Merge "Convert article delete to use OOUI"
[lhc/web/wiklou.git] / includes / jobqueue / JobRunner.php
1 <?php
2 /**
3 * Job queue runner utility methods
4 *
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.
9 *
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.
14 *
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
19 *
20 * @file
21 * @ingroup JobQueue
22 */
23
24 use MediaWiki\MediaWikiServices;
25 use MediaWiki\Logger\LoggerFactory;
26 use Liuggio\StatsdClient\Factory\StatsdDataFactory;
27 use Psr\Log\LoggerAwareInterface;
28 use Psr\Log\LoggerInterface;
29 use Wikimedia\ScopedCallback;
30 use Wikimedia\Rdbms\LBFactory;
31 use Wikimedia\Rdbms\DBError;
32 use Wikimedia\Rdbms\DBReplicationWaitError;
33
34 /**
35 * Job queue runner utility methods
36 *
37 * @ingroup JobQueue
38 * @since 1.24
39 */
40 class JobRunner implements LoggerAwareInterface {
41 /** @var Config */
42 protected $config;
43 /** @var callable|null Debug output handler */
44 protected $debug;
45
46 /**
47 * @var LoggerInterface $logger
48 */
49 protected $logger;
50
51 const MAX_ALLOWED_LAG = 3; // abort if more than this much DB lag is present
52 const LAG_CHECK_PERIOD = 1.0; // check replica DB lag this many seconds
53 const ERROR_BACKOFF_TTL = 1; // seconds to back off a queue due to errors
54 const READONLY_BACKOFF_TTL = 30; // seconds to back off a queue due to read-only errors
55
56 /**
57 * @param callable $debug Optional debug output handler
58 */
59 public function setDebugHandler( $debug ) {
60 $this->debug = $debug;
61 }
62
63 /**
64 * @param LoggerInterface $logger
65 * @return void
66 */
67 public function setLogger( LoggerInterface $logger ) {
68 $this->logger = $logger;
69 }
70
71 /**
72 * @param LoggerInterface $logger
73 */
74 public function __construct( LoggerInterface $logger = null ) {
75 if ( $logger === null ) {
76 $logger = LoggerFactory::getInstance( 'runJobs' );
77 }
78 $this->setLogger( $logger );
79 $this->config = MediaWikiServices::getInstance()->getMainConfig();
80 }
81
82 /**
83 * Run jobs of the specified number/type for the specified time
84 *
85 * The response map has a 'job' field that lists status of each job, including:
86 * - type : the job type
87 * - status : ok/failed
88 * - error : any error message string
89 * - time : the job run time in ms
90 * The response map also has:
91 * - backoffs : the (job type => seconds) map of backoff times
92 * - elapsed : the total time spent running tasks in ms
93 * - reached : the reason the script finished, one of (none-ready, job-limit, time-limit,
94 * memory-limit)
95 *
96 * This method outputs status information only if a debug handler was set.
97 * Any exceptions are caught and logged, but are not reported as output.
98 *
99 * @param array $options Map of parameters:
100 * - type : the job type (or false for the default types)
101 * - maxJobs : maximum number of jobs to run
102 * - maxTime : maximum time in seconds before stopping
103 * - throttle : whether to respect job backoff configuration
104 * @return array Summary response that can easily be JSON serialized
105 */
106 public function run( array $options ) {
107 $jobClasses = $this->config->get( 'JobClasses' );
108 $profilerLimits = $this->config->get( 'TrxProfilerLimits' );
109
110 $response = [ 'jobs' => [], 'reached' => 'none-ready' ];
111
112 $type = isset( $options['type'] ) ? $options['type'] : false;
113 $maxJobs = isset( $options['maxJobs'] ) ? $options['maxJobs'] : false;
114 $maxTime = isset( $options['maxTime'] ) ? $options['maxTime'] : false;
115 $noThrottle = isset( $options['throttle'] ) && !$options['throttle'];
116
117 // Bail if job type is invalid
118 if ( $type !== false && !isset( $jobClasses[$type] ) ) {
119 $response['reached'] = 'none-possible';
120 return $response;
121 }
122 // Bail out if DB is in read-only mode
123 if ( wfReadOnly() ) {
124 $response['reached'] = 'read-only';
125 return $response;
126 }
127
128 $lbFactory = MediaWikiServices::getInstance()->getDBLoadBalancerFactory();
129 // Bail out if there is too much DB lag.
