3 * Database-backed job queue code.
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
21 * @author Aaron Schulz
25 * Class to handle job queues stored in the DB
30 class JobQueueDB
extends JobQueue
{
31 const CACHE_TTL_SHORT
= 30; // integer; seconds to cache info without re-validating
32 const MAX_AGE_PRUNE
= 604800; // integer; seconds a job can live once claimed
33 const MAX_JOB_RANDOM
= 2147483647; // integer; 2^31 - 1, used for job_random
34 const MAX_OFFSET
= 255; // integer; maximum number of rows to skip
39 /** @var bool|string Name of an external DB cluster. False if not set */
40 protected $cluster = false;
43 * Additional parameters include:
44 * - cluster : The name of an external cluster registered via LBFactory.
45 * If not specified, the primary DB cluster for the wiki will be used.
46 * This can be overridden with a custom cluster so that DB handles will
47 * be retrieved via LBFactory::getExternalLB() and getConnection().
48 * @param array $params
50 protected function __construct( array $params ) {
53 parent
::__construct( $params );
55 $this->cluster
= isset( $params['cluster'] ) ?
$params['cluster'] : false;
56 // Make sure that we don't use the SQL cache, which would be harmful
57 $this->cache
= ( $wgMemc instanceof SqlBagOStuff
) ?
new EmptyBagOStuff() : $wgMemc;
60 protected function supportedOrders() {
61 return array( 'random', 'timestamp', 'fifo' );
64 protected function optimalOrder() {
69 * @see JobQueue::doIsEmpty()
72 protected function doIsEmpty() {
73 $dbr = $this->getSlaveDB();
75 $found = $dbr->selectField( // unclaimed job
76 'job', '1', array( 'job_cmd' => $this->type
, 'job_token' => '' ), __METHOD__
78 } catch ( DBError
$e ) {
79 $this->throwDBException( $e );
86 * @see JobQueue::doGetSize()
89 protected function doGetSize() {
90 $key = $this->getCacheKey( 'size' );
92 $size = $this->cache
->get( $key );
93 if ( is_int( $size ) ) {
98 $dbr = $this->getSlaveDB();
99 $size = (int)$dbr->selectField( 'job', 'COUNT(*)',
100 array( 'job_cmd' => $this->type
, 'job_token' => '' ),
103 } catch ( DBError
$e ) {
104 $this->throwDBException( $e );
106 $this->cache
->set( $key, $size, self
::CACHE_TTL_SHORT
);
112 * @see JobQueue::doGetAcquiredCount()
115 protected function doGetAcquiredCount() {
116 if ( $this->claimTTL
<= 0 ) {
117 return 0; // no acknowledgements
120 $key = $this->getCacheKey( 'acquiredcount' );
122 $count = $this->cache
->get( $key );
123 if ( is_int( $count ) ) {
127 $dbr = $this->getSlaveDB();
129 $count = (int)$dbr->selectField( 'job', 'COUNT(*)',
130 array( 'job_cmd' => $this->type
, "job_token != {$dbr->addQuotes( '' )}" ),
133 } catch ( DBError
$e ) {
134 $this->throwDBException( $e );
136 $this->cache
->set( $key, $count, self
::CACHE_TTL_SHORT
);
142 * @see JobQueue::doGetAbandonedCount()
144 * @throws MWException
146 protected function doGetAbandonedCount() {
149 if ( $this->claimTTL
<= 0 ) {
150 return 0; // no acknowledgements
153 $key = $this->getCacheKey( 'abandonedcount' );
155 $count = $wgMemc->get( $key );
156 if ( is_int( $count ) ) {
160 $dbr = $this->getSlaveDB();
162 $count = (int)$dbr->selectField( 'job', 'COUNT(*)',
164 'job_cmd' => $this->type
,
165 "job_token != {$dbr->addQuotes( '' )}",
166 "job_attempts >= " . $dbr->addQuotes( $this->maxTries
)
170 } catch ( DBError
$e ) {
171 $this->throwDBException( $e );
173 $wgMemc->set( $key, $count, self
::CACHE_TTL_SHORT
);
179 * @see JobQueue::doBatchPush()
182 * @throws DBError|Exception
185 protected function doBatchPush( array $jobs, $flags ) {
186 $dbw = $this->getMasterDB();
189 $method = __METHOD__
;
190 $dbw->onTransactionIdle(
191 function () use ( $dbw, $that, $jobs, $flags, $method ) {
192 $that->doBatchPushInternal( $dbw, $jobs, $flags, $method );
198 * This function should *not* be called outside of JobQueueDB
200 * @param IDatabase $dbw
203 * @param string $method
207 public function doBatchPushInternal( IDatabase
$dbw, array $jobs, $flags, $method ) {
208 if ( !count( $jobs ) ) {
212 $rowSet = array(); // (sha1 => job) map for jobs that are de-duplicated
213 $rowList = array(); // list of jobs for jobs that are not de-duplicated
214 foreach ( $jobs as $job ) {
215 $row = $this->insertFields( $job );
216 if ( $job->ignoreDuplicates() ) {
217 $rowSet[$row['job_sha1']] = $row;
223 if ( $flags & self
::QOS_ATOMIC
) {
224 $dbw->begin( $method ); // wrap all the job additions in one transaction
227 // Strip out any duplicate jobs that are already in the queue...
