7f4300ca333ba6ba5ec2d0df501d3d4220f27a5e
[lhc/web/wiklou.git] / includes / job / JobQueueDB.php
1 <?php
2 /**
3 * Database-backed job queue code.
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 * @author Aaron Schulz
22 */
23
24 /**
25 * Class to handle job queues stored in the DB
26 *
27 * @ingroup JobQueue
28 * @since 1.21
29 */
30 class JobQueueDB extends JobQueue {
31 const CACHE_TTL_SHORT = 30; // integer; seconds to cache info without re-validating
32 const CACHE_TTL_LONG = 300; // integer; seconds to cache info that is kept up to date
33 const MAX_AGE_PRUNE = 604800; // integer; seconds a job can live once claimed
34 const MAX_ATTEMPTS = 3; // integer; number of times to try a job
35 const MAX_JOB_RANDOM = 2147483647; // integer; 2^31 - 1, used for job_random
36 const MAX_OFFSET = 255; // integer; maximum number of rows to skip
37
38 protected $cluster = false; // string; name of an external DB cluster
39
40 /**
41 * Additional parameters include:
42 * - cluster : The name of an external cluster registered via LBFactory.
43 * If not specified, the primary DB cluster for the wiki will be used.
44 * This can be overridden with a custom cluster so that DB handles will
45 * be retrieved via LBFactory::getExternalLB() and getConnection().
46 * @param $params array
47 */
48 protected function __construct( array $params ) {
49 parent::__construct( $params );
50 $this->cluster = isset( $params['cluster'] ) ? $params['cluster'] : false;
51 }
52
53 /**
54 * @see JobQueue::doIsEmpty()
55 * @return bool
56 */
57 protected function doIsEmpty() {
58 global $wgMemc;
59
60 $key = $this->getCacheKey( 'empty' );
61
62 $isEmpty = $wgMemc->get( $key );
63 if ( $isEmpty === 'true' ) {
64 return true;
65 } elseif ( $isEmpty === 'false' ) {
66 return false;
67 }
68
69 list( $dbr, $scope ) = $this->getSlaveDB();
70 $found = $dbr->selectField( // unclaimed job
71 'job', '1', array( 'job_cmd' => $this->type, 'job_token' => '' ), __METHOD__
72 );
73 $wgMemc->add( $key, $found ? 'false' : 'true', self::CACHE_TTL_LONG );
74
75 return !$found;
76 }
77
78 /**
79 * @see JobQueue::doGetSize()
80 * @return integer
81 */
82 protected function doGetSize() {
83 global $wgMemc;
84
85 $key = $this->getCacheKey( 'size' );
86
87 $size = $wgMemc->get( $key );
88 if ( is_int( $size ) ) {
89 return $size;
90 }
91
92 $dbr = $this->getSlaveDB();
93 $size = (int)$dbr->selectField( 'job', 'COUNT(*)',
94 array( 'job_cmd' => $this->type, 'job_token' => '' ),
95 __METHOD__
96 );
97 $wgMemc->set( $key, $size, self::CACHE_TTL_SHORT );
98
99 return $size;
100 }
101
102 /**
103 * @see JobQueue::doGetAcquiredCount()
104 * @return integer
105 */
106 protected function doGetAcquiredCount() {
107 global $wgMemc;
108
109 $key = $this->getCacheKey( 'acquiredcount' );
110
111 $count = $wgMemc->get( $key );
112 if ( is_int( $count ) ) {
113 return $count;
114 }
115
116 $dbr = $this->getSlaveDB();
117 $count = (int)$dbr->selectField( 'job', 'COUNT(*)',
118 array( 'job_cmd' => $this->type, "job_token !={$dbr->addQuotes('')}" ),
119 __METHOD__
120 );
121 $wgMemc->set( $key, $count, self::CACHE_TTL_SHORT );
122
123 return $count;
124 }
125
126 /**
127 * @see JobQueue::doBatchPush()
128 * @param array $jobs
129 * @param $flags
130 * @throws DBError|Exception
131 * @return bool
132 */
133 protected function doBatchPush( array $jobs, $flags ) {
134 if ( count( $jobs ) ) {
135 list( $dbw, $scope ) = $this->getMasterDB();
136
137 $rowSet = array(); // (sha1 => job) map for jobs that are de-duplicated
138 $rowList = array(); // list of jobs for jobs that are are not de-duplicated
139
140 foreach ( $jobs as $job ) {
141 $row = $this->insertFields( $job );
142 if ( $job->ignoreDuplicates() ) {
143 $rowSet[$row['job_sha1']] = $row;
144 } else {
145 $rowList[] = $row;
146 }
147 }
148
149 $atomic = ( $flags & self::QoS_Atomic );
150 $key = $this->getCacheKey( 'empty' );
151 $ttl = self::CACHE_TTL_LONG;
152
153 $dbw->onTransactionIdle(
154 function() use ( $dbw, $rowSet, $rowList, $atomic, $key, $ttl, $scope
155 ) {
156 global $wgMemc;
157
158 if ( $atomic ) {
159 $dbw->begin( __METHOD__ ); // wrap all the job additions in one transaction
160 }
161 try {
162 // Strip out any duplicate jobs that are already in the queue...
163 if ( count( $rowSet ) ) {
164 $res = $dbw->select( 'job', 'job_sha1',
165 array(
166 // No job_type condition since it's part of the job_sha1 hash
167 'job_sha1' => array_keys( $rowSet ),
168 'job_token' => '' // unclaimed
169 ),
170 __METHOD__
171 );
172 foreach ( $res as $row ) {
173 wfDebug( "Job with hash '{$row->job_sha1}' is a duplicate." );
174 unset( $rowSet[$row->job_sha1] ); // already enqueued
175 }
176 }
177 // Build the full list of job rows to insert
178 $rows = array_merge( $rowList, array_values( $rowSet ) );
179 // Insert the job rows in chunks to avoid slave lag...