130 // This check should not block as we want to try other wiki queues.
131 list( , $maxLag ) = $lbFactory->getMainLB( wfWikiID() )->getMaxLag();
132 if ( $maxLag >= self::MAX_ALLOWED_LAG ) {
133 $response['reached'] = 'replica-lag-limit';
134 return $response;
135 }
136
137 // Flush any pending DB writes for sanity
138 $lbFactory->commitAll( __METHOD__ );
139
140 // Catch huge single updates that lead to replica DB lag
141 $trxProfiler = Profiler::instance()->getTransactionProfiler();
142 $trxProfiler->setLogger( LoggerFactory::getInstance( 'DBPerformance' ) );
143 $trxProfiler->setExpectations( $profilerLimits['JobRunner'], __METHOD__ );
144
145 // Some jobs types should not run until a certain timestamp
146 $backoffs = []; // map of (type => UNIX expiry)
147 $backoffDeltas = []; // map of (type => seconds)
148 $wait = 'wait'; // block to read backoffs the first time
149
150 $group = JobQueueGroup::singleton();
151 $stats = MediaWikiServices::getInstance()->getStatsdDataFactory();
152 $jobsPopped = 0;
153 $timeMsTotal = 0;
154 $startTime = microtime( true ); // time since jobs started running
155 $lastCheckTime = 1; // timestamp of last replica DB check
156 do {
157 // Sync the persistent backoffs with concurrent runners
158 $backoffs = $this->syncBackoffDeltas( $backoffs, $backoffDeltas, $wait );
159 $blacklist = $noThrottle ? [] : array_keys( $backoffs );
160 $wait = 'nowait'; // less important now
161
162 if ( $type === false ) {
163 $job = $group->pop(
164 JobQueueGroup::TYPE_DEFAULT,
165 JobQueueGroup::USE_CACHE,
166 $blacklist
167 );
168 } elseif ( in_array( $type, $blacklist ) ) {
169 $job = false; // requested queue in backoff state
170 } else {
171 $job = $group->pop( $type ); // job from a single queue
172 }
173 $lbFactory->commitMasterChanges( __METHOD__ ); // flush any JobQueueDB writes
174
175 if ( $job ) { // found a job
176 ++$jobsPopped;
177 $popTime = time();
178 $jType = $job->getType();
179
180 WebRequest::overrideRequestId( $job->getRequestId() );
181
182 // Back off of certain jobs for a while (for throttling and for errors)
183 $ttw = $this->getBackoffTimeToWait( $job );
184 if ( $ttw > 0 ) {
185 // Always add the delta for other runners in case the time running the
186 // job negated the backoff for each individually but not collectively.
187 $backoffDeltas[$jType] = isset( $backoffDeltas[$jType] )
188 ? $backoffDeltas[$jType] + $ttw
189 : $ttw;
190 $backoffs = $this->syncBackoffDeltas( $backoffs, $backoffDeltas, $wait );
191 }
192
193 $info = $this->executeJob( $job, $lbFactory, $stats, $popTime );
194 if ( $info['status'] !== false || !$job->allowRetries() ) {
195 $group->ack( $job ); // succeeded or job cannot be retried
196 $lbFactory->commitMasterChanges( __METHOD__ ); // flush any JobQueueDB writes
197 }
198
199 // Back off of certain jobs for a while (for throttling and for errors)
200 if ( $info['status'] === false && mt_rand( 0, 49 ) == 0 ) {
201 $ttw = max( $ttw, $this->getErrorBackoffTTL( $info['error'] ) );
202 $backoffDeltas[$jType] = isset( $backoffDeltas[$jType] )
203 ? $backoffDeltas[$jType] + $ttw
204 : $ttw;
205 }
206
207 $response['jobs'][] = [
208 'type' => $jType,
209 'status' => ( $info['status'] === false ) ? 'failed' : 'ok',
210 'error' => $info['error'],
211 'time' => $info['timeMs']
212 ];
213 $timeMsTotal += $info['timeMs'];
214
215 // Break out if we hit the job count or wall time limits...