228 if ( count( $rowSet ) ) {
229 $res = $dbw->select( 'job', 'job_sha1',
231 // No job_type condition since it's part of the job_sha1 hash
232 'job_sha1' => array_keys( $rowSet ),
233 'job_token' => '' // unclaimed
237 foreach ( $res as $row ) {
238 wfDebug( "Job with hash '{$row->job_sha1}' is a duplicate.\n" );
239 unset( $rowSet[$row->job_sha1
] ); // already enqueued
242 // Build the full list of job rows to insert
243 $rows = array_merge( $rowList, array_values( $rowSet ) );
244 // Insert the job rows in chunks to avoid slave lag...
245 foreach ( array_chunk( $rows, 50 ) as $rowBatch ) {
246 $dbw->insert( 'job', $rowBatch, $method );
248 JobQueue
::incrStats( 'job-insert', $this->type
, count( $rows ) );
250 'job-insert-duplicate',
252 count( $rowSet ) +
count( $rowList ) - count( $rows )
254 } catch ( DBError
$e ) {
255 if ( $flags & self
::QOS_ATOMIC
) {
256 $dbw->rollback( $method );
260 if ( $flags & self
::QOS_ATOMIC
) {
261 $dbw->commit( $method );
268 * @see JobQueue::doPop()
271 protected function doPop() {
272 $dbw = $this->getMasterDB();
274 $dbw->commit( __METHOD__
, 'flush' ); // flush existing transaction
275 $autoTrx = $dbw->getFlag( DBO_TRX
); // get current setting
276 $dbw->clearFlag( DBO_TRX
); // make each query its own transaction
277 $scopedReset = new ScopedCallback( function () use ( $dbw, $autoTrx ) {
278 $dbw->setFlag( $autoTrx ? DBO_TRX
: 0 ); // restore old setting
281 $uuid = wfRandomString( 32 ); // pop attempt
282 $job = false; // job popped off
283 do { // retry when our row is invalid or deleted as a duplicate
284 // Try to reserve a row in the DB...
285 if ( in_array( $this->order
, array( 'fifo', 'timestamp' ) ) ) {
286 $row = $this->claimOldest( $uuid );
287 } else { // random first
288 $rand = mt_rand( 0, self
::MAX_JOB_RANDOM
); // encourage concurrent UPDATEs
289 $gte = (bool)mt_rand( 0, 1 ); // find rows with rand before/after $rand
290 $row = $this->claimRandom( $uuid, $rand, $gte );
292 // Check if we found a row to reserve...
294 break; // nothing to do
296 JobQueue
::incrStats( 'job-pop', $this->type
);
297 // Get the job object from the row...
