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\Logger\LoggerFactory
;
25 use Psr\Log\LoggerAwareInterface
;
26 use Psr\Log\LoggerInterface
;
29 * Job queue runner utility methods
34 class JobRunner
implements LoggerAwareInterface
{
35 /** @var callable|null Debug output handler */
39 * @var LoggerInterface $logger
43 const MAX_ALLOWED_LAG
= 3; // abort if more than this much DB lag is present
44 const LAG_CHECK_PERIOD
= 1.0; // check slave lag this many seconds
45 const ERROR_BACKOFF_TTL
= 1; // seconds to back off a queue due to errors
48 * @param callable $debug Optional debug output handler
50 public function setDebugHandler( $debug ) {
51 $this->debug
= $debug;
55 * @param LoggerInterface $logger
58 public function setLogger( LoggerInterface
$logger ) {
59 $this->logger
= $logger;
63 * @param LoggerInterface $logger
65 public function __construct( LoggerInterface
$logger = null ) {
66 if ( $logger === null ) {
67 $logger = LoggerFactory
::getInstance( 'runJobs' );
69 $this->setLogger( $logger );
73 * Run jobs of the specified number/type for the specified time
75 * The response map has a 'job' field that lists status of each job, including:
76 * - type : the job type
77 * - status : ok/failed
78 * - error : any error message string
79 * - time : the job run time in ms
80 * The response map also has:
81 * - backoffs : the (job type => seconds) map of backoff times
82 * - elapsed : the total time spent running tasks in ms
83 * - reached : the reason the script finished, one of (none-ready, job-limit, time-limit,
86 * This method outputs status information only if a debug handler was set.
87 * Any exceptions are caught and logged, but are not reported as output.
89 * @param array $options Map of parameters:
90 * - type : the job type (or false for the default types)
91 * - maxJobs : maximum number of jobs to run
92 * - maxTime : maximum time in seconds before stopping
93 * - throttle : whether to respect job backoff configuration
94 * @return array Summary response that can easily be JSON serialized
96 public function run( array $options ) {
97 global $wgJobClasses, $wgTrxProfilerLimits;
99 $response = [ 'jobs' => [], 'reached' => 'none-ready' ];
101 $type = isset( $options['type'] ) ?
$options['type'] : false;
102 $maxJobs = isset( $options['maxJobs'] ) ?
$options['maxJobs'] : false;
103 $maxTime = isset( $options['maxTime'] ) ?
$options['maxTime'] : false;
104 $noThrottle = isset( $options['throttle'] ) && !$options['throttle'];
106 // Bail if job type is invalid
107 if ( $type !== false && !isset( $wgJobClasses[$type] ) ) {
108 $response['reached'] = 'none-possible';
111 // Bail out if DB is in read-only mode
112 if ( wfReadOnly() ) {
113 $response['reached'] = 'read-only';
116 // Bail out if there is too much DB lag.
117 // This check should not block as we want to try other wiki queues.
118 list( , $maxLag ) = wfGetLB( wfWikiID() )->getMaxLag();
119 if ( $maxLag >= self
::MAX_ALLOWED_LAG
) {
120 $response['reached'] = 'slave-lag-limit';
124 // Flush any pending DB writes for sanity
125 wfGetLBFactory()->commitAll( __METHOD__
);
127 // Catch huge single updates that lead to slave lag
128 $trxProfiler = Profiler
::instance()->getTransactionProfiler();
129 $trxProfiler->setLogger( LoggerFactory
::getInstance( 'DBPerformance' ) );
130 $trxProfiler->setExpectations( $wgTrxProfilerLimits['JobRunner'], __METHOD__
);
132 // Some jobs types should not run until a certain timestamp
133 $backoffs = []; // map of (type => UNIX expiry)
134 $backoffDeltas = []; // map of (type => seconds)
135 $wait = 'wait'; // block to read backoffs the first time
137 $group = JobQueueGroup
::singleton();
138 $stats = RequestContext
::getMain()->getStats();
141 $startTime = microtime( true ); // time since jobs started running
142 $lastCheckTime = 1; // timestamp of last slave check
144 // Sync the persistent backoffs with concurrent runners
145 $backoffs = $this->syncBackoffDeltas( $backoffs, $backoffDeltas, $wait );
146 $blacklist = $noThrottle ?