216 if ( $maxJobs && $jobsPopped >= $maxJobs ) {
217 $response['reached'] = 'job-limit';
218 break;
219 } elseif ( $maxTime && ( microtime( true ) - $startTime ) > $maxTime ) {
220 $response['reached'] = 'time-limit';
221 break;
222 }
223
224 // Don't let any of the main DB replica DBs get backed up.
225 // This only waits for so long before exiting and letting
226 // other wikis in the farm (on different masters) get a chance.
227 $timePassed = microtime( true ) - $lastCheckTime;
228 if ( $timePassed >= self::LAG_CHECK_PERIOD || $timePassed < 0 ) {
229 try {
230 $lbFactory->waitForReplication( [
231 'ifWritesSince' => $lastCheckTime,
232 'timeout' => self::MAX_ALLOWED_LAG
233 ] );
234 } catch ( DBReplicationWaitError $e ) {
235 $response['reached'] = 'replica-lag-limit';
236 break;
237 }
238 $lastCheckTime = microtime( true );
239 }
240 // Don't let any queue replica DBs/backups fall behind
241 if ( $jobsPopped > 0 && ( $jobsPopped % 100 ) == 0 ) {
242 $group->waitForBackups();
243 }
244
245 // Bail if near-OOM instead of in a job
246 if ( !$this->checkMemoryOK() ) {
247 $response['reached'] = 'memory-limit';
248 break;
249 }
250 }
251 } while ( $job ); // stop when there are no jobs
252
253 // Sync the persistent backoffs for the next runJobs.php pass
254 if ( $backoffDeltas ) {
255 $this->syncBackoffDeltas( $backoffs, $backoffDeltas, 'wait' );
256 }
257
258 $response['backoffs'] = $backoffs;
259 $response['elapsed'] = $timeMsTotal;
260
261 return $response;
262 }
263
264 /**
265 * @param string $error
266 * @return int TTL in seconds
267 */
268 private function getErrorBackoffTTL( $error ) {
269 return strpos( $error, 'DBReadOnlyError' ) !== false
270 ? self::READONLY_BACKOFF_TTL
271 : self::ERROR_BACKOFF_TTL;
272 }
273
274 /**
275 * @param Job $job
276 * @param LBFactory $lbFactory
277 * @param StatsdDataFactory $stats
278 * @param float $popTime
279 * @return array Map of status/error/timeMs
280 */
281 private function executeJob( Job $job, LBFactory $lbFactory, $stats, $popTime ) {
282 $jType = $job->getType();
283 $msg = $job->toString() . " STARTING";
284 $this->logger->debug( $msg );
285 $this->debugCallback( $msg );
286
287 // Run the job...
288 $rssStart = $this->getMaxRssKb();
289 $jobStartTime = microtime( true );
290 try {
291 $fnameTrxOwner = get_class( $job ) . '::run'; // give run() outer scope
292 $lbFactory->beginMasterChanges( $fnameTrxOwner );
293 $status = $job->run();
294 $error = $job->getLastError();
295 $this->commitMasterChanges( $lbFactory, $job, $fnameTrxOwner );
296 // Important: this must be the last deferred update added (T100085, T154425)
297 DeferredUpdates::addCallableUpdate( [ JobQueueGroup::class, 'pushLazyJobs' ] );
298 // Run any deferred update tasks; doUpdates() manages transactions itself
299 DeferredUpdates::doUpdates();
300 } catch ( Exception $e ) {
301 MWExceptionHandler::rollbackMasterChangesAndLog( $e );
302 $status = false;
303 $error = get_class( $e ) . ': ' . $e->getMessage();
304 }
305 // Always attempt to call teardown() even if Job throws exception.