298 $title = Title
::makeTitle( $row->job_namespace
, $row->job_title
);
299 $job = Job
::factory( $row->job_cmd
, $title,
300 self
::extractBlob( $row->job_params
), $row->job_id
);
301 $job->metadata
['id'] = $row->job_id
;
302 $job->metadata
['timestamp'] = $row->job_timestamp
;
306 if ( !$job ||
mt_rand( 0, 9 ) == 0 ) {
307 // Handled jobs that need to be recycled/deleted;
308 // any recycled jobs will be picked up next attempt
309 $this->recycleAndDeleteStaleJobs();
311 } catch ( DBError
$e ) {
312 $this->throwDBException( $e );
319 * Reserve a row with a single UPDATE without holding row locks over RTTs...
321 * @param string $uuid 32 char hex string
322 * @param int $rand Random unsigned integer (31 bits)
323 * @param bool $gte Search for job_random >= $random (otherwise job_random <= $random)
324 * @return stdClass|bool Row|false
326 protected function claimRandom( $uuid, $rand, $gte ) {
327 $dbw = $this->getMasterDB();
328 // Check cache to see if the queue has <= OFFSET items
329 $tinyQueue = $this->cache
->get( $this->getCacheKey( 'small' ) );
331 $row = false; // the row acquired
332 $invertedDirection = false; // whether one job_random direction was already scanned
333 // This uses a replication safe method for acquiring jobs. One could use UPDATE+LIMIT
334 // instead, but that either uses ORDER BY (in which case it deadlocks in MySQL) or is
335 // not replication safe. Due to http://bugs.mysql.com/bug.php?id=6980, subqueries cannot
336 // be used here with MySQL.
338 if ( $tinyQueue ) { // queue has <= MAX_OFFSET rows
339 // For small queues, using OFFSET will overshoot and return no rows more often.
340 // Instead, this uses job_random to pick a row (possibly checking both directions).
341 $ineq = $gte ?
'>=' : '<=';
342 $dir = $gte ?
'ASC' : 'DESC';
343 $row = $dbw->selectRow( 'job', self
::selectFields(), // find a random job
345 'job_cmd' => $this->type
,
346 'job_token' => '', // unclaimed
347 "job_random {$ineq} {$dbw->addQuotes( $rand )}" ),
349 array( 'ORDER BY' => "job_random {$dir}" )
351 if ( !$row && !$invertedDirection ) {
353 $invertedDirection = true;
354 continue; // try the other direction
356 } else { // table *may* have >= MAX_OFFSET rows
357 // Bug 42614: "ORDER BY job_random" with a job_random inequality causes high CPU
358 // in MySQL if there are many rows for some reason. This uses a small OFFSET
359 // instead of job_random for reducing excess claim retries.
360 $row = $dbw->selectRow( 'job', self
::selectFields(), // find a random job
362 'job_cmd' => $this->type
,
363 'job_token' => '', // unclaimed
366 array( 'OFFSET' => mt_rand( 0, self
::MAX_OFFSET
) )
369 $tinyQueue = true; // we know the queue must have <= MAX_OFFSET rows
370 $this->cache
->set( $this->getCacheKey( 'small' ), 1, 30 );
371 continue; // use job_random
375 if ( $row ) { // claim the job
376 $dbw->update( 'job', // update by PK
378 'job_token' => $uuid,
379 'job_token_timestamp' => $dbw->timestamp(),
380 'job_attempts = job_attempts+1' ),
381 array( 'job_cmd' => $this->type
, 'job_id' => $row->job_id
, 'job_token' => '' ),
384 // This might get raced out by another runner when claiming the previously
385 // selected row. The use of job_random should minimize this problem, however.
386 if ( !$dbw->affectedRows() ) {
387 $row = false; // raced out
390 break; // nothing to do
398 * Reserve a row with a single UPDATE without holding row locks over RTTs...
400 * @param string $uuid 32 char hex string
401 * @return stdClass|bool Row|false
403 protected function claimOldest( $uuid ) {
404 $dbw = $this->getMasterDB();
406 $row = false; // the row acquired
408 if ( $dbw->getType() === 'mysql' ) {
409 // Per http://bugs.mysql.com/bug.php?id=6980, we can't use subqueries on the
410 // same table being changed in an UPDATE query in MySQL (gives Error: 1093).
411 // Oracle and Postgre have no such limitation. However, MySQL offers an
412 // alternative here by supporting ORDER BY + LIMIT for UPDATE queries.
413 $dbw->query( "UPDATE {$dbw->tableName( 'job' )} " .