[] : array_keys( $backoffs );
147 $wait = 'nowait'; // less important now
149 if ( $type === false ) {
151 JobQueueGroup
::TYPE_DEFAULT
,
152 JobQueueGroup
::USE_CACHE
,
155 } elseif ( in_array( $type, $blacklist ) ) {
156 $job = false; // requested queue in backoff state
158 $job = $group->pop( $type ); // job from a single queue
161 if ( $job ) { // found a job
164 $jType = $job->getType();
166 // Back off of certain jobs for a while (for throttling and for errors)
167 $ttw = $this->getBackoffTimeToWait( $job );
169 // Always add the delta for other runners in case the time running the
170 // job negated the backoff for each individually but not collectively.
171 $backoffDeltas[$jType] = isset( $backoffDeltas[$jType] )
172 ?
$backoffDeltas[$jType] +
$ttw
174 $backoffs = $this->syncBackoffDeltas( $backoffs, $backoffDeltas, $wait );
177 $info = $this->executeJob( $job, $stats, $popTime );
178 if ( $info['status'] !== false ||
!$job->allowRetries() ) {
179 $group->ack( $job ); // succeeded or job cannot be retried
182 // Back off of certain jobs for a while (for throttling and for errors)
183 if ( $info['status'] === false && mt_rand( 0, 49 ) == 0 ) {
184 $ttw = max( $ttw, self
::ERROR_BACKOFF_TTL
); // too many errors
185 $backoffDeltas[$jType] = isset( $backoffDeltas[$jType] )
186 ?
$backoffDeltas[$jType] +
$ttw
190 $response['jobs'][] = [
192 'status' => ( $info['status'] === false ) ?
'failed' : 'ok',
193 'error' => $info['error'],
194 'time' => $info['timeMs']
196 $timeMsTotal +
= $info['timeMs'];
198 // Break out if we hit the job count or wall time limits...
199 if ( $maxJobs && $jobsPopped >= $maxJobs ) {
200 $response['reached'] = 'job-limit';
202 } elseif ( $maxTime && ( microtime( true ) - $startTime ) > $maxTime ) {
203 $response['reached'] = 'time-limit';
207 // Don't let any of the main DB slaves get backed up.
208 // This only waits for so long before exiting and letting
209 // other wikis in the farm (on different masters) get a chance.
210 $timePassed = microtime( true ) - $lastCheckTime;
211 if ( $timePassed >= self
::LAG_CHECK_PERIOD ||
$timePassed < 0 ) {
213 wfGetLBFactory()->waitForReplication( [
214 'ifWritesSince' => $lastCheckTime,
215 'timeout' => self
::MAX_ALLOWED_LAG
217 } catch ( DBReplicationWaitError
$e ) {
218 $response['reached'] = 'slave-lag-limit';
221 $lastCheckTime = microtime( true );
223 // Don't let any queue slaves/backups fall behind
224 if ( $jobsPopped > 0 && ( $jobsPopped %
100 ) == 0 ) {
225 $group->waitForBackups();
228 // Bail if near-OOM instead of in a job
229 if ( !$this->checkMemoryOK() ) {
230 $response['reached'] = 'memory-limit';
234 } while ( $job ); // stop when there are no jobs
236 // Sync the persistent backoffs for the next runJobs.php pass
237 if ( $backoffDeltas ) {
238 $this->syncBackoffDeltas( $backoffs, $backoffDeltas, 'wait' );
241 $response['backoffs'] = $backoffs;
242 $response['elapsed'] = $timeMsTotal;
249 * @param BufferingStatsdDataFactory $stats
250 * @param float $popTime
251 * @return array Map of status/error/timeMs
253 private function executeJob( Job
$job, $stats, $popTime ) {
254 $jType = $job->getType();
255 $msg = $job->toString() . " STARTING";
256 $this->logger
->debug( $msg );
257 $this->debugCallback( $msg );
260 $rssStart = $this->getMaxRssKb();
261 $jobStartTime = microtime( true );
263 $status = $job->run();
264 $error = $job->getLastError();
265 $this->commitMasterChanges( $job );
267 DeferredUpdates
::doUpdates();
268 $this->commitMasterChanges( $job );
270 } catch ( Exception
$e ) {
271 MWExceptionHandler
::rollbackMasterChangesAndLog( $e );
273 $error = get_class( $e ) . ': ' . $e->getMessage();
274 MWExceptionHandler
::logException( $e );
276 // Commit all outstanding connections that are in a transaction
277 // to get a fresh repeatable read snapshot on every connection.
278 // Note that jobs are still responsible for handling slave lag.
