wiki.techinc.nl/includes/jobqueue/JobRunner.php
Aaron Schulz 6aa41cbd37 jobqueue: cleanup JobRunner for reability and code reuse
Consolidate more logic into JobRunner::execute() and make
it public. Add "caught" field to the resulting map. The
intended use case for this method is JobExecutor. Calling
this method from there could cut down on code duplication.

Also:
* Use try/finally to restore state instead of ScopedCallback.
* Use more generic Throwable instead of Exception.
* Reorganize JobRunner::run() slightly for readability.
* Set class constant visibility and improve code comments.

Bug: T243492
Change-Id: I90566a49c603aa78f45b35c0d3fc1925d2cfe2f8
2020-01-29 21:00:39 +00:00

653 lines
22 KiB
PHP

<?php
/**
* Job queue runner utility methods
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License along
* with this program; if not, write to the Free Software Foundation, Inc.,
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
* http://www.gnu.org/copyleft/gpl.html
*
* @file
* @ingroup JobQueue
*/
use Liuggio\StatsdClient\Factory\StatsdDataFactoryInterface;
use MediaWiki\Logger\LoggerFactory;
use MediaWiki\MediaWikiServices;
use Psr\Log\LoggerAwareInterface;
use Psr\Log\LoggerInterface;
use Wikimedia\Rdbms\DBError;
use Wikimedia\Rdbms\DBReadOnlyError;
use Wikimedia\Rdbms\LBFactory;
use Wikimedia\ScopedCallback;
/**
* Job queue runner utility methods
*
* @ingroup JobQueue
* @since 1.24
*/
class JobRunner implements LoggerAwareInterface {
/** @var Config */
protected $config;
/** @var callable|null Debug output handler */
protected $debug;
/** @var LoggerInterface */
protected $logger;
/** @var int Abort if more than this much DB lag is present */
private const MAX_ALLOWED_LAG = 3;
/** @var int An appropriate timeout to balance lag avoidance and job progress */
private const SYNC_TIMEOUT = self::MAX_ALLOWED_LAG;
/** @var float Check replica DB lag this many seconds */
private const LAG_CHECK_PERIOD = 1.0;
/** @var int Seconds to back off a queue due to errors */
private const ERROR_BACKOFF_TTL = 1;
/** @var int Seconds to back off a queue due to read-only errors */
private const READONLY_BACKOFF_TTL = 30;
/**
* @param callable $debug Optional debug output handler
*/
public function setDebugHandler( $debug ) {
$this->debug = $debug;
}
/**
* @param LoggerInterface $logger
* @return void
*/
public function setLogger( LoggerInterface $logger ) {
$this->logger = $logger;
}
/**
* @param LoggerInterface|null $logger
*/
public function __construct( LoggerInterface $logger = null ) {
if ( $logger === null ) {
$logger = LoggerFactory::getInstance( 'runJobs' );
}
$this->setLogger( $logger );
$this->config = MediaWikiServices::getInstance()->getMainConfig();
}
/**
* Run jobs of the specified number/type for the specified time
*
* The response map has a 'job' field that lists status of each job, including:
* - type : the job/queue type
* - status : ok/failed
* - error : any error message string
* - time : the job run time in ms
* The response map also has:
* - backoffs : the (job/queue type => seconds) map of backoff times
* - elapsed : the total time spent running tasks in ms
* - reached : the reason the script finished, one of (none-ready, job-limit, time-limit,
* memory-limit)
*
* This method outputs status information only if a debug handler was set.
* Any exceptions are caught and logged, but are not reported as output.
