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
{
43 /** @var callable|null Debug output handler */
47 * @var LoggerInterface $logger
51 const MAX_ALLOWED_LAG
= 3; // abort if more than this much DB lag is present
52 const LAG_CHECK_PERIOD
= 1.0; // check replica DB lag this many seconds
53 const ERROR_BACKOFF_TTL
= 1; // seconds to back off a queue due to errors
54 const READONLY_BACKOFF_TTL
= 30; // seconds to back off a queue due to read-only errors
57 * @param callable $debug Optional debug output handler
59 public function setDebugHandler( $debug ) {
60 $this->debug
= $debug;
64 * @param LoggerInterface $logger
67 public function setLogger( LoggerInterface
$logger ) {
68 $this->logger
= $logger;
72 * @param LoggerInterface $logger
74 public function __construct( LoggerInterface
$logger = null ) {
75 if ( $logger === null ) {
76 $logger = LoggerFactory
::getInstance( 'runJobs' );
78 $this->setLogger( $logger );
79 $this->config
= MediaWikiServices
::getInstance()->getMainConfig();
83 * Run jobs of the specified number/type for the specified time
85 * The response map has a 'job' field that lists status of each job, including:
86 * - type : the job type
87 * - status : ok/failed
88 * - error : any error message string
89 * - time : the job run time in ms
90 * The response map also has:
91 * - backoffs : the (job type => seconds) map of backoff times
92 * - elapsed : the total time spent running tasks in ms
93 * - reached : the reason the script finished, one of (none-ready, job-limit, time-limit,
96 * This method outputs status information only if a debug handler was set.
97 * Any exceptions are caught and logged, but are not reported as output.
99 * @param array $options Map of parameters:
100 * - type : the job type (or false for the default types)
101 * - maxJobs : maximum number of jobs to run
102 * - maxTime : maximum time in seconds before stopping
103 * - throttle : whether to respect job backoff configuration
104 * @return array Summary response that can easily be JSON serialized
106 public function run( array $options ) {
107 $jobClasses = $this->config
->get( 'JobClasses' );
108 $profilerLimits = $this->config
->get( 'TrxProfilerLimits' );
110 $response = [ 'jobs' => [], 'reached' => 'none-ready' ];
112 $type = isset( $options['type'] ) ?
$options['type'] : false;
113 $maxJobs = isset( $options['maxJobs'] ) ?
$options['maxJobs'] : false;
114 $maxTime = isset( $options['maxTime'] ) ?
$options['maxTime'] : false;
115 $noThrottle = isset( $options['throttle'] ) && !$options['throttle'];
117 // Bail if job type is invalid
118 if ( $type !== false && !isset( $jobClasses[$type] ) ) {
119 $response['reached'] = 'none-possible';
122 // Bail out if DB is in read-only mode
123 if ( wfReadOnly() ) {
124 $response['reached'] = 'read-only';
128 $lbFactory = MediaWikiServices
::getInstance()->getDBLoadBalancerFactory();
129 // Bail out if there is too much DB lag.
130 // This check should not block as we want to try other wiki queues.
131 list( , $maxLag ) = $lbFactory->getMainLB( wfWikiID() )->getMaxLag();
132 if ( $maxLag >= self
::MAX_ALLOWED_LAG
) {
133 $response['reached'] = 'replica-lag-limit';
137 // Flush any pending DB writes for sanity
138 $lbFactory->commitAll( __METHOD__
);
140 // Catch huge single updates that lead to replica DB lag
141 $trxProfiler = Profiler
::instance()->getTransactionProfiler();
142 $trxProfiler->setLogger( LoggerFactory
::getInstance( 'DBPerformance' ) );
143 $trxProfiler->setExpectations( $profilerLimits['JobRunner'], __METHOD__
);
145 // Some jobs types should not run until a certain timestamp
146 $backoffs = []; // map of (type => UNIX expiry)
147 $backoffDeltas = []; // map of (type => seconds)
148 $wait = 'wait'; // block to read backoffs the first time
150 $group = JobQueueGroup
::singleton();
151 $stats = MediaWikiServices
::getInstance()->getStatsdDataFactory();
154 $startTime = microtime( true ); // time since jobs started running
155 $lastCheckTime = 1; // timestamp of last replica DB check
157 // Sync the persistent backoffs with concurrent runners
158 $backoffs = $this->syncBackoffDeltas( $backoffs, $backoffDeltas, $wait );
159 $blacklist = $noThrottle ?