279 wfGetLBFactory()->commitAll( __METHOD__
);
280 // Clear out title cache data from prior snapshots
281 LinkCache
::singleton()->clear();
282 $timeMs = intval( ( microtime( true ) - $jobStartTime ) * 1000 );
283 $rssEnd = $this->getMaxRssKb();
285 // Record how long jobs wait before getting popped
286 $readyTs = $job->getReadyTimestamp();
288 $pickupDelay = max( 0, $popTime - $readyTs );
289 $stats->timing( 'jobqueue.pickup_delay.all', 1000 * $pickupDelay );
290 $stats->timing( "jobqueue.pickup_delay.$jType", 1000 * $pickupDelay );
292 // Record root job age for jobs being run
293 $rootTimestamp = $job->getRootJobParams()['rootJobTimestamp'];
294 if ( $rootTimestamp ) {
295 $age = max( 0, $popTime - wfTimestamp( TS_UNIX
, $rootTimestamp ) );
296 $stats->timing( "jobqueue.pickup_root_age.$jType", 1000 * $age );
298 // Track the execution time for jobs
299 $stats->timing( "jobqueue.run.$jType", $timeMs );
300 // Track RSS increases for jobs (in case of memory leaks)
301 if ( $rssStart && $rssEnd ) {
302 $stats->increment( "jobqueue.rss_delta.$jType", $rssEnd - $rssStart );
305 if ( $status === false ) {
306 $msg = $job->toString() . " t=$timeMs error={$error}";
307 $this->logger
->error( $msg );
308 $this->debugCallback( $msg );
310 $msg = $job->toString() . " t=$timeMs good";
311 $this->logger
->info( $msg );
312 $this->debugCallback( $msg );
315 return [ 'status' => $status, 'error' => $error, 'timeMs' => $timeMs ];
319 * @return int|null Max memory RSS in kilobytes
321 private function getMaxRssKb() {
322 $info = wfGetRusage() ?
: [];
323 // see http://linux.die.net/man/2/getrusage
324 return isset( $info['ru_maxrss'] ) ?
(int)$info['ru_maxrss'] : null;
329 * @return int Seconds for this runner to avoid doing more jobs of this type
330 * @see $wgJobBackoffThrottling
332 private function getBackoffTimeToWait( Job
$job ) {
333 global $wgJobBackoffThrottling;
335 if ( !isset( $wgJobBackoffThrottling[$job->getType()] ) ||
336 $job instanceof DuplicateJob
// no work was done
338 return 0; // not throttled
341 $itemsPerSecond = $wgJobBackoffThrottling[$job->getType()];
342 if ( $itemsPerSecond <= 0 ) {
343 return 0; // not throttled
347 if ( $job->workItemCount() > 0 ) {
348 $exactSeconds = $job->workItemCount() / $itemsPerSecond;
349 // use randomized rounding
350 $seconds = floor( $exactSeconds );
351 $remainder = $exactSeconds - $seconds;
352 $seconds +
= ( mt_rand() / mt_getrandmax() < $remainder ) ?
1 : 0;
355 return (int)$seconds;
359 * Get the previous backoff expiries from persistent storage
360 * On I/O or lock acquisition failure this returns the original $backoffs.
362 * @param array $backoffs Map of (job type => UNIX timestamp)
363 * @param string $mode Lock wait mode - "wait" or "nowait"
364 * @return array Map of (job type => backoff expiry timestamp)
366 private function loadBackoffs( array $backoffs, $mode = 'wait' ) {
367 $file = wfTempDir() . '/mw-runJobs-backoffs.json';
368 if ( is_file( $file ) ) {
369 $noblock = ( $mode === 'nowait' ) ? LOCK_NB
: 0;
370 $handle = fopen( $file, 'rb' );
371 if ( !flock( $handle, LOCK_SH |
$noblock ) ) {
373 return $backoffs; // don't wait on lock
375 $content = stream_get_contents( $handle );
376 flock( $handle, LOCK_UN
);
378 $ctime = microtime( true );
379 $cBackoffs = json_decode( $content, true ) ?
: [];
380 foreach ( $cBackoffs as $type => $timestamp ) {
381 if ( $timestamp < $ctime ) {
382 unset( $cBackoffs[$type] );
393 * Merge the current backoff expiries from persistent storage
395 * The $deltas map is set to an empty array on success.
396 * On I/O or lock acquisition failure this returns the original $backoffs.
398 * @param array $backoffs Map of (job type => UNIX timestamp)
399 * @param array $deltas Map of (job type => seconds)
400 * @param string $mode Lock wait mode - "wait" or "nowait"
401 * @return array The new backoffs account for $backoffs and the latest file data
403 private function syncBackoffDeltas( array $backoffs, array &$deltas, $mode = 'wait' ) {
405 return $this->loadBackoffs( $backoffs, $mode );
408 $noblock = ( $mode === 'nowait' ) ? LOCK_NB
: 0;
409 $file = wfTempDir() . '/mw-runJobs-backoffs.json';
410 $handle = fopen( $file, 'wb+' );
411 if ( !flock( $handle, LOCK_EX |
$noblock ) ) {
413 return $backoffs; // don't wait on lock
415 $ctime = microtime( true );
416 $content = stream_get_contents( $handle );
417 $cBackoffs = json_decode( $content, true ) ?
