cluster = isset( $params['cluster'] ) ? $params['cluster'] : false; // Make sure that we don't use the SQL cache, which would be harmful $this->cache = ( $wgMemc instanceof SqlBagOStuff ) ? new EmptyBagOStuff() : $wgMemc; } protected function supportedOrders() { return array( 'random', 'timestamp', 'fifo' ); } protected function optimalOrder() { return 'random'; } /** * @see JobQueue::doIsEmpty() * @return bool */ protected function doIsEmpty() { $key = $this->getCacheKey( 'empty' ); $isEmpty = $this->cache->get( $key ); if ( $isEmpty === 'true' ) { return true; } elseif ( $isEmpty === 'false' ) { return false; } $dbr = $this->getSlaveDB(); try { $found = $dbr->selectField( // unclaimed job 'job', '1', array( 'job_cmd' => $this->type, 'job_token' => '' ), __METHOD__ ); } catch ( DBError $e ) { $this->throwDBException( $e ); } $this->cache->add( $key, $found ? 'false' : 'true', self::CACHE_TTL_LONG ); return !$found; } /** * @see JobQueue::doGetSize() * @return int */ protected function doGetSize() { $key = $this->getCacheKey( 'size' ); $size = $this->cache->get( $key ); if ( is_int( $size ) ) { return $size; } try { $dbr = $this->getSlaveDB(); $size = (int)$dbr->selectField( 'job', 'COUNT(*)', array( 'job_cmd' => $this->type, 'job_token' => '' ), __METHOD__ ); } catch ( DBError $e ) { $this->throwDBException( $e ); } $this->cache->set( $key, $size, self::CACHE_TTL_SHORT ); return $size; } /** * @see JobQueue::doGetAcquiredCount() * @return int */ protected function doGetAcquiredCount() { if ( $this->claimTTL <= 0 ) { return 0; // no acknowledgements } $key = $this->getCacheKey( 'acquiredcount' ); $count = $this->cache->get( $key ); if ( is_int( $count ) ) { return $count; } $dbr = $this->getSlaveDB(); try { $count = (int)$dbr->selectField( 'job', 'COUNT(*)', array( 'job_cmd' => $this->type, "job_token != {$dbr->addQuotes( '' )}" ), __METHOD__ ); } catch ( DBError $e ) { $this->throwDBException( $e ); } $this->cache->set( $key, $count, self::CACHE_TTL_SHORT ); return $count; } /** * @see JobQueue::doGetAbandonedCount() * @return int * @throws MWException */ protected function doGetAbandonedCount() { global $wgMemc; if ( $this->claimTTL <= 0 ) { return 0; // no acknowledgements } $key = $this->getCacheKey( 'abandonedcount' ); $count = $wgMemc->get( $key ); if ( is_int( $count ) ) { return $count; } $dbr = $this->getSlaveDB(); try { $count = (int)$dbr->selectField( 'job', 'COUNT(*)', array( 'job_cmd' => $this->type, "job_token != {$dbr->addQuotes( '' )}", "job_attempts >= " . $dbr->addQuotes( $this->maxTries ) ), __METHOD__ ); } catch ( DBError $e ) { $this->throwDBException( $e ); } $wgMemc->set( $key, $count, self::CACHE_TTL_SHORT ); return $count; } /** * @see JobQueue::doBatchPush() * @param array $jobs * @param int $flags * @throws DBError|Exception * @return void */ protected function doBatchPush( array $jobs, $flags ) { $dbw = $this->getMasterDB(); $that = $this; $method = __METHOD__; $dbw->onTransactionIdle( function () use ( $dbw, $that, $jobs, $flags, $method ) { $that->doBatchPushInternal( $dbw, $jobs, $flags, $method ); } ); } /** * This function should *not* be called outside of JobQueueDB * * @param IDatabase $dbw * @param array $jobs * @param int $flags * @param string $method * @throws DBError * @return void */ public function doBatchPushInternal( IDatabase $dbw, array $jobs, $flags, $method ) { if ( !count( $jobs ) ) { return; } $rowSet = array(); // (sha1 => job) map for jobs that are de-duplicated $rowList = array(); // list of jobs for jobs that are not de-duplicated foreach ( $jobs as $job ) { $row = $this->insertFields( $job ); if ( $job->ignoreDuplicates() ) { $rowSet[$row['job_sha1']] = $row; } else { $rowList[] = $row; } } if ( $flags & self::QOS_ATOMIC ) { $dbw->begin( $method ); // wrap all the job additions in one transaction } try { // Strip out any duplicate jobs that are already in the queue... if ( count( $rowSet ) ) { $res = $dbw->select( 'job', 'job_sha1', array( // No job_type condition since it's part of the job_sha1 hash 'job_sha1' => array_keys( $rowSet ), 'job_token' => '' // unclaimed ), $method ); foreach ( $res as $row ) { wfDebug( "Job with hash '{$row->job_sha1}' is a duplicate.\n" ); unset( $rowSet[$row->job_sha1] ); // already enqueued } } // Build the full list of job rows to insert $rows = array_merge( $rowList, array_values( $rowSet ) ); // Insert the job rows in chunks to avoid slave lag... foreach ( array_chunk( $rows, 50 ) as $rowBatch ) { $dbw->insert( 'job', $rowBatch, $method ); } JobQueue::incrStats( 'job-insert', $this->type, count( $rows ), $this->wiki ); JobQueue::incrStats( 'job-insert-duplicate', $this->type, count( $rowSet ) + count( $rowList ) - count( $rows ), $this->wiki ); } catch ( DBError $e ) { if ( $flags & self::QOS_ATOMIC ) { $dbw->rollback( $method ); } throw $e; } if ( $flags & self::QOS_ATOMIC ) { $dbw->commit( $method ); } $this->cache->set( $this->getCacheKey( 'empty' ), 'false', JobQueueDB::CACHE_TTL_LONG ); return; } /** * @see JobQueue::doPop() * @return Job|bool */ protected function doPop() { if ( $this->cache->get( $this->getCacheKey( 'empty' ) ) === 'true' ) { return false; // queue is empty } $dbw = $this->getMasterDB(); try { $dbw->commit( __METHOD__, 'flush' ); // flush existing transaction $autoTrx = $dbw->getFlag( DBO_TRX ); // get current setting $dbw->clearFlag( DBO_TRX ); // make each query its own transaction $scopedReset = new ScopedCallback( function () use ( $dbw, $autoTrx ) { $dbw->setFlag( $autoTrx ? DBO_TRX : 0 ); // restore old setting } ); $uuid = wfRandomString( 32 ); // pop attempt $job = false; // job popped off do { // retry when our row is invalid or deleted as a duplicate // Try to reserve a row in the DB... if ( in_array( $this->order, array( 'fifo', 'timestamp' ) ) ) { $row = $this->claimOldest( $uuid ); } else { // random first $rand = mt_rand( 0, self::MAX_JOB_RANDOM ); // encourage concurrent UPDATEs $gte = (bool)mt_rand( 0, 1 ); // find rows with rand before/after $rand $row = $this->claimRandom( $uuid, $rand, $gte ); } // Check if we found a row to reserve... if ( !$row ) { $this->cache->set( $this->getCacheKey( 'empty' ), 'true', self::CACHE_TTL_LONG ); break; // nothing to do } JobQueue::incrStats( 'job-pop', $this->type, 1, $this->wiki ); // Get the job object from the row... $title = Title::makeTitleSafe( $row->job_namespace, $row->job_title ); if ( !$title ) { $dbw->delete( 'job', array( 'job_id' => $row->job_id ), __METHOD__ ); wfDebug( "Row has invalid title '{$row->job_title}'.\n" ); continue; // try again } $job = Job::factory( $row->job_cmd, $title, self::extractBlob( $row->job_params ), $row->job_id ); $job->metadata['id'] = $row->job_id; break; // done } while ( true ); } catch ( DBError $e ) { $this->throwDBException( $e ); } return $job; } /** * Reserve a row with a single UPDATE without holding row locks over RTTs... * * @param string $uuid 32 char hex string * @param int $rand Random unsigned integer (31 bits) * @param bool $gte Search for job_random >= $random (otherwise job_random <= $random) * @return stdClass|bool Row|false */ protected function