[] : array_keys( $backoffs );
160 $wait = 'nowait'; // less important now
162 if ( $type === false ) {
164 JobQueueGroup
::TYPE_DEFAULT
,
165 JobQueueGroup
::USE_CACHE
,
168 } elseif ( in_array( $type, $blacklist ) ) {
169 $job = false; // requested queue in backoff state
171 $job = $group->pop( $type ); // job from a single queue
173 $lbFactory->commitMasterChanges( __METHOD__
); // flush any JobQueueDB writes
175 if ( $job ) { // found a job
178 $jType = $job->getType();
180 WebRequest
::overrideRequestId( $job->getRequestId() );
182 // Back off of certain jobs for a while (for throttling and for errors)
183 $ttw = $this->getBackoffTimeToWait( $job );
185 // Always add the delta for other runners in case the time running the
186 // job negated the backoff for each individually but not collectively.
187 $backoffDeltas[$jType] = isset( $backoffDeltas[$jType] )
188 ?
$backoffDeltas[$jType] +
$ttw
190 $backoffs = $this->syncBackoffDeltas( $backoffs, $backoffDeltas, $wait );
193 $info = $this->executeJob( $job, $lbFactory, $stats, $popTime );
194 if ( $info['status'] !== false ||
!$job->allowRetries() ) {
195 $group->ack( $job ); // succeeded or job cannot be retried
196 $lbFactory->commitMasterChanges( __METHOD__
); // flush any JobQueueDB writes
199 // Back off of certain jobs for a while (for throttling and for errors)
200 if ( $info['status'] === false && mt_rand( 0, 49 ) == 0 ) {
201 $ttw = max( $ttw, $this->getErrorBackoffTTL( $info['error'] ) );
202 $backoffDeltas[$jType] = isset( $backoffDeltas[$jType] )
203 ?
$backoffDeltas[$jType] +
$ttw
207 $response['jobs'][] = [
209 'status' => ( $info['status'] === false ) ?
'failed' : 'ok',
210 'error' => $info['error'],
211 'time' => $info['timeMs']
213 $timeMsTotal +
= $info['timeMs'];
215 // Break out if we hit the job count or wall time limits...
216 if ( $maxJobs && $jobsPopped >= $maxJobs ) {
217 $response['reached'] = 'job-limit';
219 } elseif ( $maxTime && ( microtime( true ) - $startTime ) > $maxTime ) {
220 $response['reached'] = 'time-limit';
224 // Don't let any of the main DB replica DBs get backed up.
225 // This only waits for so long before exiting and letting
226 // other wikis in the farm (on different masters) get a chance.