: [];
418 foreach ( $deltas as $type => $seconds ) {
419 $cBackoffs[$type] = isset( $cBackoffs[$type] ) && $cBackoffs[$type] >= $ctime
420 ?
$cBackoffs[$type] +
$seconds
423 foreach ( $cBackoffs as $type => $timestamp ) {
424 if ( $timestamp < $ctime ) {
425 unset( $cBackoffs[$type] );
428 ftruncate( $handle, 0 );
429 fwrite( $handle, json_encode( $cBackoffs ) );
430 flock( $handle, LOCK_UN
);
439 * Make sure that this script is not too close to the memory usage limit.
440 * It is better to die in between jobs than OOM right in the middle of one.
443 private function checkMemoryOK() {
444 static $maxBytes = null;
445 if ( $maxBytes === null ) {
447 if ( preg_match( '!^(\d+)(k|m|g|)$!i', ini_get( 'memory_limit' ), $m ) ) {
448 list( , $num, $unit ) = $m;
449 $conv = [ 'g' => 1073741824, 'm' => 1048576, 'k' => 1024, '' => 1 ];
450 $maxBytes = $num * $conv[strtolower( $unit )];
455 $usedBytes = memory_get_usage();
456 if ( $maxBytes && $usedBytes >= 0.95 * $maxBytes ) {
457 $msg = "Detected excessive memory usage ($usedBytes/$maxBytes).";
458 $this->debugCallback( $msg );
459 $this->logger
->error( $msg );
468 * Log the job message
469 * @param string $msg The message to log
471 private function debugCallback( $msg ) {
472 if ( $this->debug
) {
473 call_user_func_array( $this->debug
, [ wfTimestamp( TS_DB
) . " $msg\n" ] );
478 * Issue a commit on all masters who are currently in a transaction and have
479 * made changes to the database. It also supports sometimes waiting for the
480 * local wiki's slaves to catch up. See the documentation for
481 * $wgJobSerialCommitThreshold for more.
486 private function commitMasterChanges( Job
$job ) {
487 global $wgJobSerialCommitThreshold;
489 $lb = wfGetLB( wfWikiID() );
490 if ( $wgJobSerialCommitThreshold !== false && $lb->getServerCount() > 1 ) {
491 // Generally, there is one master connection to the local DB
492 $dbwSerial = $lb->getAnyOpenConnection( $lb->getWriterIndex() );
498 ||
!$dbwSerial->namedLocksEnqueue()
499 ||
$dbwSerial->pendingWriteQueryDuration() < $wgJobSerialCommitThreshold
501 // Writes are all to foreign DBs, named locks don't form queues,
502 // or $wgJobSerialCommitThreshold is not reached; commit changes now
503 wfGetLBFactory()->commitMasterChanges( __METHOD__
);
507 $ms = intval( 1000 * $dbwSerial->pendingWriteQueryDuration() );
508 $msg = $job->toString() . " COMMIT ENQUEUED [{$ms}ms of writes]";
509 $this->logger
->warning( $msg );
510 $this->debugCallback( $msg );
512 // Wait for an exclusive lock to commit
513 if ( !$dbwSerial->lock( 'jobrunner-serial-commit', __METHOD__
, 30 ) ) {
514 // This will trigger a rollback in the main loop
515 throw new DBError( $dbwSerial, "Timed out waiting on commit queue." );
517 // Wait for the generic slave to catch up
518 $pos = $lb->getMasterPos();
520 $lb->waitForOne( $pos );
524 // Re-ping all masters with transactions. This throws DBError if some
525 // connection died while waiting on locks/slaves, triggering a rollback.
526 wfGetLBFactory()->forEachLB( function( LoadBalancer
$lb ) use ( $fname ) {
527 $lb->forEachOpenConnection( function( IDatabase
$conn ) use ( $fname ) {
528 if ( $conn->writesOrCallbacksPending() ) {
529 $conn->query( "SELECT 1", $fname );
534 // Actually commit the DB master changes
535 wfGetLBFactory()->commitMasterChanges( __METHOD__
);
538 $dbwSerial->unlock( 'jobrunner-serial-commit', __METHOD__
);