claimRandom( $uuid, $rand, $gte ) { $dbw = $this->getMasterDB(); // Check cache to see if the queue has <= OFFSET items $tinyQueue = $this->cache->get( $this->getCacheKey( 'small' ) ); $row = false; // the row acquired $invertedDirection = false; // whether one job_random direction was already scanned // This uses a replication safe method for acquiring jobs. One could use UPDATE+LIMIT // instead, but that either uses ORDER BY (in which case it deadlocks in MySQL) or is // not replication safe. Due to http://bugs.mysql.com/bug.php?id=6980, subqueries cannot // be used here with MySQL. do { if ( $tinyQueue ) { // queue has <= MAX_OFFSET rows // For small queues, using OFFSET will overshoot and return no rows more often. // Instead, this uses job_random to pick a row (possibly checking both directions). $ineq = $gte ? '>=' : '<='; $dir = $gte ? 'ASC' : 'DESC'; $row = $dbw->selectRow( 'job', self::selectFields(), // find a random job array( 'job_cmd' => $this->type, 'job_token' => '', // unclaimed "job_random {$ineq} {$dbw->addQuotes( $rand )}" ), __METHOD__, array( 'ORDER BY' => "job_random {$dir}" ) ); if ( !$row && !$invertedDirection ) { $gte = !$gte; $invertedDirection = true; continue; // try the other direction } } else { // table *may* have >= MAX_OFFSET rows // Bug 42614: "ORDER BY job_random" with a job_random inequality causes high CPU // in MySQL if there are many rows for some reason. This uses a small OFFSET // instead of job_random for reducing excess claim retries. $row = $dbw->selectRow( 'job', self::selectFields(), // find a random job array( 'job_cmd' => $this->type, 'job_token' => '', // unclaimed ), __METHOD__, array( 'OFFSET' => mt_rand( 0, self::MAX_OFFSET ) ) ); if ( !$row ) { $tinyQueue = true; // we know the queue must have <= MAX_OFFSET rows $this->cache->set( $this->getCacheKey( 'small' ), 1, 30 ); continue; // use job_random } } if ( $row ) { // claim the job $dbw->update( 'job', // update by PK array( 'job_token' => $uuid, 'job_token_timestamp' => $dbw->timestamp(), 'job_attempts = job_attempts+1' ), array( 'job_cmd' => $this->type, 'job_id' => $row->job_id, 'job_token' => '' ), __METHOD__ ); // This might get raced out by another runner when claiming the previously // selected row. The use of job_random should minimize this problem, however. if ( !$dbw->affectedRows() ) { $row = false; // raced out } } else { break; // nothing to do } } while ( !$row ); return $row; } /** * Reserve a row with a single UPDATE without holding row locks over RTTs... * * @param string $uuid 32 char hex string * @return stdClass|bool Row|false */ protected function claimOldest( $uuid ) { $dbw = $this->getMasterDB(); $row = false; // the row acquired do { if ( $dbw->getType() === 'mysql' ) { // Per http://bugs.mysql.com/bug.php?id=6980, we can't use subqueries on the // same table being changed in an UPDATE query in MySQL (gives Error: 1093). // Oracle and Postgre have no such limitation. However, MySQL offers an // alternative here by supporting ORDER BY + LIMIT for UPDATE queries. $dbw->query( "UPDATE {$dbw->tableName( 'job' )} " . "SET " . "job_token = {$dbw->addQuotes( $uuid ) }, " . "job_token_timestamp = {$dbw->addQuotes( $dbw->timestamp() )}, " . "job_attempts = job_attempts+1 " . "WHERE ( " . "job_cmd = {$dbw->addQuotes( $this->type )} " . "AND job_token = {$dbw->addQuotes( '' )} " . ") ORDER BY job_id ASC LIMIT 1", __METHOD__ ); } else { // Use a subquery to find the job, within an UPDATE to claim it. // This uses as much of the DB wrapper functions as possible. $dbw->update( 'job', array( 'job_token' => $uuid, 'job_token_timestamp' => $dbw->timestamp(), 'job_attempts = job_attempts+1' ), array( 'job_id = (' . $dbw->selectSQLText( 'job', 'job_id', array( 'job_cmd' => $this->type, 'job_token' => '' ), __METHOD__, array( 'ORDER BY' => 'job_id ASC', 'LIMIT' => 1 ) ) . ')' ), __METHOD__ ); } // Fetch any row that we just reserved... if ( $dbw->affectedRows() ) { $row = $dbw->selectRow( 'job', self::selectFields(), array( 'job_cmd' => $this->type, 'job_token' => $uuid ), __METHOD__ ); if ( !