227 $timePassed = microtime( true ) - $lastCheckTime;
228 if ( $timePassed >= self
::LAG_CHECK_PERIOD ||
$timePassed < 0 ) {
230 $lbFactory->waitForReplication( [
231 'ifWritesSince' => $lastCheckTime,
232 'timeout' => self
::MAX_ALLOWED_LAG
234 } catch ( DBReplicationWaitError
$e ) {
235 $response['reached'] = 'replica-lag-limit';
238 $lastCheckTime = microtime( true );
240 // Don't let any queue replica DBs/backups fall behind
241 if ( $jobsPopped > 0 && ( $jobsPopped %
100 ) == 0 ) {
242 $group->waitForBackups();
245 // Bail if near-OOM instead of in a job
246 if ( !$this->checkMemoryOK() ) {
247 $response['reached'] = 'memory-limit';
251 } while ( $job ); // stop when there are no jobs
253 // Sync the persistent backoffs for the next runJobs.php pass
254 if ( $backoffDeltas ) {
255 $this->syncBackoffDeltas( $backoffs, $backoffDeltas, 'wait' );
258 $response['backoffs'] = $backoffs;
259 $response['elapsed'] = $timeMsTotal;
265 * @param string $error
266 * @return int TTL in seconds
268 private function getErrorBackoffTTL( $error ) {
269 return strpos( $error, 'DBReadOnlyError' ) !== false
270 ? self
::READONLY_BACKOFF_TTL
271 : self
::ERROR_BACKOFF_TTL
;
276 * @param LBFactory $lbFactory
277 * @param StatsdDataFactory $stats
278 * @param float $popTime
279 * @return array Map of status/error/timeMs
281 private function executeJob( Job
$job, LBFactory
$lbFactory, $stats, $popTime ) {
282 $jType = $job->getType();
283 $msg = $job->toString() . " STARTING";
284 $this->logger
->debug( $msg, [
285 'job_type' => $job->getType(),
287 $this->debugCallback( $msg );
290 $rssStart = $this->getMaxRssKb();
291 $jobStartTime = microtime( true );
293 $fnameTrxOwner = get_class( $job ) . '::run'; // give run() outer scope
294 $lbFactory->beginMasterChanges( $fnameTrxOwner );
295 $status = $job->run();
296 $error = $job->getLastError();
297 $this->commitMasterChanges( $lbFactory, $job, $fnameTrxOwner );
298 // Important: this must be the last deferred update added (T100085, T154425)
299 DeferredUpdates
::addCallableUpdate( [ JobQueueGroup
::class, 'pushLazyJobs' ] );
300 // Run any deferred update tasks; doUpdates() manages transactions itself
301 DeferredUpdates
::doUpdates();
302 } catch ( Exception
$e ) {
303 MWExceptionHandler
::rollbackMasterChangesAndLog( $e );
305 $error = get_class( $e ) . ': ' . $e->getMessage();
307 // Always attempt to call teardown() even if Job throws exception.
309 $job->teardown( $status );
310 } catch ( Exception
$e ) {
311 MWExceptionHandler
::logException( $e );
314 // Commit all outstanding connections that are in a transaction
315 // to get a fresh repeatable read snapshot on every connection.
316 // Note that jobs are still responsible for handling replica DB lag.
317 $lbFactory->flushReplicaSnapshots( __METHOD__
);
318 // Clear out title cache data from prior snapshots
319 MediaWikiServices
::getInstance()->getLinkCache()->clear();
320 $timeMs = intval( ( microtime( true ) - $jobStartTime ) * 1000 );
321 $rssEnd = $this->getMaxRssKb();
323 // Record how long jobs wait before getting popped
324 $readyTs = $job->getReadyTimestamp();
326 $pickupDelay = max( 0, $popTime - $readyTs );
327 $stats->timing( 'jobqueue.pickup_delay.all', 1000 * $pickupDelay );
328 $stats->timing( "jobqueue.pickup_delay.$jType", 1000 * $pickupDelay );
330 // Record root job age for jobs being run
331 $rootTimestamp = $job->getRootJobParams()['rootJobTimestamp'];
332 if ( $rootTimestamp ) {
333 $age = max( 0, $popTime - wfTimestamp( TS_UNIX
, $rootTimestamp ) );
334 $stats->timing( "jobqueue.pickup_root_age.$jType", 1000 * $age );
336 // Track the execution time for jobs
337 $stats->timing( "jobqueue.run.$jType", $timeMs );
338 // Track RSS increases for jobs (in case of memory leaks)
339 if ( $rssStart && $rssEnd ) {
340 $stats->updateCount( "jobqueue.rss_delta.$jType", $rssEnd - $rssStart );
343 if ( $status === false ) {
344 $msg = $job->toString() . " t={job_duration} error={job_error}";
345 $this->logger
->error( $msg, [
346 'job_type' => $job->getType(),
347 'job_duration' => $timeMs,
348 'job_error' => $error,
351 $msg = $job->toString() . " t=$timeMs error={$error}";
352 $this->debugCallback( $msg );
354 $msg = $job->toString() . " t={job_duration} good";
355 $this->logger
->info( $msg, [
356 'job_type' => $job->getType(),
357 'job_duration' => $timeMs,
360 $msg = $job->toString() . " t=$timeMs good";
361 $this->debugCallback( $msg );
364 return [ 'status' => $status, 'error' => $error, 'timeMs' => $timeMs ];
368 * @return int|null Max memory RSS in kilobytes
370 private function getMaxRssKb() {
371 $info = wfGetRusage() ?