415 "job_token = {$dbw->addQuotes( $uuid ) }, " .
416 "job_token_timestamp = {$dbw->addQuotes( $dbw->timestamp() )}, " .
417 "job_attempts = job_attempts+1 " .
419 "job_cmd = {$dbw->addQuotes( $this->type )} " .
420 "AND job_token = {$dbw->addQuotes( '' )} " .
421 ") ORDER BY job_id ASC LIMIT 1",
425 // Use a subquery to find the job, within an UPDATE to claim it.
426 // This uses as much of the DB wrapper functions as possible.
429 'job_token' => $uuid,
430 'job_token_timestamp' => $dbw->timestamp(),
431 'job_attempts = job_attempts+1' ),
432 array( 'job_id = (' .
433 $dbw->selectSQLText( 'job', 'job_id',
434 array( 'job_cmd' => $this->type
, 'job_token' => '' ),
436 array( 'ORDER BY' => 'job_id ASC', 'LIMIT' => 1 ) ) .
442 // Fetch any row that we just reserved...
443 if ( $dbw->affectedRows() ) {
444 $row = $dbw->selectRow( 'job', self
::selectFields(),
445 array( 'job_cmd' => $this->type
, 'job_token' => $uuid ), __METHOD__
447 if ( !$row ) { // raced out by duplicate job removal
448 wfDebug( "Row deleted as duplicate by another process.\n" );
451 break; // nothing to do
459 * @see JobQueue::doAck()
461 * @throws MWException
464 protected function doAck( Job
$job ) {
465 if ( !isset( $job->metadata
['id'] ) ) {
466 throw new MWException( "Job of type '{$job->getType()}' has no ID." );
469 $dbw = $this->getMasterDB();
471 $dbw->commit( __METHOD__
, 'flush' ); // flush existing transaction
472 $autoTrx = $dbw->getFlag( DBO_TRX
); // get current setting
473 $dbw->clearFlag( DBO_TRX
); // make each query its own transaction
474 $scopedReset = new ScopedCallback( function () use ( $dbw, $autoTrx ) {
475 $dbw->setFlag( $autoTrx ? DBO_TRX
: 0 ); // restore old setting
478 // Delete a row with a single DELETE without holding row locks over RTTs...
480 array( 'job_cmd' => $this->type
, 'job_id' => $job->metadata
['id'] ), __METHOD__
);
482 JobQueue
::incrStats( 'job-ack', $this->type
);
483 } catch ( DBError
$e ) {
484 $this->throwDBException( $e );
491 * @see JobQueue::doDeduplicateRootJob()
492 * @param IJobSpecification $job
493 * @throws MWException
496 protected function doDeduplicateRootJob( IJobSpecification
$job ) {
497 $params = $job->getParams();
498 if ( !isset( $params['rootJobSignature'] ) ) {
499 throw new MWException( "Cannot register root job; missing 'rootJobSignature'." );
500 } elseif ( !isset( $params['rootJobTimestamp'] ) ) {
501 throw new MWException( "Cannot register root job; missing 'rootJobTimestamp'." );
503 $key = $this->getRootJobCacheKey( $params['rootJobSignature'] );
504 // Callers should call batchInsert() and then this function so that if the insert
505 // fails, the de-duplication registration will be aborted. Since the insert is
506 // deferred till "transaction idle", do the same here, so that the ordering is
507 // maintained. Having only the de-duplication registration succeed would cause
508 // jobs to become no-ops without any actual jobs that made them redundant.
509 $dbw = $this->getMasterDB();
510 $cache = $this->dupCache
;
511 $dbw->onTransactionIdle( function () use ( $cache, $params, $key, $dbw ) {
512 $timestamp = $cache->get( $key ); // current last timestamp of this job
513 if ( $timestamp && $timestamp >= $params['rootJobTimestamp'] ) {
514 return true; // a newer version of this root job was enqueued
517 // Update the timestamp of the last root job started at the location...