306 try {
307 $job->teardown( $status );
308 } catch ( Exception $e ) {
309 MWExceptionHandler::logException( $e );
310 }
311
312 // Commit all outstanding connections that are in a transaction
313 // to get a fresh repeatable read snapshot on every connection.
314 // Note that jobs are still responsible for handling replica DB lag.
315 $lbFactory->flushReplicaSnapshots( __METHOD__ );
316 // Clear out title cache data from prior snapshots
317 MediaWikiServices::getInstance()->getLinkCache()->clear();
318 $timeMs = intval( ( microtime( true ) - $jobStartTime ) * 1000 );
319 $rssEnd = $this->getMaxRssKb();
320
321 // Record how long jobs wait before getting popped
322 $readyTs = $job->getReadyTimestamp();
323 if ( $readyTs ) {
324 $pickupDelay = max( 0, $popTime - $readyTs );
325 $stats->timing( 'jobqueue.pickup_delay.all', 1000 * $pickupDelay );
326 $stats->timing( "jobqueue.pickup_delay.$jType", 1000 * $pickupDelay );
327 }
328 // Record root job age for jobs being run
329 $rootTimestamp = $job->getRootJobParams()['rootJobTimestamp'];
330 if ( $rootTimestamp ) {
331 $age = max( 0, $popTime - wfTimestamp( TS_UNIX, $rootTimestamp ) );
332 $stats->timing( "jobqueue.pickup_root_age.$jType", 1000 * $age );
333 }
334 // Track the execution time for jobs
335 $stats->timing( "jobqueue.run.$jType", $timeMs );
336 // Track RSS increases for jobs (in case of memory leaks)
337 if ( $rssStart && $rssEnd ) {
338 $stats->updateCount( "jobqueue.rss_delta.$jType", $rssEnd - $rssStart );
339 }
340
341 if ( $status === false ) {
342 $msg = $job->toString() . " t=$timeMs error={$error}";
343 $this->logger->error( $msg );
344 $this->debugCallback( $msg );
345 } else {
346 $msg = $job->toString() . " t=$timeMs good";
347 $this->logger->info( $msg );
348 $this->debugCallback( $msg );
349 }
350
351 return [ 'status' => $status, 'error' => $error, 'timeMs' => $timeMs ];
352 }
353
354 /**
355 * @return int|null Max memory RSS in kilobytes
356 */
357 private function getMaxRssKb() {
358 $info = wfGetRusage() ?: [];
359 // see https://linux.die.net/man/2/getrusage
360 return isset( $info['ru_maxrss'] ) ? (int)$info['ru_maxrss'] : null;
361 }
362
363 /**
364 * @param Job $job
365 * @return int Seconds for this runner to avoid doing more jobs of this type
366 * @see $wgJobBackoffThrottling
367 */
368 private function getBackoffTimeToWait( Job $job ) {
369 $throttling = $this->config->get( 'JobBackoffThrottling' );
370
371 if ( !isset( $throttling[$job->getType()] ) || $job instanceof DuplicateJob ) {
372 return 0; // not throttled
373 }
374
375 $itemsPerSecond = $throttling[$job->getType()];
376 if ( $itemsPerSecond <= 0 ) {
377 return 0; // not throttled
378 }
379
380 $seconds = 0;
381 if ( $job->workItemCount() > 0 ) {
382 $exactSeconds = $job->workItemCount() / $itemsPerSecond;
383 // use randomized rounding
384 $seconds = floor( $exactSeconds );
385 $remainder = $exactSeconds - $seconds;
386 $seconds += ( mt_rand() / mt_getrandmax() < $remainder ) ? 1 : 0;
387 }
388
389 return (int)$seconds;
390 }
391
392 /**
393 * Get the previous backoff expiries from persistent storage
394 * On I/O or lock acquisition failure this returns the original $backoffs.