*
* @param array $options Map of parameters:
* - type : specified job/queue type (or false for the default types)
* - maxJobs : maximum number of jobs to run
* - maxTime : maximum time in seconds before stopping
* - throttle : whether to respect job backoff configuration
* @return array Summary response that can easily be JSON serialized
*/
public function run( array $options ) {
$type = $options['type'] ?? false;
$maxJobs = $options['maxJobs'] ?? false;
$maxTime = $options['maxTime'] ?? false;
$throttle = $options['throttle'] ?? true;
$services = MediaWikiServices::getInstance();
$jobClasses = $this->config->get( 'JobClasses' );
$profilerLimits = $this->config->get( 'TrxProfilerLimits' );
$response = [ 'jobs' => [], 'reached' => 'none-ready' ];
if ( $type !== false && !isset( $jobClasses[$type] ) ) {
// Invalid job type specified
$response['reached'] = 'none-possible';
return $response;
}
if ( $services->getReadOnlyMode()->isReadOnly() ) {
// Any jobs popped off the queue might fail to run and thus might end up lost
$response['reached'] = 'read-only';
return $response;
}
$lbFactory = $services->getDBLoadBalancerFactory();
list( , $maxLag ) = $lbFactory->getMainLB()->getMaxLag();
if ( $maxLag >= self::MAX_ALLOWED_LAG ) {
// DB lag is already too high; caller can immediately try other wikis if applicable
$response['reached'] = 'replica-lag-limit';
return $response;
}
// Narrow DB query expectations for this HTTP request
$trxProfiler = Profiler::instance()->getTransactionProfiler();
$trxProfiler->setLogger( LoggerFactory::getInstance( 'DBPerformance' ) );
$trxProfiler->setExpectations( $profilerLimits['JobRunner'], __METHOD__ );
// Error out if an explicit DB transaction round is somehow active
if ( $lbFactory->hasTransactionRound() ) {
throw new LogicException( __METHOD__ . ' called with an active transaction round.' );
}
// Some jobs types should not run until a certain timestamp
$backoffs = []; // map of (type => UNIX expiry)
$backoffDeltas = []; // map of (type => seconds)
$wait = 'wait'; // block to read backoffs the first time
$group = JobQueueGroup::singleton();
$stats = $services->getStatsdDataFactory();
$loopStartTime = microtime( true );
$jobsPopped = 0;
$timeMsTotal = 0;
$lastSyncTime = 1; // initialize "last sync check timestamp" to "ages ago"
// Keep popping and running jobs until there are no more...
do {
// Sync the persistent backoffs with concurrent runners
$backoffs = $this->syncBackoffDeltas( $backoffs, $backoffDeltas, $wait );
$blacklist = $throttle ? array_keys( $backoffs ) : [];
$wait = 'nowait'; // less important now
if ( $type === false ) {
// Treat the default job type queues as a single queue and pop off a job
$job = $group->pop( $group::TYPE_DEFAULT, $group::USE_CACHE, $blacklist );
} else {
// Pop off a job from the specified job type queue unless the execution of
// that type of job is currently rate-limited by the back-off blacklist
$job = in_array( $type, $blacklist ) ? false : $group->pop( $type );
}
if ( $job ) {
++$jobsPopped;
$jType = $job->getType();
// Back off of certain jobs for a while (for throttling and for errors)
$ttw = $this->getBackoffTimeToWait( $job );
if ( $ttw > 0 ) {
// Always add the delta for other runners in case the time running the
// job negated the backoff for each individually but not collectively.
$backoffDeltas[$jType] = ( $backoffDeltas[$jType] ?? 0 ) + $ttw;
$backoffs = $this->syncBackoffDeltas( $backoffs, $backoffDeltas, $wait );
}
$info = $this->executeJob( $job, $lbFactory, $stats );
// Mark completed or "one shot only" jobs as resolved
if ( $info['status'] !== false || !$job->allowRetries() ) {
$group->ack( $job );
}
// Back off of certain jobs for a while (for throttling and for errors)
if ( $info['status'] === false && mt_rand( 0, 49 ) == 0 ) {
$ttw = max( $ttw, $this->getErrorBackoffTTL( $info['caught'] ) );
$backoffDeltas[$jType] = isset( $backoffDeltas[$jType] )
? $backoffDeltas[$jType] + $ttw
: $ttw;
}
$response['jobs'][] = [
'type' => $jType,
'status' => ( $info['status'] === false ) ? 'failed' : 'ok',
'error' => $info['error'],
'time' => $info['timeMs']
];
$timeMsTotal += $info['timeMs'];
// Break out if we hit the job count or wall time limits
if ( $maxJobs && $jobsPopped >= $maxJobs ) {
$response['reached'] = 'job-limit';
break;
} elseif ( $maxTime && ( microtime( true ) - $loopStartTime ) > $maxTime ) {
$response['reached'] = 'time-limit';
break;
}
// Don't let any of the main DB replica DBs get backed up.