518 return $cache->set( $key, $params['rootJobTimestamp'], JobQueueDB
::ROOTJOB_TTL
);
525 * @see JobQueue::doDelete()
528 protected function doDelete() {
529 $dbw = $this->getMasterDB();
531 $dbw->delete( 'job', array( 'job_cmd' => $this->type
) );
532 } catch ( DBError
$e ) {
533 $this->throwDBException( $e );
540 * @see JobQueue::doWaitForBackups()
543 protected function doWaitForBackups() {
544 wfWaitForSlaves( false, $this->wiki
, $this->cluster ?
: false );
550 protected function doFlushCaches() {
551 foreach ( array( 'size', 'acquiredcount' ) as $type ) {
552 $this->cache
->delete( $this->getCacheKey( $type ) );
557 * @see JobQueue::getAllQueuedJobs()
560 public function getAllQueuedJobs() {
561 $dbr = $this->getSlaveDB();
563 return new MappedIterator(
564 $dbr->select( 'job', self
::selectFields(),
565 array( 'job_cmd' => $this->getType(), 'job_token' => '' ) ),
566 function ( $row ) use ( $dbr ) {
569 Title
::makeTitle( $row->job_namespace
, $row->job_title
),
570 strlen( $row->job_params
) ?
unserialize( $row->job_params
) : false
572 $job->metadata
['id'] = $row->job_id
;
573 $job->metadata
['timestamp'] = $row->job_timestamp
;
577 } catch ( DBError
$e ) {
578 $this->throwDBException( $e );
582 public function getCoalesceLocationInternal() {
583 return $this->cluster
584 ?
"DBCluster:{$this->cluster}:{$this->wiki}"
585 : "LBFactory:{$this->wiki}";
588 protected function doGetSiblingQueuesWithJobs( array $types ) {
589 $dbr = $this->getSlaveDB();
590 // @note: this does not check whether the jobs are claimed or not.
591 // This is useful so JobQueueGroup::pop() also sees queues that only
592 // have stale jobs. This lets recycleAndDeleteStaleJobs() re-enqueue
593 // failed jobs so that they can be popped again for that edge case.
594 $res = $dbr->select( 'job', 'DISTINCT job_cmd',
595 array( 'job_cmd' => $types ), __METHOD__
);
598 foreach ( $res as $row ) {
599 $types[] = $row->job_cmd
;
605 protected function doGetSiblingQueueSizes( array $types ) {
606 $dbr = $this->getSlaveDB();
607 $res = $dbr->select( 'job', array( 'job_cmd', 'COUNT(*) AS count' ),
608 array( 'job_cmd' => $types ), __METHOD__
, array( 'GROUP BY' => 'job_cmd' ) );
611 foreach ( $res as $row ) {
612 $sizes[$row->job_cmd
] = (int)$row->count
;
619 * Recycle or destroy any jobs that have been claimed for too long
621 * @return int Number of jobs recycled/deleted
623 public function recycleAndDeleteStaleJobs() {
625 $count = 0; // affected rows
626 $dbw = $this->getMasterDB();
629 if ( !$dbw->lock( "jobqueue-recycle-{$this->type}", __METHOD__
, 1 ) ) {
630 return $count; // already in progress
633 // Remove claims on jobs acquired for too long if enabled...
634 if ( $this->claimTTL
> 0 ) {
635 $claimCutoff = $dbw->timestamp( $now - $this->claimTTL
);
636 // Get the IDs of jobs that have be claimed but not finished after too long.
637 // These jobs can be recycled into the queue by expiring the claim. Selecting
638 // the IDs first means that the UPDATE can be done by primary key (less deadlocks).
639 $res = $dbw->select( 'job', 'job_id',
641 'job_cmd' => $this->type
,
642 "job_token != {$dbw->addQuotes( '' )}", // was acquired
643 "job_token_timestamp < {$dbw->addQuotes( $claimCutoff )}", // stale
644 "job_attempts < {$dbw->addQuotes( $this->maxTries )}" ), // retries left
650 }, iterator_to_array( $res )
652 if ( count( $ids ) ) {
653 // Reset job_token for these jobs so that other runners will pick them up.
654 // Set the timestamp to the current time, as it is useful to now that the job
655 // was already tried before (the timestamp becomes the "released" time).