: [];
372 // see https://linux.die.net/man/2/getrusage
373 return isset( $info['ru_maxrss'] ) ?
(int)$info['ru_maxrss'] : null;
378 * @return int Seconds for this runner to avoid doing more jobs of this type
379 * @see $wgJobBackoffThrottling
381 private function getBackoffTimeToWait( Job
$job ) {
382 $throttling = $this->config
->get( 'JobBackoffThrottling' );
384 if ( !isset( $throttling[$job->getType()] ) ||
$job instanceof DuplicateJob
) {
385 return 0; // not throttled
388 $itemsPerSecond = $throttling[$job->getType()];
389 if ( $itemsPerSecond <= 0 ) {
390 return 0; // not throttled
394 if ( $job->workItemCount() > 0 ) {
395 $exactSeconds = $job->workItemCount() / $itemsPerSecond;
396 // use randomized rounding
397 $seconds = floor( $exactSeconds );
398 $remainder = $exactSeconds - $seconds;
399 $seconds +
= ( mt_rand() / mt_getrandmax() < $remainder ) ?
1 : 0;
402 return (int)$seconds;
406 * Get the previous backoff expiries from persistent storage
407 * On I/O or lock acquisition failure this returns the original $backoffs.
409 * @param array $backoffs Map of (job type => UNIX timestamp)
410 * @param string $mode Lock wait mode - "wait" or "nowait"
411 * @return array Map of (job type => backoff expiry timestamp)
413 private function loadBackoffs( array $backoffs, $mode = 'wait' ) {
414 $file = wfTempDir() . '/mw-runJobs-backoffs.json';
415 if ( is_file( $file ) ) {
416 $noblock = ( $mode === 'nowait' ) ? LOCK_NB
: 0;
417 $handle = fopen( $file, 'rb' );
418 if ( !flock( $handle, LOCK_SH |
$noblock ) ) {
420 return $backoffs; // don't wait on lock
422 $content = stream_get_contents( $handle );
423 flock( $handle, LOCK_UN
);
425 $ctime = microtime( true );
426 $cBackoffs = json_decode( $content, true ) ?
: [];
427 foreach ( $cBackoffs as $type => $timestamp ) {
428 if ( $timestamp < $ctime ) {
429 unset( $cBackoffs[$type] );
440 * Merge the current backoff expiries from persistent storage
442 * The $deltas map is set to an empty array on success.
443 * On I/O or lock acquisition failure this returns the original $backoffs.
445 * @param array $backoffs Map of (job type => UNIX timestamp)
446 * @param array $deltas Map of (job type => seconds)
447 * @param string $mode Lock wait mode - "wait" or "nowait"
448 * @return array The new backoffs account for $backoffs and the latest file data
450 private function syncBackoffDeltas( array $backoffs, array &$deltas, $mode = 'wait' ) {
452 return $this->loadBackoffs( $backoffs, $mode );
455 $noblock = ( $mode === 'nowait' ) ? LOCK_NB
: 0;
456 $file = wfTempDir() . '/mw-runJobs-backoffs.json';
457 $handle = fopen( $file, 'wb+' );
458 if ( !flock( $handle, LOCK_EX |
$noblock ) ) {
460 return $backoffs; // don't wait on lock
462 $ctime = microtime( true );
463 $content = stream_get_contents( $handle );
464 $cBackoffs = json_decode( $content, true ) ?
: [];
465 foreach ( $deltas as $type => $seconds ) {
466 $cBackoffs[$type] = isset( $cBackoffs[$type] ) && $cBackoffs[$type] >= $ctime
467 ?
$cBackoffs[$type] +
$seconds
470 foreach ( $cBackoffs as $type => $timestamp ) {
471 if ( $timestamp < $ctime ) {
472 unset( $cBackoffs[$type] );
475 ftruncate( $handle, 0 );
476 fwrite( $handle, json_encode( $cBackoffs ) );
477 flock( $handle, LOCK_UN
);
486 * Make sure that this script is not too close to the memory usage limit.
