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
;
34 * Job queue runner utility methods
39 class JobRunner
implements LoggerAwareInterface
{
42 /** @var callable|null Debug output handler */
46 * @var LoggerInterface $logger
50 const MAX_ALLOWED_LAG
= 3; // abort if more than this much DB lag is present
51 const LAG_CHECK_PERIOD
= 1.0; // check replica DB lag this many seconds
52 const ERROR_BACKOFF_TTL
= 1; // seconds to back off a queue due to errors
53 const READONLY_BACKOFF_TTL
= 30; // seconds to back off a queue due to read-only errors
56 * @param callable $debug Optional debug output handler
58 public function setDebugHandler( $debug ) {
59 $this->debug
= $debug;
63 * @param LoggerInterface $logger
66 public function setLogger( LoggerInterface
$logger ) {
67 $this->logger
= $logger;
71 * @param LoggerInterface|null $logger
73 public function __construct( LoggerInterface
$logger = null ) {
74 if ( $logger === null ) {
75 $logger = LoggerFactory
::getInstance( 'runJobs' );
77 $this->setLogger( $logger );
78 $this->config
= MediaWikiServices
::getInstance()->getMainConfig();
82 * Run jobs of the specified number/type for the specified time
84 * The response map has a 'job' field that lists status of each job, including:
85 * - type : the job type
86 * - status : ok/failed
87 * - error : any error message string
88 * - time : the job run time in ms
89 * The response map also has:
90 * - backoffs : the (job type => seconds) map of backoff times
91 * - elapsed : the total time spent running tasks in ms
92 * - reached : the reason the script finished, one of (none-ready, job-limit, time-limit,
95 * This method outputs status information only if a debug handler was set.
96 * Any exceptions are caught and logged, but are not reported as output.
98 * @param array $options Map of parameters:
99 * - type : the job type (or false for the default types)
100 * - maxJobs : maximum number of jobs to run
101 * - maxTime : maximum time in seconds before stopping
102 * - throttle : whether to respect job backoff configuration
103 * @return array Summary response that can easily be JSON serialized
105 public function run( array $options ) {
106 $jobClasses = $this->config
->get( 'JobClasses' );
107 $profilerLimits = $this->config
->get( 'TrxProfilerLimits' );
109 $response = [ 'jobs' => [], 'reached' => 'none-ready' ];
111 $type = $options['type'] ??
false;
112 $maxJobs = $options['maxJobs'] ??
false;
113 $maxTime = $options['maxTime'] ??
false;
114 $noThrottle = isset( $options['throttle'] ) && !$options['throttle'];
116 // Bail if job type is invalid
117 if ( $type !== false && !isset( $jobClasses[$type] ) ) {
118 $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 if ( $lbFactory->hasTransactionRound() ) {
130 throw new LogicException( __METHOD__
. ' called with an active transaction round.' );
132 // Bail out if there is too much DB lag.
133 // This check should not block as we want to try other wiki queues.
134 list( , $maxLag ) = $lbFactory->getMainLB()->getMaxLag();
135 if ( $maxLag >= self
::MAX_ALLOWED_LAG
) {
136 $response['reached'] = 'replica-lag-limit';
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
174 if ( $job ) { // found a job
177 $jType = $job->getType();
179 WebRequest
::overrideRequestId( $job->getRequestId() );
181 // Back off of certain jobs for a while (for throttling and for errors)
182 $ttw = $this->getBackoffTimeToWait( $job );
184 // Always add the delta for other runners in case the time running the
185 // job negated the backoff for each individually but not collectively.
186 $backoffDeltas[$jType] = isset( $backoffDeltas[$jType] )
187 ?
$backoffDeltas[$jType] +
$ttw
189 $backoffs = $this->syncBackoffDeltas( $backoffs, $backoffDeltas, $wait );
192 $info = $this->executeJob( $job, $lbFactory, $stats, $popTime );
193 if ( $info['status'] !== false ||
!$job->allowRetries() ) {
194 $group->ack( $job ); // succeeded or job cannot be retried
197 // Back off of certain jobs for a while (for throttling and for errors)
198 if ( $info['status'] === false && mt_rand( 0, 49 ) == 0 ) {
199 $ttw = max( $ttw, $this->getErrorBackoffTTL( $info['error'] ) );
200 $backoffDeltas[$jType] = isset( $backoffDeltas[$jType] )
201 ?
$backoffDeltas[$jType] +
$ttw
205 $response['jobs'][] = [
207 'status' => ( $info['status'] === false ) ?
'failed' : 'ok',
208 'error' => $info['error'],
209 'time' => $info['timeMs']
211 $timeMsTotal +
= $info['timeMs'];
213 // Break out if we hit the job count or wall time limits...
