X-Git-Url: https://git.cyclocoop.org/%242?a=blobdiff_plain;f=includes%2Fjob%2FJobQueueDB.php;h=4d0f294bc0478a7ab344d10c26f4418105b88dae;hb=2df4cb35478f9946b3538a8b0ddc86b2dfad2124;hp=cbb239156557a1126413f44f9f64e8c0399979a3;hpb=34a65359cca7e6afe7480494973e3bbd853c7b19;p=lhc%2Fweb%2Fwiklou.git diff --git a/includes/job/JobQueueDB.php b/includes/job/JobQueueDB.php index cbb2391565..4d0f294bc0 100644 --- a/includes/job/JobQueueDB.php +++ b/includes/job/JobQueueDB.php @@ -49,54 +49,79 @@ class JobQueueDB extends JobQueue { return false; } - $found = $this->getSlaveDB()->selectField( - 'job', '1', array( 'job_cmd' => $this->type ), __METHOD__ + $found = $this->getSlaveDB()->selectField( // unclaimed job + 'job', '1', array( 'job_cmd' => $this->type, 'job_token' => '' ), __METHOD__ ); $wgMemc->add( $key, $found ? 'false' : 'true', self::CACHE_TTL ); + return (bool)$found; } /** * @see JobQueue::doBatchPush() + * @param array $jobs + * @param $flags + * @throws DBError|Exception * @return bool */ protected function doBatchPush( array $jobs, $flags ) { if ( count( $jobs ) ) { $dbw = $this->getMasterDB(); - $rows = array(); + $rowSet = array(); // (sha1 => job) map for jobs that are de-duplicated + $rowList = array(); // list of jobs for jobs that are are not de-duplicated + foreach ( $jobs as $job ) { - $rows[] = $this->insertFields( $job ); + $row = $this->insertFields( $job ); + if ( $job->ignoreDuplicates() ) { + $rowSet[$row['job_sha1']] = $row; + } else { + $rowList[] = $row; + } } + $atomic = ( $flags & self::QoS_Atomic ); $key = $this->getEmptinessCacheKey(); $ttl = self::CACHE_TTL; - $dbw->onTransactionIdle( function() use ( $dbw, $rows, $atomic, $key, $ttl ) { + $dbw->onTransactionIdle( + function() use ( $dbw, $rowSet, $rowList, $atomic, $key, $ttl + ) { global $wgMemc; - $autoTrx = $dbw->getFlag( DBO_TRX ); // automatic begin() enabled? if ( $atomic ) { - $dbw->begin(); // wrap all the job additions in one transaction - } else { - $dbw->clearFlag( DBO_TRX ); // make each query its own transaction + $dbw->begin( __METHOD__ ); // wrap all the job additions in one transaction } try { - foreach ( array_chunk( $rows, 50 ) as $rowBatch ) { // avoid slave lag + // 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." ); + 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__ ); } } catch ( DBError $e ) { if ( $atomic ) { - $dbw->rollback(); - } else { - $dbw->setFlag( $autoTrx ? DBO_TRX : 0 ); // restore automatic begin() + $dbw->rollback( __METHOD__ ); } throw $e; } if ( $atomic ) { - $dbw->commit(); - } else { - $dbw->setFlag( $autoTrx ? DBO_TRX : 0 ); // restore automatic begin() + $dbw->commit( __METHOD__ ); } $wgMemc->set( $key, 'false', $ttl ); // queue is not empty @@ -122,56 +147,44 @@ class JobQueueDB extends JobQueue { $uuid = wfRandomString( 32 ); // pop attempt $job = false; // job popped off - $autoTrx = $dbw->getFlag( DBO_TRX ); // automatic begin() enabled? - $dbw->clearFlag( DBO_TRX ); // make each query its own transaction - try { - // Occasionally recycle jobs back into the queue that have been claimed too long - if ( mt_rand( 0, 99 ) == 0 ) { - $this->recycleStaleJobs(); - } - 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, $gte ); - if ( !$row ) { // need to try the other direction - $row = $this->claimRandom( $uuid, $rand, !$gte ); - } - } - // Check if we found a row to reserve... - if ( !$row ) { - $wgMemc->set( $this->getEmptinessCacheKey(), 'true', self::CACHE_TTL ); - break; // nothing to do - } - // Get the job object from the row... - $title = Title::makeTitleSafe( $row->job_namespace, $row->job_title ); - if ( !