// This only waits for so long before exiting and letting
// other wikis in the farm (on different masters) get a chance.
$timePassed = microtime( true ) - $lastSyncTime;
if ( $timePassed >= self::LAG_CHECK_PERIOD || $timePassed < 0 ) {
$opts = [ 'ifWritesSince' => $lastSyncTime, 'timeout' => self::SYNC_TIMEOUT ];
if ( !$lbFactory->waitForReplication( $opts ) ) {
$response['reached'] = 'replica-lag-limit';
break;
}
$lastSyncTime = microtime( true );
}
// Abort if nearing OOM to avoid erroring out in the middle of a job
if ( !$this->checkMemoryOK() ) {
$response['reached'] = 'memory-limit';
break;
}
}
} while ( $job );
// Sync the persistent backoffs for the next runJobs.php pass
if ( $backoffDeltas ) {
$this->syncBackoffDeltas( $backoffs, $backoffDeltas, 'wait' );
}
$response['backoffs'] = $backoffs;
$response['elapsed'] = $timeMsTotal;
return $response;
}
/**
* Run a specific job in a manner appropriate for mass use by job dispatchers
*
* Wraps the job's run() and tearDown() methods into appropriate transaction rounds.
* During execution, SPI-based logging will use the ID of the HTTP request that spawned
* the job (instead of the current one). Large DB write transactions will be subject to
* $wgJobSerialCommitThreshold and $wgMaxJobDBWriteDuration.
*
* This should never be called if there are explicit transaction rounds or pending DB writes
*
* @param RunnableJob $job
* @param LBFactory $lbFactory
* @param StatsdDataFactoryInterface $stats
* @return array Map of:
* - status: boolean; whether the job succeed
* - error: error string; empty if there was no error specified
* - caught: list of FQCNs corresponding to any exceptions caught
* - timeMs: float; job execution time in milliseconds
* @since 1.35
*/
public function executeJob(
RunnableJob $job,
LBFactory $lbFactory,
StatsdDataFactoryInterface $stats
) {
$oldRequestId = WebRequest::getRequestId();
// Temporarily inherit the original ID of the web request that spawned this job
WebRequest::overrideRequestId( $job->getRequestId() );
// Use an appropriate timeout to balance lag avoidance and job progress
$oldTimeout = $lbFactory->setDefaultReplicationWaitTimeout( self::SYNC_TIMEOUT );
try {
return $this->doExecuteJob( $job, $lbFactory, $stats );
} finally {
$lbFactory->setDefaultReplicationWaitTimeout( $oldTimeout );
WebRequest::overrideRequestId( $oldRequestId );
}
}
/**
* @param RunnableJob $job
* @param LBFactory $lbFactory
* @param StatsdDataFactoryInterface $stats
* @return array Map of:
* - status: boolean; whether the job succeed
* - error: error string; empty if there was no error specified
* - caught: list of FQCNs corresponding to any exceptions caught
* - timeMs: float; job execution time in milliseconds
*/
private function doExecuteJob(
RunnableJob $job,
LBFactory $lbFactory,
StatsdDataFactoryInterface $stats
) {
$jType = $job->getType();
$msg = $job->toString() . " STARTING";
$this->logger->debug( $msg, [ 'job_type' => $job->getType() ] );
$this->debugCallback( $msg );
// Clear out title cache data from prior snapshots
// (e.g. from before JobRunner was invoked in this process)
MediaWikiServices::getInstance()->getLinkCache()->clear();
// Run the job...