214 if ( $maxJobs && $jobsPopped >= $maxJobs ) {
215 $response['reached'] = 'job-limit';
217 } elseif ( $maxTime && ( microtime( true ) - $startTime ) > $maxTime ) {
218 $response['reached'] = 'time-limit';
222 // Don't let any of the main DB replica DBs get backed up.
223 // This only waits for so long before exiting and letting
224 // other wikis in the farm (on different masters) get a chance.
225 $timePassed = microtime( true ) - $lastCheckTime;
226 if ( $timePassed >= self
::LAG_CHECK_PERIOD ||
$timePassed < 0 ) {
227 $success = $lbFactory->waitForReplication( [
228 'ifWritesSince' => $lastCheckTime,
229 'timeout' => self
::MAX_ALLOWED_LAG
,
232 $response['reached'] = 'replica-lag-limit';
235 $lastCheckTime = microtime( true );
238 // Bail if near-OOM instead of in a job
239 if ( !$this->checkMemoryOK() ) {
240 $response['reached'] = 'memory-limit';
244 } while ( $job ); // stop when there are no jobs
246 // Sync the persistent backoffs for the next runJobs.php pass
247 if ( $backoffDeltas ) {
248 $this->syncBackoffDeltas( $backoffs, $backoffDeltas, 'wait' );
251 $response['backoffs'] = $backoffs;
252 $response['elapsed'] = $timeMsTotal;
258 * @param string $error
259 * @return int TTL in seconds
261 private function getErrorBackoffTTL( $error ) {
262 return strpos( $error, 'DBReadOnlyError' ) !== false
263 ? self
::READONLY_BACKOFF_TTL
264 : self
::ERROR_BACKOFF_TTL
;
269 * @param LBFactory $lbFactory
270 * @param StatsdDataFactory $stats
271 * @param float $popTime
272 * @return array Map of status/error/timeMs
274 private function executeJob( Job
$job, LBFactory
$lbFactory, $stats, $popTime ) {
275 $jType = $job->getType();
276 $msg = $job->toString() . " STARTING";
277 $this->logger
->debug( $msg, [
278 'job_type' => $job->getType(),
280 $this->debugCallback( $msg );
283 $rssStart = $this->getMaxRssKb();
284 $jobStartTime = microtime( true );
286 $fnameTrxOwner = get_class( $job ) . '::run'; // give run() outer scope
287 if ( !$job->hasExecutionFlag( $job::JOB_NO_EXPLICIT_TRX_ROUND
) ) {
288 $lbFactory->beginMasterChanges( $fnameTrxOwner );
290 $status = $job->run();
291 $error = $job->getLastError();
292 $this->commitMasterChanges( $lbFactory, $job, $fnameTrxOwner );
293 // Run any deferred update tasks; doUpdates() manages transactions itself
294 DeferredUpdates
::doUpdates();
295 } catch ( Exception
$e ) {
296 MWExceptionHandler
::rollbackMasterChangesAndLog( $e );
298 $error = get_class( $e ) . ': ' . $e->getMessage();
300 // Always attempt to call teardown() even if Job throws exception.
302 $job->teardown( $status );
303 } catch ( Exception
$e ) {
304 MWExceptionHandler
::logException( $e );
307 // Commit all outstanding connections that are in a transaction
308 // to get a fresh repeatable read snapshot on every connection.
309 // Note that jobs are still responsible for handling replica DB lag.