$title ) { - $dbw->delete( 'job', array( 'job_id' => $row->job_id ), __METHOD__ ); - wfIncrStats( 'job-pop' ); - wfDebugLog( 'JobQueueDB', "Row has invalid title '{$row->job_title}'." ); - continue; // try again - } - $job = Job::factory( $row->job_cmd, $title, - self::extractBlob( $row->job_params ), $row->job_id ); - // Delete any *other* duplicate jobs in the queue... - if ( $job->ignoreDuplicates() && strlen( $row->job_sha1 ) ) { - $dbw->delete( 'job', - array( 'job_sha1' => $row->job_sha1, - "job_id != {$dbw->addQuotes( $row->job_id )}" ), - __METHOD__ - ); - wfIncrStats( 'job-pop', $dbw->affectedRows() ); - } - break; // done - } while( true ); - } catch ( DBError $e ) { - $dbw->setFlag( $autoTrx ? DBO_TRX : 0 ); // restore automatic begin() - throw $e; + // Occasionally recycle jobs back into the queue that have been claimed too long + if ( mt_rand( 0, 99 ) == 0 ) { + $this->recycleStaleJobs(); } - $dbw->setFlag( $autoTrx ? DBO_TRX : 0 ); // restore automatic begin() + 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, $gte ); + if ( !$row ) { // need to try the other direction + $row = $this->claimRandom( $uuid, $rand, !$gte ); + } + } + // Check if we found a row to reserve... + if ( !$row ) { + $wgMemc->set( $this->getEmptinessCacheKey(), 'true', self::CACHE_TTL ); + break; // nothing to do + } + // Get the job object from the row... + $title = Title::makeTitleSafe( $row->job_namespace, $row->job_title ); + if ( !$title ) { + $dbw->delete( 'job', array( 'job_id' => $row->job_id ), __METHOD__ ); + wfIncrStats( 'job-pop' ); + wfDebugLog( 'JobQueueDB', "Row has invalid title '{$row->job_title}'." ); + continue; // try again + } + $job = Job::factory( $row->job_cmd, $title, + self::extractBlob( $row->job_params ), $row->job_id ); + $job->id = $row->job_id; // XXX: work around broken subclasses + // Flag this job as an old duplicate based on its "root" job... + if ( $this->isRootJobOldDuplicate( $job ) ) { + $job = DuplicateJob::newFromJob( $job ); // convert to a no-op + } + break; // done + } while( true ); return $job; } @@ -296,26 +309,42 @@ class JobQueueDB extends JobQueue { $dbw = $this->getMasterDB(); $count = 0; // affected rows - if ( $this->claimTTL > 0 ) { // re-try stale jobs... + if ( !$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 ); - // 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. - $dbw->update( 'job', - array( - 'job_token' => '', - 'job_token_timestamp' => $dbw->timestamp( $now ) ), // time of release + // 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( self::MAX_ATTEMPTS )}" ), + "job_attempts < {$dbw->addQuotes( self::MAX_ATTEMPTS )}" ), // retries left __METHOD__ ); - $count += $dbw->affectedRows(); + $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__ + ); + $count += $dbw->affectedRows(); + } } - // Just destroy stale jobs... + // Just destroy any stale jobs... $pruneCutoff = $dbw->timestamp( $now - self::MAX_AGE_PRUNE ); $conds = array( 'job_cmd' => $this->type, @@ -325,34 +354,99 @@ class JobQueueDB extends JobQueue { if ( $this->claimTTL > 0 ) { // only prune jobs attempted too many times... $conds[] = "job_attempts >= {$dbw->addQuotes( self::MAX_ATTEMPTS )}"; } - $dbw->delete( 'job', $conds, __METHOD__ ); - $count += $dbw->affectedRows(); + // 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__ ); + $count += $dbw->affectedRows(); + } + + $dbw->unlock( "jobqueue-recycle-{$this->type}", __METHOD__ ); return $count; } /** * @see JobQueue::doAck() + * @param Job $job + * @throws MWException * @return Job|bool */ protected function doAck( Job $job ) { + if ( !