one.
475 private function checkMemoryOK() {
476 static $maxBytes = null;
477 if ( $maxBytes === null ) {
479 if ( preg_match( '!^(\d+)(k|m|g|)$!i', ini_get( 'memory_limit' ), $m ) ) {
480 list( , $num, $unit ) = $m;
481 $conv = [ 'g' => 1073741824, 'm' => 1048576, 'k' => 1024, '' => 1 ];
482 $maxBytes = $num * $conv[strtolower( $unit )];
487 $usedBytes = memory_get_usage();
488 if ( $maxBytes && $usedBytes >= 0.95 * $maxBytes ) {
489 $msg = "Detected excessive memory usage ($usedBytes/$maxBytes).";
490 $this->debugCallback( $msg );
491 $this->logger
->error( $msg );
500 * Log the job message
501 * @param string $msg The message to log
503 private function debugCallback( $msg ) {
504 if ( $this->debug
) {
505 call_user_func_array( $this->debug
, [ wfTimestamp( TS_DB
) . " $msg\n" ] );
510 * Issue a commit on all masters who are currently in a transaction and have
511 * made changes to the database. It also supports sometimes waiting for the
512 * local wiki's replica DBs to catch up. See the documentation for
513 * $wgJobSerialCommitThreshold for more.
515 * @param LBFactory $lbFactory
517 * @param string $fnameTrxOwner
520 private function commitMasterChanges( LBFactory
$lbFactory, Job
$job, $fnameTrxOwner ) {
521 global $wgJobSerialCommitThreshold;
524 $lb = $lbFactory->getMainLB( wfWikiID() );
525 if ( $wgJobSerialCommitThreshold !== false && $lb->getServerCount() > 1 ) {
526 // Generally, there is one master connection to the local DB
527 $dbwSerial = $lb->getAnyOpenConnection( $lb->getWriterIndex() );
528 // We need natively blocking fast locks
529 if ( $dbwSerial && $dbwSerial->namedLocksEnqueue() ) {
530 $time = $dbwSerial->pendingWriteQueryDuration( $dbwSerial::ESTIMATE_DB_APPLY
);
531 if ( $time < $wgJobSerialCommitThreshold ) {
538 // There are no replica DBs or writes are all to foreign DB (we don't handle that)
543 $lbFactory->commitMasterChanges( $fnameTrxOwner );
547 $ms = intval( 1000 * $time );
548 $msg = $job->toString() . " COMMIT ENQUEUED [{$ms}ms of writes]";
549 $this->logger
->info( $msg );
550 $this->debugCallback( $msg );
552 // Wait for an exclusive lock to commit
553 if ( !$dbwSerial->lock( 'jobrunner-serial-commit', __METHOD__
, 30 ) ) {
554 // This will trigger a rollback in the main loop
555 throw new DBError( $dbwSerial, "Timed out waiting on commit queue." );
557 $unlocker = new ScopedCallback( function () use ( $dbwSerial ) {
558 $dbwSerial->unlock( 'jobrunner-serial-commit', __METHOD__
);
561 // Wait for the replica DBs to catch up
562 $pos = $lb->getMasterPos();
564 $lb->waitForAll( $pos );
567 // Actually commit the DB master changes
568 $lbFactory->commitMasterChanges( $fnameTrxOwner );
569 ScopedCallback
::consume( $unlocker );