310 $lbFactory->flushReplicaSnapshots( __METHOD__
);
311 // Clear out title cache data from prior snapshots
312 MediaWikiServices
::getInstance()->getLinkCache()->clear();
313 $timeMs = intval( ( microtime( true ) - $jobStartTime ) * 1000 );
314 $rssEnd = $this->getMaxRssKb();
316 // Record how long jobs wait before getting popped
317 $readyTs = $job->getReadyTimestamp();
319 $pickupDelay = max( 0, $popTime - $readyTs );
320 $stats->timing( 'jobqueue.pickup_delay.all', 1000 * $pickupDelay );
321 $stats->timing( "jobqueue.pickup_delay.$jType", 1000 * $pickupDelay );
323 // Record root job age for jobs being run
324 $rootTimestamp = $job->getRootJobParams()['rootJobTimestamp'];
325 if ( $rootTimestamp ) {
326 $age = max( 0, $popTime - wfTimestamp( TS_UNIX
, $rootTimestamp ) );
327 $stats->timing( "jobqueue.pickup_root_age.$jType", 1000 * $age );
329 // Track the execution time for jobs
330 $stats->timing( "jobqueue.run.$jType", $timeMs );
331 // Track RSS increases for jobs (in case of memory leaks)
332 if ( $rssStart && $rssEnd ) {
333 $stats->updateCount( "jobqueue.rss_delta.$jType", $rssEnd - $rssStart );
336 if ( $status === false ) {
337 $msg = $job->toString() . " t={job_duration} error={job_error}";
338 $this->logger
->error( $msg, [
339 'job_type' => $job->getType(),
340 'job_duration' => $timeMs,
341 'job_error' => $error,
344 $msg = $job->toString() . " t=$timeMs error={$error}";
345 $this->debugCallback( $msg );
347 $msg = $job->toString() . " t={job_duration} good";
348 $this->logger
->info( $msg, [
349 'job_type' => $job->getType(),
350 'job_duration' => $timeMs,
353 $msg = $job->toString() . " t=$timeMs good";
354 $this->debugCallback( $msg );
357 return [ 'status' => $status, 'error' => $error, 'timeMs' => $timeMs ];
361 * @return int|null Max memory RSS in kilobytes
363 private function getMaxRssKb() {
364 $info = wfGetRusage() ?
: [];
365 // see https://linux.die.net/man/2/getrusage
366 return isset( $info['ru_maxrss'] ) ?
(int)$info['ru_maxrss'] : null;
371 * @return int Seconds for this runner to avoid doing more jobs of this type
372 * @see $wgJobBackoffThrottling
374 private function getBackoffTimeToWait( Job
$job ) {
375 $throttling = $this->config
->get( 'JobBackoffThrottling' );
377 if ( !isset( $throttling[$job->getType()] ) ||
$job instanceof DuplicateJob
) {
378 return 0; // not throttled
381 $itemsPerSecond = $throttling[$job->getType()];
382 if ( $itemsPerSecond <= 0 ) {
383 return 0; // not throttled
387 if ( $job->workItemCount() > 0 ) {
388 $exactSeconds = $job->workItemCount() / $itemsPerSecond;
389 // use randomized rounding
390 $seconds = floor( $exactSeconds );
391 $remainder = $exactSeconds - $seconds;
392 $seconds +
= ( mt_rand() / mt_getrandmax() < $remainder ) ?
1 : 0;
395 return (int)$seconds;
399 * Get the previous backoff expiries from persistent storage
400 * On I/O or lock acquisition failure this returns the original $backoffs.
402 * @param array $backoffs Map of (job type => UNIX timestamp)
403 * @param string $mode Lock wait mode - "wait" or "nowait"
404 * @return array Map of (job type => backoff expiry timestamp)
406 private function loadBackoffs( array $backoffs, $mode = 'wait' ) {
407 $file = wfTempDir() . '/mw-runJobs-backoffs.json';
408 if ( is_file( $file ) ) {
409 $noblock = ( $mode === 'nowait' ) ? LOCK_NB
: 0;
410 $handle = fopen( $file, 'rb' );
411 if ( !flock( $handle, LOCK_SH |
$noblock ) ) {
413 return $backoffs; // don't wait on lock
415 $content = stream_get_contents( $handle );
416 flock( $handle, LOCK_UN
);
418 $ctime = microtime( true );
419 $cBackoffs = json_decode( $content, true ) ?
: [];
420 foreach ( $cBackoffs as $type => $timestamp ) {
421 if ( $timestamp < $ctime ) {
422 unset( $cBackoffs[$type] );
433 * Merge the current backoff expiries from persistent storage
435 * The $deltas map is set to an empty array on success.
436 * On I/O or lock acquisition failure this returns the original $backoffs.
438 * @param array $backoffs Map of (job type => UNIX timestamp)
439 * @param array $deltas Map of (job type => seconds)
440 * @param string $mode Lock wait mode - "wait" or "nowait"
441 * @return array The new backoffs account for $backoffs and the latest file data
443 private function syncBackoffDeltas( array $backoffs, array &$deltas, $mode = 'wait' ) {
445 return $this->loadBackoffs( $backoffs, $mode );
448 $noblock = ( $mode === 'nowait' ) ? LOCK_NB
: 0;
449 $file = wfTempDir() . '/mw-runJobs-backoffs.json';
450 $handle = fopen( $file, 'wb+' );
451 if ( !flock( $handle, LOCK_EX |
$noblock ) ) {
453 return $backoffs; // don't wait on lock
455 $ctime = microtime( true );
456 $content = stream_get_contents( $handle );
457 $cBackoffs = json_decode( $content, true ) ?