$row ) { // raced out by duplicate job removal wfDebug( "Row deleted as duplicate by another process.\n" ); } } else { break; // nothing to do } } while ( !$row ); return $row; } /** * @see JobQueue::doAck() * @param Job $job * @throws MWException * @return Job|bool */ protected function doAck( Job $job ) { if ( !isset( $job->metadata['id'] ) ) { throw new MWException( "Job of type '{$job->getType()}' has no ID." ); } $dbw = $this->getMasterDB(); try { $dbw->commit( __METHOD__, 'flush' ); // flush existing transaction $autoTrx = $dbw->getFlag( DBO_TRX ); // get current setting $dbw->clearFlag( DBO_TRX ); // make each query its own transaction $scopedReset = new ScopedCallback( function () use ( $dbw, $autoTrx ) { $dbw->setFlag( $autoTrx ? DBO_TRX : 0 ); // restore old setting } ); // Delete a row with a single DELETE without holding row locks over RTTs... $dbw->delete( 'job', array( 'job_cmd' => $this->type, 'job_id' => $job->metadata['id'] ), __METHOD__ ); } catch ( DBError $e ) { $this->throwDBException( $e ); } return true; } /** * @see JobQueue::doDeduplicateRootJob() * @param Job $job * @throws MWException * @return bool */ protected function doDeduplicateRootJob( Job $job ) { $params = $job->getParams(); if ( !isset( $params['rootJobSignature'] ) ) { throw new MWException( "Cannot register root job; missing 'rootJobSignature'." ); } elseif ( !isset( $params['rootJobTimestamp'] ) ) { throw new MWException( "Cannot register root job; missing 'rootJobTimestamp'." ); } $key = $this->getRootJobCacheKey( $params['rootJobSignature'] ); // Callers should call batchInsert() and then this function so that if the insert // fails, the de-duplication registration will be aborted. Since the insert is // deferred till "transaction idle", do the same here, so that the ordering is // maintained. Having only the de-duplication registration succeed would cause // jobs to become no-ops without any actual jobs that made them redundant. $dbw = $this->getMasterDB(); $cache = $this->dupCache; $dbw->onTransactionIdle( function () use ( $cache, $params, $key, $dbw ) { $timestamp = $cache->get( $key ); // current last timestamp of this job if ( $timestamp && $timestamp >= $params['rootJobTimestamp'] ) { return true; // a newer version of this root job was enqueued } // Update the timestamp of the last root job started at the location... return $cache->set( $key, $params['rootJobTimestamp'], JobQueueDB::ROOTJOB_TTL ); } ); return true; } /** * @see JobQueue::doDelete() * @return bool */ protected function doDelete() { $dbw = $this->getMasterDB(); try { $dbw->delete( 'job', array( 'job_cmd' => $this->type ) ); } catch ( DBError $e ) { $this->throwDBException( $e ); } return true; } /** * @see JobQueue::doWaitForBackups() * @return void */ protected function doWaitForBackups() { wfWaitForSlaves( false, $this->wiki, $this->cluster ?: false ); } /** * @return array */ protected function doGetPeriodicTasks() { return array( 'recycleAndDeleteStaleJobs' => array( 'callback' => array( $this, 'recycleAndDeleteStaleJobs' ), 