$job->getId() ) { + throw new MWException( "Job of type '{$job->getType()}' has no ID." ); + } + $dbw = $this->getMasterDB(); $dbw->commit( __METHOD__, 'flush' ); // flush existing transaction - $autoTrx = $dbw->getFlag( DBO_TRX ); // automatic begin() enabled? - $dbw->clearFlag( DBO_TRX ); // make each query its own transaction - try { - // Delete a row with a single DELETE without holding row locks over RTTs... - $dbw->delete( 'job', array( 'job_cmd' => $this->type, 'job_id' => $job->getId() ) ); - } catch ( Exception $e ) { - $dbw->setFlag( $autoTrx ? DBO_TRX : 0 ); // restore automatic begin() - throw $e; + // Delete a row with a single DELETE without holding row locks over RTTs... + $dbw->delete( 'job', + array( 'job_cmd' => $this->type, 'job_id' => $job->getId() ), __METHOD__ ); + + 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'." ); } - $dbw->setFlag( $autoTrx ? DBO_TRX : 0 ); // restore automatic begin() + $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 that 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. + $this->getMasterDB()->onTransactionIdle( function() use ( $params, $key ) { + global $wgMemc; + + $timestamp = $wgMemc->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 $wgMemc->set( $key, $params['rootJobTimestamp'], 14*86400 ); // 2 weeks + } ); return true; } + /** + * Check if the "root" job of a given job has been superseded by a newer one + * + * @param $job Job + * @return bool + */ + protected function isRootJobOldDuplicate( Job $job ) { + global $wgMemc; + + $params = $job->getParams(); + if ( !isset( $params['rootJobSignature'] ) ) { + return false; // job has no de-deplication info + } elseif ( !isset( $params['rootJobTimestamp'] ) ) { + trigger_error( "Cannot check root job; missing 'rootJobTimestamp'." ); + return false; + } + + // Get the last time this root job was enqueued + $timestamp = $wgMemc->get( $this->getRootJobCacheKey( $params['rootJobSignature'] ) ); + + // Check if a new root job was started at the location after this one's... + return ( $timestamp && $timestamp > $params['rootJobTimestamp'] ); + } + /** * @see JobQueue::doWaitForBackups() * @return void @@ -380,22 +474,22 @@ class JobQueueDB extends JobQueue { * @return array */ protected function insertFields( Job $job ) { - // Rows that describe the nature of the job - $descFields = array( + $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 - $dbw = $this->getMasterDB(); - $metaFields = array( + // Additional job metadata 'job_id' => $dbw->nextSequenceValue( 'job_job_id_seq' ), 'job_timestamp' => $dbw->timestamp(), - 'job_sha1' => wfBaseConvert( sha1( serialize( $descFields ) ), 16, 36, 32 ), + 'job_sha1' => wfBaseConvert( + sha1( serialize( $job->getDeduplicationInfo() ) ), + 16, 36, 31 + ), 'job_random' => mt_rand( 0, self::MAX_JOB_RANDOM ) ); - return ( $descFields + $metaFields ); } /** @@ -406,6 +500,15 @@ class JobQueueDB extends JobQueue { return wfForeignMemcKey( $db, $prefix, 'jobqueue', $this->type, 'isempty' ); } + /** + * @param string $signature Hash identifier of the root job + * @return string + */ + private function getRootJobCacheKey( $signature ) { + list( $db, $prefix ) = wfSplitWikiID( $this->wiki ); + return wfForeignMemcKey( $db, $prefix, 'jobqueue', $this->type, 'rootjob', $signature ); + } + /** * @param $params * @return string