487 * It is better to die in between jobs than OOM right in the middle of one.
490 private function checkMemoryOK() {
491 static $maxBytes = null;
492 if ( $maxBytes === null ) {
494 if ( preg_match( '!^(\d+)(k|m|g|)$!i', ini_get( 'memory_limit' ), $m ) ) {
495 list( , $num, $unit ) = $m;
496 $conv = [ 'g' => 1073741824, 'm' => 1048576, 'k' => 1024, '' => 1 ];
497 $maxBytes = $num * $conv[strtolower( $unit )];
502 $usedBytes = memory_get_usage();
503 if ( $maxBytes && $usedBytes >= 0.95 * $maxBytes ) {
504 $msg = "Detected excessive memory usage ({used_bytes}/{max_bytes}).";
505 $this->logger
->error( $msg, [
506 'used_bytes' => $usedBytes,
507 'max_bytes' => $maxBytes,
510 $msg = "Detected excessive memory usage ($usedBytes/$maxBytes).";
511 $this->debugCallback( $msg );
520 * Log the job message
521 * @param string $msg The message to log
523 private function debugCallback( $msg ) {
524 if ( $this->debug
) {
525 call_user_func_array( $this->debug
, [ wfTimestamp( TS_DB
) . " $msg\n" ] );
530 * Issue a commit on all masters who are currently in a transaction and have
531 * made changes to the database. It also supports sometimes waiting for the
532 * local wiki's replica DBs to catch up. See the documentation for
533 * $wgJobSerialCommitThreshold for more.
535 * @param LBFactory $lbFactory
537 * @param string $fnameTrxOwner
540 private function commitMasterChanges( LBFactory
$lbFactory, Job
$job, $fnameTrxOwner ) {
541 $syncThreshold = $this->config
->get( 'JobSerialCommitThreshold' );
544 $lb = $lbFactory->getMainLB( wfWikiID() );
545 if ( $syncThreshold !== false && $lb->getServerCount() > 1 ) {
546 // Generally, there is one master connection to the local DB
547 $dbwSerial = $lb->getAnyOpenConnection( $lb->getWriterIndex() );
548 // We need natively blocking fast locks
549 if ( $dbwSerial && $dbwSerial->namedLocksEnqueue() ) {
550 $time = $dbwSerial->pendingWriteQueryDuration( $dbwSerial::ESTIMATE_DB_APPLY
);
551 if ( $time < $syncThreshold ) {
558 // There are no replica DBs or writes are all to foreign DB (we don't handle that)
563 $lbFactory->commitMasterChanges(
565 // Abort if any transaction was too big
566 [ 'maxWriteDuration' => $this->config
->get( 'MaxJobDBWriteDuration' ) ]
572 $ms = intval( 1000 * $time );
574 $msg = $job->toString() . " COMMIT ENQUEUED [{job_commit_write_ms}ms of writes]";
575 $this->logger
->info( $msg, [
576 'job_type' => $job->getType(),
577 'job_commit_write_ms' => $ms,
580 $msg = $job->toString() . " COMMIT ENQUEUED [{$ms}ms of writes]";
581 $this->debugCallback( $msg );
583 // Wait for an exclusive lock to commit
584 if ( !$dbwSerial->lock( 'jobrunner-serial-commit', __METHOD__
, 30 ) ) {
585 // This will trigger a rollback in the main loop
586 throw new DBError( $dbwSerial, "Timed out waiting on commit queue." );
588 $unlocker = new ScopedCallback( function () use ( $dbwSerial ) {
589 $dbwSerial->unlock( 'jobrunner-serial-commit', __METHOD__
);
592 // Wait for the replica DBs to catch up
593 $pos = $lb->getMasterPos();
595 $lb->waitForAll( $pos );
598 // Actually commit the DB master changes
599 $lbFactory->commitMasterChanges(
601 // Abort if any transaction was too big
602 [ 'maxWriteDuration' => $this->config
->get( 'MaxJobDBWriteDuration' ) ]
604 ScopedCallback
::consume( $unlocker );