659 'job_token_timestamp' => $dbw->timestamp( $now ) ), // time of release
664 $affected = $dbw->affectedRows();
666 JobQueue
::incrStats( 'job-recycle', $this->type
, $affected );
667 $this->aggr
->notifyQueueNonEmpty( $this->wiki
, $this->type
);
671 // Just destroy any stale jobs...
672 $pruneCutoff = $dbw->timestamp( $now - self
::MAX_AGE_PRUNE
);
674 'job_cmd' => $this->type
,
675 "job_token != {$dbw->addQuotes( '' )}", // was acquired
676 "job_token_timestamp < {$dbw->addQuotes( $pruneCutoff )}" // stale
678 if ( $this->claimTTL
> 0 ) { // only prune jobs attempted too many times...
679 $conds[] = "job_attempts >= {$dbw->addQuotes( $this->maxTries )}";
681 // Get the IDs of jobs that are considered stale and should be removed. Selecting
682 // the IDs first means that the UPDATE can be done by primary key (less deadlocks).
683 $res = $dbw->select( 'job', 'job_id', $conds, __METHOD__
);
687 }, iterator_to_array( $res )
689 if ( count( $ids ) ) {
690 $dbw->delete( 'job', array( 'job_id' => $ids ), __METHOD__
);
691 $affected = $dbw->affectedRows();
693 JobQueue
::incrStats( 'job-abandon', $this->type
, $affected );
696 $dbw->unlock( "jobqueue-recycle-{$this->type}", __METHOD__
);
697 } catch ( DBError
$e ) {
698 $this->throwDBException( $e );
705 * @param IJobSpecification $job
708 protected function insertFields( IJobSpecification
$job ) {
709 $dbw = $this->getMasterDB();
712 // Fields that describe the nature of the job
713 'job_cmd' => $job->getType(),
714 'job_namespace' => $job->getTitle()->getNamespace(),
715 'job_title' => $job->getTitle()->getDBkey(),
716 'job_params' => self
::makeBlob( $job->getParams() ),
717 // Additional job metadata
718 'job_id' => $dbw->nextSequenceValue( 'job_job_id_seq' ),
719 'job_timestamp' => $dbw->timestamp(),
720 'job_sha1' => wfBaseConvert(
721 sha1( serialize( $job->getDeduplicationInfo() ) ),
724 'job_random' => mt_rand( 0, self
::MAX_JOB_RANDOM
)
729 * @throws JobQueueConnectionError
732 protected function getSlaveDB() {
734 return $this->getDB( DB_SLAVE
);
735 } catch ( DBConnectionError
$e ) {
736 throw new JobQueueConnectionError( "DBConnectionError:" . $e->getMessage() );
741 * @throws JobQueueConnectionError
744 protected function getMasterDB() {
746 return $this->getDB( DB_MASTER
);
747 } catch ( DBConnectionError
$e ) {
748 throw new JobQueueConnectionError( "DBConnectionError:" . $e->getMessage() );
753 * @param int $index (DB_SLAVE/DB_MASTER)
756 protected function getDB( $index ) {
757 $lb = ( $this->cluster
!== false )
758 ?
wfGetLBFactory()->getExternalLB( $this->cluster
, $this->wiki
)
759 : wfGetLB( $this->wiki
);
761 return $lb->getConnectionRef( $index, array(), $this->wiki
);
765 * @param string $property
768 private function getCacheKey( $property ) {
769 list( $db, $prefix ) = wfSplitWikiID( $this->wiki
);
770 $cluster = is_string( $this->cluster
) ?
$this->cluster
: 'main';
772 return wfForeignMemcKey( $db, $prefix, 'jobqueue', $cluster, $this->type
, $property );
776 * @param array|bool $params
779 protected static function makeBlob( $params ) {
780 if ( $params !== false ) {
781 return serialize( $params );
788 * @param string $blob
791 protected static function extractBlob( $blob ) {
792 if ( (string)$blob !== '' ) {
793 return unserialize( $blob );
801 * @throws JobQueueError
803 protected function throwDBException( DBError
$e ) {
804 throw new JobQueueError( get_class( $e ) . ": " . $e->getMessage() );
808 * Return the list of job fields that should be selected.
812 public static function selectFields() {
823 'job_token_timestamp',