'period' => ceil( $this->claimTTL / 2 ) ) ); } /** * @return void */ protected function doFlushCaches() { foreach ( array( 'empty', 'size', 'acquiredcount' ) as $type ) { $this->cache->delete( $this->getCacheKey( $type ) ); } } /** * @see JobQueue::getAllQueuedJobs() * @return Iterator */ public function getAllQueuedJobs() { $dbr = $this->getSlaveDB(); try { return new MappedIterator( $dbr->select( 'job', self::selectFields(), array( 'job_cmd' => $this->getType(), 'job_token' => '' ) ), function ( $row ) use ( $dbr ) { $job = Job::factory( $row->job_cmd, Title::makeTitle( $row->job_namespace, $row->job_title ), strlen( $row->job_params ) ? unserialize( $row->job_params ) : false ); $job->metadata['id'] = $row->job_id; return $job; } ); } catch ( DBError $e ) { $this->throwDBException( $e ); } } public function getCoalesceLocationInternal() { return $this->cluster ? "DBCluster:{$this->cluster}:{$this->wiki}" : "LBFactory:{$this->wiki}"; } protected function doGetSiblingQueuesWithJobs( array $types ) { $dbr = $this->getSlaveDB(); $res = $dbr->select( 'job', 'DISTINCT job_cmd', array( 'job_cmd' => $types ), __METHOD__ ); $types = array(); foreach ( $res as $row ) { $types[] = $row->job_cmd; } return $types; } protected function doGetSiblingQueueSizes( array $types ) { $dbr = $this->getSlaveDB(); $res = $dbr->select( 'job', array( 'job_cmd', 'COUNT(*) AS count' ), array( 'job_cmd' => $types ), __METHOD__, array( 'GROUP BY' => 'job_cmd' ) ); $sizes = array(); foreach ( $res as $row ) { $sizes[$row->job_cmd] = (int)$row->count; } return $sizes; } /** * Recycle or destroy any jobs that have been claimed for too long * * @return int Number of jobs recycled/deleted */ public function recycleAndDeleteStaleJobs() { $now = time(); $count = 0; // affected rows $dbw = $this->getMasterDB(); try { if ( !$dbw->lock( "jobqueue-recycle-{$this->type}", __METHOD__, 1 ) ) { return $count; // already in progress } // Remove claims on jobs acquired for too long if enabled... if ( $this->claimTTL > 0 ) { $claimCutoff = $dbw->timestamp( $now - $this->claimTTL ); // Get the IDs of jobs that have be claimed but not finished after too long. // These jobs can be recycled into the queue by expiring the claim. Selecting // the IDs first means that the UPDATE can be done by primary key (less deadlocks). $res = $dbw->select( 'job', 'job_id', array( 'job_cmd' => $this->type, "job_token != {$dbw->addQuotes( '' )}", // was acquired "job_token_timestamp < {$dbw->addQuotes( $claimCutoff )}", // stale "job_attempts < {$dbw->addQuotes( $this->maxTries )}" ), // retries left __METHOD__ ); $ids = array_map( function ( $o ) { return $o->job_id; }, iterator_to_array( $res ) ); if ( count( $ids ) ) { // Reset job_token for these jobs so that other runners will pick them up. // Set the timestamp to the current time, as it is useful to now that the job // was already tried before (the timestamp becomes the "released" time). $dbw->update( 'job', array( 'job_token' => '', 'job_token_timestamp' => $dbw->timestamp( $now ) ), // time of release array( 'job_id' => $ids ), __METHOD__ ); $affected = $dbw->affectedRows(); $count += $affected; JobQueue::incrStats( 'job-recycle', $this->type, $affected, $this->wiki ); // The tasks recycled jobs or release delayed jobs into the queue $this->cache->set( $this->getCacheKey( 'empty' ), 'false', self::CACHE_TTL_LONG ); $this->aggr->notifyQueueNonEmpty( $this->wiki, $this->type ); } } // Just destroy any stale jobs... $pruneCutoff = $dbw->timestamp( $now - self::MAX_AGE_PRUNE ); $conds = array( 