$caught = [];
$rssStart = $this->getMaxRssKb();
$jobStartTime = microtime( true );
try {
$fnameTrxOwner = get_class( $job ) . '::run'; // give run() outer scope
// Flush any pending changes left over from an implicit transaction round
if ( $job->hasExecutionFlag( $job::JOB_NO_EXPLICIT_TRX_ROUND ) ) {
$lbFactory->commitMasterChanges( $fnameTrxOwner ); // new implicit round
} else {
$lbFactory->beginMasterChanges( $fnameTrxOwner ); // new explicit round
}
// Clear any stale REPEATABLE-READ snapshots from replica DB connections
$lbFactory->flushReplicaSnapshots( $fnameTrxOwner );
$status = $job->run();
$error = $job->getLastError();
// Commit all pending changes from this job
$this->commitMasterChanges( $lbFactory, $job, $fnameTrxOwner );
// Run any deferred update tasks; doUpdates() manages transactions itself
DeferredUpdates::doUpdates();
} catch ( Throwable $e ) {
MWExceptionHandler::rollbackMasterChangesAndLog( $e );
$status = false;
$error = get_class( $e ) . ': ' . $e->getMessage();
$caught[] = get_class( $e );
}
// Always attempt to call teardown(), even if Job throws exception
try {
$job->tearDown( $status );
} catch ( Throwable $e ) {
MWExceptionHandler::logException( $e );
}
$timeMs = intval( ( microtime( true ) - $jobStartTime ) * 1000 );
$rssEnd = $this->getMaxRssKb();
// Record how long jobs wait before getting popped
$readyTs = $job->getReadyTimestamp();
if ( $readyTs ) {
$pickupDelay = max( 0, $jobStartTime - $readyTs );
$stats->timing( 'jobqueue.pickup_delay.all', 1000 * $pickupDelay );
$stats->timing( "jobqueue.pickup_delay.$jType", 1000 * $pickupDelay );
}
// Record root job age for jobs being run
$rootTimestamp = $job->getRootJobParams()['rootJobTimestamp'];
if ( $rootTimestamp ) {
$age = max( 0, $jobStartTime - wfTimestamp( TS_UNIX, $rootTimestamp ) );
$stats->timing( "jobqueue.pickup_root_age.$jType", 1000 * $age );
}
// Track the execution time for jobs
$stats->timing( "jobqueue.run.$jType", $timeMs );
// Track RSS increases for jobs (in case of memory leaks)
if ( $rssStart && $rssEnd ) {
$stats->updateCount( "jobqueue.rss_delta.$jType", $rssEnd - $rssStart );
}
if ( $status === false ) {
$msg = $job->toString() . " t={job_duration} error={job_error}";
$this->logger->error( $msg, [
'job_type' => $job->getType(),
'job_duration' => $timeMs,
'job_error' => $error,
] );
$msg = $job->toString() . " t=$timeMs error={$error}";
$this->debugCallback( $msg );
} else {
$msg = $job->toString() . " t={job_duration} good";
$this->logger->info( $msg, [
'job_type' => $job->getType(),
'job_duration' => $timeMs,
] );
$msg = $job->toString() . " t=$timeMs good";
$this->debugCallback( $msg );
}
return [
'status' => $status,
'error' => $error,
'caught' => $caught,
'timeMs' => $timeMs
];
}
/**
* @param string[] $caught List of FQCNs corresponding to any exceptions caught
* @return int TTL in seconds
*/
private function getErrorBackoffTTL( array $caught ) {
return in_array( DBReadOnlyError::class, $caught )
? self::READONLY_BACKOFF_TTL
: self::ERROR_BACKOFF_TTL;
}
/**
* @return int|null Max memory RSS in kilobytes
*/
private function getMaxRssKb() {
$info = getrusage( 0 /* RUSAGE_SELF */ );
// see https://linux.die.net/man/2/getrusage
return isset( $info['ru_maxrss'] ) ? (int)$info['ru_maxrss'] : null;
}
/**
* @param RunnableJob $job
* @return int Seconds for this runner to avoid doing more jobs of this type
* @see $wgJobBackoffThrottling
*/
private function getBackoffTimeToWait( RunnableJob $job ) {
$throttling = $this->config->get( 'JobBackoffThrottling' );
if ( !isset( $throttling[$job->getType()] ) || $job instanceof DuplicateJob ) {
return 0; // not throttled
}
$itemsPerSecond = $throttling[$job->getType()];
if ( $itemsPerSecond <= 0 ) {
return 0; // not throttled
}
$seconds = 0;
if ( $job->workItemCount() > 0 ) {
$exactSeconds = $job->workItemCount() / $itemsPerSecond;
// use randomized rounding
$seconds = floor( $exactSeconds );
$remainder = $exactSeconds - $seconds;
$seconds += ( mt_rand() / mt_getrandmax() < $remainder ) ? 1 : 0;
}
return (int)$seconds;
}
/**
* Get the previous backoff expiries from persistent storage
* On I/O or lock acquisition failure this returns the original $backoffs.