395 *
396 * @param array $backoffs Map of (job type => UNIX timestamp)
397 * @param string $mode Lock wait mode - "wait" or "nowait"
398 * @return array Map of (job type => backoff expiry timestamp)
399 */
400 private function loadBackoffs( array $backoffs, $mode = 'wait' ) {
401 $file = wfTempDir() . '/mw-runJobs-backoffs.json';
402 if ( is_file( $file ) ) {
403 $noblock = ( $mode === 'nowait' ) ? LOCK_NB : 0;
404 $handle = fopen( $file, 'rb' );
405 if ( !flock( $handle, LOCK_SH | $noblock ) ) {
406 fclose( $handle );
407 return $backoffs; // don't wait on lock
408 }
409 $content = stream_get_contents( $handle );
410 flock( $handle, LOCK_UN );
411 fclose( $handle );
412 $ctime = microtime( true );
413 $cBackoffs = json_decode( $content, true ) ?: [];
414 foreach ( $cBackoffs as $type => $timestamp ) {
415 if ( $timestamp < $ctime ) {
416 unset( $cBackoffs[$type] );
417 }
418 }
419 } else {
420 $cBackoffs = [];
421 }
422
423 return $cBackoffs;
424 }
425
426 /**
427 * Merge the current backoff expiries from persistent storage
428 *
429 * The $deltas map is set to an empty array on success.
430 * On I/O or lock acquisition failure this returns the original $backoffs.
431 *
432 * @param array $backoffs Map of (job type => UNIX timestamp)
433 * @param array $deltas Map of (job type => seconds)
434 * @param string $mode Lock wait mode - "wait" or "nowait"
435 * @return array The new backoffs account for $backoffs and the latest file data
436 */
437 private function syncBackoffDeltas( array $backoffs, array &$deltas, $mode = 'wait' ) {
438 if ( !$deltas ) {
439 return $this->loadBackoffs( $backoffs, $mode );
440 }
441
442 $noblock = ( $mode === 'nowait' ) ? LOCK_NB : 0;
443 $file = wfTempDir() . '/mw-runJobs-backoffs.json';
444 $handle = fopen( $file, 'wb+' );
445 if ( !flock( $handle, LOCK_EX | $noblock ) ) {
446 fclose( $handle );
447 return $backoffs; // don't wait on lock
448 }
449 $ctime = microtime( true );
450 $content = stream_get_contents( $handle );
451 $cBackoffs = json_decode( $content, true ) ?: [];
452 foreach ( $deltas as $type => $seconds ) {
453 $cBackoffs[$type] = isset( $cBackoffs[$type] ) && $cBackoffs[$type] >= $ctime
454 ? $cBackoffs[$type] + $seconds
455 : $ctime + $seconds;
456 }
457 foreach ( $cBackoffs as $type => $timestamp ) {
458 if ( $timestamp < $ctime ) {
459 unset( $cBackoffs[$type] );
460 }
461 }
462 ftruncate( $handle, 0 );
463 fwrite( $handle, json_encode( $cBackoffs ) );
464 flock( $handle, LOCK_UN );
465 fclose( $handle );
466
467 $deltas = [];
468
469 return $cBackoffs;
470 }
471
472 /**
473 * Make sure that this script is not too close to the memory usage limit.
474 * It is better to die in between jobs than OOM right in the middle of one.