: [];
458 foreach ( $deltas as $type => $seconds ) {
459 $cBackoffs[$type] = isset( $cBackoffs[$type] ) && $cBackoffs[$type] >= $ctime
460 ?
$cBackoffs[$type] +
$seconds
463 foreach ( $cBackoffs as $type => $timestamp ) {
464 if ( $timestamp < $ctime ) {
465 unset( $cBackoffs[$type] );
468 ftruncate( $handle, 0 );
469 fwrite( $handle, json_encode( $cBackoffs ) );
470 flock( $handle, LOCK_UN
);
479 * Make sure that this script is not too close to the memory usage limit.
480 * It is better to die in between jobs than OOM right in the middle of one.
483 private function checkMemoryOK() {
484 static $maxBytes = null;
485 if ( $maxBytes === null ) {
487 if ( preg_match( '!^(\d+)(k|m|g|)$!i', ini_get( 'memory_limit' ), $m ) ) {
488 list( , $num, $unit ) = $m;
489 $conv = [ 'g' => 1073741824, 'm' => 1048576, 'k' => 1024, '' => 1 ];
490 $maxBytes = $num * $conv[strtolower( $unit )];
495 $usedBytes = memory_get_usage();
496 if ( $maxBytes && $usedBytes >= 0.95 * $maxBytes ) {
497 $msg = "Detected excessive memory usage ({used_bytes}/{max_bytes}).";
498 $this->logger
->error( $msg, [
499 'used_bytes' => $usedBytes,
500 'max_bytes' => $maxBytes,
503 $msg = "Detected excessive memory usage ($usedBytes/$maxBytes).";
504 $this->debugCallback( $msg );
513 * Log the job message
514 * @param string $msg The message to log
516 private function debugCallback( $msg ) {
517 if ( $this->debug
) {
518 call_user_func_array( $this->debug
, [ wfTimestamp( TS_DB
) . " $msg\n" ] );
523 * Issue a commit on all masters who are currently in a transaction and have
524 * made changes to the database. It also supports sometimes waiting for the
525 * local wiki's replica DBs to catch up. See the documentation for
526 * $wgJobSerialCommitThreshold for more.
528 * @param LBFactory $lbFactory
530 * @param string $fnameTrxOwner
533 private function commitMasterChanges( LBFactory
$lbFactory, Job
$job, $fnameTrxOwner ) {
534 $syncThreshold = $this->config
->get( 'JobSerialCommitThreshold' );
537 $lb = $lbFactory->getMainLB();
538 if ( $syncThreshold !== false && $lb->hasStreamingReplicaServers() ) {
539 // Generally, there is one master connection to the local DB
540 $dbwSerial = $lb->getAnyOpenConnection( $lb->getWriterIndex() );
541 // We need natively blocking fast locks
542 if ( $dbwSerial && $dbwSerial->namedLocksEnqueue() ) {
543 $time = $dbwSerial->pendingWriteQueryDuration( $dbwSerial::ESTIMATE_DB_APPLY
);
544 if ( $time < $syncThreshold ) {
551 // There are no replica DBs or writes are all to foreign DB (we don't handle that)
556 $lbFactory->commitMasterChanges(
558 // Abort if any transaction was too big
559 [ 'maxWriteDuration' => $this->config
->get( 'MaxJobDBWriteDuration' ) ]
565 $ms = intval( 1000 * $time );
567 $msg = $job->toString() . " COMMIT ENQUEUED [{job_commit_write_ms}ms of writes]";
568 $this->logger
->info( $msg, [
569 'job_type' => $job->getType(),
570 'job_commit_write_ms' => $ms,
573 $msg = $job->toString() . " COMMIT ENQUEUED [{$ms}ms of writes]";
574 $this->debugCallback( $msg );
576 // Wait for an exclusive lock to commit
577 if ( !$dbwSerial->lock( 'jobrunner-serial-commit', $fnameTrxOwner, 30 ) ) {
578 // This will trigger a rollback in the main loop
579 throw new DBError( $dbwSerial, "Timed out waiting on commit queue." );
581 $unlocker = new ScopedCallback( function () use ( $dbwSerial, $fnameTrxOwner ) {
582 $dbwSerial->unlock( 'jobrunner-serial-commit', $fnameTrxOwner );
585 // Wait for the replica DBs to catch up
586 $pos = $lb->getMasterPos();
588 $lb->waitForAll( $pos );
591 // Actually commit the DB master changes
592 $lbFactory->commitMasterChanges(
594 // Abort if any transaction was too big
595 [ 'maxWriteDuration' => $this->config
->get( 'MaxJobDBWriteDuration' ) ]
597 ScopedCallback
::consume( $unlocker );