*
* @param array $backoffs Map of (job type => UNIX timestamp)
* @param string $mode Lock wait mode - "wait" or "nowait"
* @return array Map of (job type => backoff expiry timestamp)
*/
private function loadBackoffs( array $backoffs, $mode = 'wait' ) {
$file = wfTempDir() . '/mw-runJobs-backoffs.json';
if ( is_file( $file ) ) {
$noblock = ( $mode === 'nowait' ) ? LOCK_NB : 0;
$handle = fopen( $file, 'rb' );
if ( !flock( $handle, LOCK_SH | $noblock ) ) {
fclose( $handle );
return $backoffs; // don't wait on lock
}
$content = stream_get_contents( $handle );
flock( $handle, LOCK_UN );
fclose( $handle );
$ctime = microtime( true );
$cBackoffs = json_decode( $content, true ) ?: [];
foreach ( $cBackoffs as $type => $timestamp ) {
if ( $timestamp < $ctime ) {
unset( $cBackoffs[$type] );
}
}
} else {
$cBackoffs = [];
}
return $cBackoffs;
}
/**
* Merge the current backoff expiries from persistent storage
*
* The $deltas map is set to an empty array on success.
* On I/O or lock acquisition failure this returns the original $backoffs.
*
* @param array $backoffs Map of (job type => UNIX timestamp)
* @param array &$deltas Map of (job type => seconds)
* @param string $mode Lock wait mode - "wait" or "nowait"
* @return array The new backoffs account for $backoffs and the latest file data
*/
private function syncBackoffDeltas( array $backoffs, array &$deltas, $mode = 'wait' ) {
if ( !$deltas ) {
return $this->loadBackoffs( $backoffs, $mode );
}
$noblock = ( $mode === 'nowait' ) ? LOCK_NB : 0;
$file = wfTempDir() . '/mw-runJobs-backoffs.json';
$handle = fopen( $file, 'wb+' );
if ( !flock( $handle, LOCK_EX | $noblock ) ) {
fclose( $handle );
return $backoffs; // don't wait on lock
}
$ctime = microtime( true );
$content = stream_get_contents( $handle );
$cBackoffs = json_decode( $content, true ) ?: [];
foreach ( $deltas as $type => $seconds ) {
$cBackoffs[$type] = isset( $cBackoffs[$type] ) && $cBackoffs[$type] >= $ctime
? $cBackoffs[$type] + $seconds
: $ctime + $seconds;
}
foreach ( $cBackoffs as $type => $timestamp ) {
if ( $timestamp < $ctime ) {
unset( $cBackoffs[$type] );
}
}
ftruncate( $handle, 0 );
fwrite( $handle, json_encode( $cBackoffs ) );
flock( $handle, LOCK_UN );
fclose( $handle );
$deltas = [];
return $cBackoffs;
}
/**
* Make sure that this script is not too close to the memory usage limit.
* It is better to die in between jobs than OOM right in the middle of one.