475 * @return bool
476 */
477 private function checkMemoryOK() {
478 static $maxBytes = null;
479 if ( $maxBytes === null ) {
480 $m = [];
481 if ( preg_match( '!^(\d+)(k|m|g|)$!i', ini_get( 'memory_limit' ), $m ) ) {
482 list( , $num, $unit ) = $m;
483 $conv = [ 'g' => 1073741824, 'm' => 1048576, 'k' => 1024, '' => 1 ];
484 $maxBytes = $num * $conv[strtolower( $unit )];
485 } else {
486 $maxBytes = 0;
487 }
488 }
489 $usedBytes = memory_get_usage();
490 if ( $maxBytes && $usedBytes >= 0.95 * $maxBytes ) {
491 $msg = "Detected excessive memory usage ($usedBytes/$maxBytes).";
492 $this->debugCallback( $msg );
493 $this->logger->error( $msg );
494
495 return false;
496 }
497
498 return true;
499 }
500
501 /**
502 * Log the job message
503 * @param string $msg The message to log
504 */
505 private function debugCallback( $msg ) {
506 if ( $this->debug ) {
507 call_user_func_array( $this->debug, [ wfTimestamp( TS_DB ) . " $msg\n" ] );
508 }
509 }
510
511 /**
512 * Issue a commit on all masters who are currently in a transaction and have
513 * made changes to the database. It also supports sometimes waiting for the
514 * local wiki's replica DBs to catch up. See the documentation for
515 * $wgJobSerialCommitThreshold for more.
516 *
517 * @param LBFactory $lbFactory
518 * @param Job $job
519 * @param string $fnameTrxOwner
520 * @throws DBError
521 */
522 private function commitMasterChanges( LBFactory $lbFactory, Job $job, $fnameTrxOwner ) {
523 $syncThreshold = $this->config->get( 'JobSerialCommitThreshold' );
524
525 $time = false;
526 $lb = $lbFactory->getMainLB( wfWikiID() );
527 if ( $syncThreshold !== false && $lb->getServerCount() > 1 ) {
528 // Generally, there is one master connection to the local DB
529 $dbwSerial = $lb->getAnyOpenConnection( $lb->getWriterIndex() );
530 // We need natively blocking fast locks
531 if ( $dbwSerial && $dbwSerial->namedLocksEnqueue() ) {
532 $time = $dbwSerial->pendingWriteQueryDuration( $dbwSerial::ESTIMATE_DB_APPLY );
533 if ( $time < $syncThreshold ) {
534 $dbwSerial = false;
535 }
536 } else {
537 $dbwSerial = false;
538 }
539 } else {
540 // There are no replica DBs or writes are all to foreign DB (we don't handle that)
541 $dbwSerial = false;
542 }
543
544 if ( !$dbwSerial ) {
545 $lbFactory->commitMasterChanges(
546 $fnameTrxOwner,
547 // Abort if any transaction was too big
548 [ 'maxWriteDuration' => $this->config->get( 'MaxJobDBWriteDuration' ) ]
549 );
550
551 return;
552 }
553
554 $ms = intval( 1000 * $time );
555 $msg = $job->toString() . " COMMIT ENQUEUED [{$ms}ms of writes]";
556 $this->logger->info( $msg );
557 $this->debugCallback( $msg );
558
559 // Wait for an exclusive lock to commit
560 if ( !$dbwSerial->lock( 'jobrunner-serial-commit', __METHOD__, 30 ) ) {
561 // This will trigger a rollback in the main loop
562 throw new DBError( $dbwSerial, "Timed out waiting on commit queue." );
563 }
564 $unlocker = new ScopedCallback( function () use ( $dbwSerial ) {
565 $dbwSerial->unlock( 'jobrunner-serial-commit', __METHOD__ );
566 } );
567
568 // Wait for the replica DBs to catch up
569 $pos = $lb->getMasterPos();
570 if ( $pos ) {
571 $lb->waitForAll( $pos );
572 }
573
574 // Actually commit the DB master changes
575 $lbFactory->commitMasterChanges(
576 $fnameTrxOwner,
577 // Abort if any transaction was too big
578 [ 'maxWriteDuration' => $this->config->get( 'MaxJobDBWriteDuration' ) ]
579 );
580 ScopedCallback::consume( $unlocker );
581 }
582 }