'job_cmd' => $this->type, "job_token != {$dbw->addQuotes( '' )}", // was acquired "job_token_timestamp < {$dbw->addQuotes( $pruneCutoff )}" // stale ); if ( $this->claimTTL > 0 ) { // only prune jobs attempted too many times... $conds[] = "job_attempts >= {$dbw->addQuotes( $this->maxTries )}"; } // Get the IDs of jobs that are considered stale and should be removed. Selecting // the IDs first means that the UPDATE can be done by primary key (less deadlocks). $res = $dbw->select( 'job', 'job_id', $conds, __METHOD__ ); $ids = array_map( function ( $o ) { return $o->job_id; }, iterator_to_array( $res ) ); if ( count( $ids ) ) { $dbw->delete( 'job', array( 'job_id' => $ids ), __METHOD__ ); $affected = $dbw->affectedRows(); $count += $affected; JobQueue::incrStats( 'job-abandon', $this->type, $affected, $this->wiki ); } $dbw->unlock( "jobqueue-recycle-{$this->type}", __METHOD__ ); } catch ( DBError $e ) { $this->throwDBException( $e ); } return $count; } /** * @param IJobSpecification $job * @return array */ protected function insertFields( IJobSpecification $job ) { $dbw = $this->getMasterDB(); return array( // Fields that describe the nature of the job 'job_cmd' => $job->getType(), 'job_namespace' => $job->getTitle()->getNamespace(), 'job_title' => $job->getTitle()->getDBkey(), 'job_params' => self::makeBlob( $job->getParams() ), // Additional job metadata 'job_id' => $dbw->nextSequenceValue( 'job_job_id_seq' ), 'job_timestamp' => $dbw->timestamp(), 'job_sha1' => wfBaseConvert( sha1( serialize( $job->getDeduplicationInfo() ) ), 16, 36, 31 ), 'job_random' => mt_rand( 0, self::MAX_JOB_RANDOM ) ); } /** * @throws JobQueueConnectionError * @return DBConnRef */ protected function getSlaveDB() { try { return $this->getDB( DB_SLAVE ); } catch ( DBConnectionError $e ) { throw new JobQueueConnectionError( "DBConnectionError:" . $e->getMessage() ); } } /** * @throws JobQueueConnectionError * @return DBConnRef */ protected function getMasterDB() { try { return $this->getDB( DB_MASTER ); } catch ( DBConnectionError $e ) { throw new JobQueueConnectionError( "DBConnectionError:" . $e->getMessage() ); } } /** * @param int $index (DB_SLAVE/DB_MASTER) * @return DBConnRef */ protected function getDB( $index ) { $lb = ( $this->cluster !== false ) ? wfGetLBFactory()->getExternalLB( $this->cluster, $this->wiki ) : wfGetLB( $this->wiki ); return $lb->getConnectionRef( $index, array(), $this->wiki ); } /** * @param string $property * @return string */ private function getCacheKey( $property ) { list( $db, $prefix ) = wfSplitWikiID( $this->wiki ); $cluster = is_string( $this->cluster ) ? $this->cluster : 'main'; return wfForeignMemcKey( $db, $prefix, 'jobqueue', $cluster, $this->type, $property ); } /** * @param array|bool $params * @return string */ protected static function makeBlob( $params ) { if ( $params !== false ) { return serialize( $params ); } else { return ''; } } /** * @param string $blob * @return bool|mixed */ protected static function extractBlob( $blob ) { if ( (string)$blob !== '' ) { return unserialize( $blob ); } else { return false; } } /** * @param DBError $e * @throws JobQueueError */ protected function throwDBException( DBError $e ) { throw new JobQueueError( get_class( $e ) . ": " . $e->getMessage() ); } /** * Return the list of job fields that should be selected. * @since 1.23 * @return array */ public static function selectFields() { return array( 'job_id', 'job_cmd', 'job_namespace', 'job_title', 'job_timestamp', 'job_params', 'job_random', 'job_attempts', 'job_token', 'job_token_timestamp', 'job_sha1', ); } }