* @return bool
*/
private function checkMemoryOK() {
static $maxBytes = null;
if ( $maxBytes === null ) {
$m = [];
if ( preg_match( '!^(\d+)(k|m|g|)$!i', ini_get( 'memory_limit' ), $m ) ) {
list( , $num, $unit ) = $m;
$conv = [ 'g' => 1073741824, 'm' => 1048576, 'k' => 1024, '' => 1 ];
$maxBytes = $num * $conv[strtolower( $unit )];
} else {
$maxBytes = 0;
}
}
$usedBytes = memory_get_usage();
if ( $maxBytes && $usedBytes >= 0.95 * $maxBytes ) {
$msg = "Detected excessive memory usage ({used_bytes}/{max_bytes}).";
$this->logger->error( $msg, [
'used_bytes' => $usedBytes,
'max_bytes' => $maxBytes,
] );
$msg = "Detected excessive memory usage ($usedBytes/$maxBytes).";
$this->debugCallback( $msg );
return false;
}
return true;
}
/**
* Log the job message
* @param string $msg The message to log
*/
private function debugCallback( $msg ) {
if ( $this->debug ) {
call_user_func_array( $this->debug, [ wfTimestamp( TS_DB ) . " $msg\n" ] );
}
}
/**
* Issue a commit on all masters who are currently in a transaction and have
* made changes to the database. It also supports sometimes waiting for the
* local wiki's replica DBs to catch up. See the documentation for
* $wgJobSerialCommitThreshold for more.
*
* @param LBFactory $lbFactory
* @param RunnableJob $job
* @param string $fnameTrxOwner
* @throws DBError
*/
private function commitMasterChanges( LBFactory $lbFactory, RunnableJob $job, $fnameTrxOwner ) {
$syncThreshold = $this->config->get( 'JobSerialCommitThreshold' );
$time = false;
$lb = $lbFactory->getMainLB();
if ( $syncThreshold !== false && $lb->hasStreamingReplicaServers() ) {
// Generally, there is one master connection to the local DB
$dbwSerial = $lb->getAnyOpenConnection( $lb->getWriterIndex() );
// We need natively blocking fast locks
if ( $dbwSerial && $dbwSerial->namedLocksEnqueue() ) {
$time = $dbwSerial->pendingWriteQueryDuration( $dbwSerial::ESTIMATE_DB_APPLY );
if ( $time < $syncThreshold ) {
$dbwSerial = false;
}
} else {
$dbwSerial = false;
}
} else {
// There are no replica DBs or writes are all to foreign DB (we don't handle that)
$dbwSerial = false;
}
if ( !$dbwSerial ) {
$lbFactory->commitMasterChanges(
$fnameTrxOwner,
// Abort if any transaction was too big
[ 'maxWriteDuration' => $this->config->get( 'MaxJobDBWriteDuration' ) ]
);
return;
}
$ms = intval( 1000 * $time );
$msg = $job->toString() . " COMMIT ENQUEUED [{job_commit_write_ms}ms of writes]";
$this->logger->info( $msg, [
'job_type' => $job->getType(),
'job_commit_write_ms' => $ms,
] );
$msg = $job->toString() . " COMMIT ENQUEUED [{$ms}ms of writes]";
$this->debugCallback( $msg );
// Wait for an exclusive lock to commit
if ( !$dbwSerial->lock( 'jobrunner-serial-commit', $fnameTrxOwner, 30 ) ) {
// This will trigger a rollback in the main loop
throw new DBError( $dbwSerial, "Timed out waiting on commit queue." );
}
$unlocker = new ScopedCallback( function () use ( $dbwSerial, $fnameTrxOwner ) {
$dbwSerial->unlock( 'jobrunner-serial-commit', $fnameTrxOwner );
} );
// Wait for the replica DBs to catch up
$pos = $lb->getMasterPos();
if ( $pos ) {
$lb->waitForAll( $pos );
}
// Actually commit the DB master changes
$lbFactory->commitMasterChanges(
$fnameTrxOwner,
// Abort if any transaction was too big
[ 'maxWriteDuration' => $this->config->get( 'MaxJobDBWriteDuration' ) ]
);
ScopedCallback::consume( $unlocker );
}
}