180 foreach ( array_chunk( $rows, 50 ) as $rowBatch ) {
181 $dbw->insert( 'job', $rowBatch, __METHOD__ );
182 }
183 wfIncrStats( 'job-insert', count( $rows ) );
184 } catch ( DBError $e ) {
185 if ( $atomic ) {
186 $dbw->rollback( __METHOD__ );
187 }
188 throw $e;
189 }
190 if ( $atomic ) {
191 $dbw->commit( __METHOD__ );
192 }
193
194 $wgMemc->set( $key, 'false', $ttl ); // queue is not empty
195 } );
196 }
197
198 return true;
199 }
200
201 /**
202 * @see JobQueue::doPop()
203 * @return Job|bool
204 */
205 protected function doPop() {
206 global $wgMemc;
207
208 if ( $wgMemc->get( $this->getCacheKey( 'empty' ) ) === 'true' ) {
209 return false; // queue is empty
210 }
211
212 list( $dbw, $scope ) = $this->getMasterDB();
213 $dbw->commit( __METHOD__, 'flush' ); // flush existing transaction
214
215 $uuid = wfRandomString( 32 ); // pop attempt
216 $job = false; // job popped off
217 // Occasionally recycle jobs back into the queue that have been claimed too long
218 if ( mt_rand( 0, 99 ) == 0 ) {
219 $this->recycleStaleJobs();
220 }
221 do { // retry when our row is invalid or deleted as a duplicate
222 // Try to reserve a row in the DB...
223 if ( in_array( $this->order, array( 'fifo', 'timestamp' ) ) ) {
224 $row = $this->claimOldest( $uuid );
225 } else { // random first
226 $rand = mt_rand( 0, self::MAX_JOB_RANDOM ); // encourage concurrent UPDATEs
227 $gte = (bool)mt_rand( 0, 1 ); // find rows with rand before/after $rand
228 $row = $this->claimRandom( $uuid, $rand, $gte );
229 }
230 // Check if we found a row to reserve...
231 if ( !$row ) {
232 $wgMemc->set( $this->getCacheKey( 'empty' ), 'true', self::CACHE_TTL_LONG );
233 break; // nothing to do
234 }
235 wfIncrStats( 'job-pop' );
236 // Get the job object from the row...
237 $title = Title::makeTitleSafe( $row->job_namespace, $row->job_title );
238 if ( !$title ) {
239 $dbw->delete( 'job', array( 'job_id' => $row->job_id ), __METHOD__ );
240 wfDebugLog( 'JobQueueDB', "Row has invalid title '{$row->job_title}'." );
241 continue; // try again
242 }
243 $job = Job::factory( $row->job_cmd, $title,
244 self::extractBlob( $row->job_params ), $row->job_id );
245 $job->id = $row->job_id; // XXX: work around broken subclasses
246 // Flag this job as an old duplicate based on its "root" job...
247 if ( $this->isRootJobOldDuplicate( $job ) ) {
248 wfIncrStats( 'job-duplicate' );
249 $job = DuplicateJob::newFromJob( $job ); // convert to a no-op
250 }
251 break; // done
252 } while( true );
253
254 return $job;
255 }
256
257 /**
258 * Reserve a row with a single UPDATE without holding row locks over RTTs...
259 *
260 * @param $uuid string 32 char hex string
261 * @param $rand integer Random unsigned integer (31 bits)
262 * @param $gte bool Search for job_random >= $random (otherwise job_random <= $random)
263 * @return Row|false
264 */
265 protected function claimRandom( $uuid, $rand, $gte ) {
266 global $wgMemc;
267
268 list( $dbw, $scope ) = $this->getMasterDB();
269 // Check cache to see if the queue has <= OFFSET items
270 $tinyQueue = $wgMemc->get( $this->getCacheKey( 'small' ) );
271
272 $row = false; // the row acquired
273 $invertedDirection = false; // whether one job_random direction was already scanned
274 // This uses a replication safe method for acquiring jobs. One could use UPDATE+LIMIT
275 // instead, but that either uses ORDER BY (in which case it deadlocks in MySQL) or is
276 // not replication safe. Due to http://bugs.mysql.com/bug.php?id=6980, subqueries cannot
277 // be used here with MySQL.
278 do {
279 if ( $tinyQueue ) { // queue has <= MAX_OFFSET rows
280 // For small queues, using OFFSET will overshoot and return no rows more often.
281 // Instead, this uses job_random to pick a row (possibly checking both directions).
282 $ineq = $gte ? '>=' : '<=';
283 $dir = $gte ? 'ASC' : 'DESC';
284 $row = $dbw->selectRow( 'job', '*', // find a random job
285 array(
286 'job_cmd' => $this->type,
287 'job_token' => '', // unclaimed
288 "job_random {$ineq} {$dbw->addQuotes( $rand )}" ),
289 __METHOD__,
290 array( 'ORDER BY' => "job_random {$dir}" )
291 );
292 if ( !$row && !$invertedDirection ) {
293 $gte = !$gte;
294 $invertedDirection = true;
295 continue; // try the other direction
296 }
297 } else { // table *may* have >= MAX_OFFSET rows
298 // Bug 42614: "ORDER BY job_random" with a job_random inequality causes high CPU
299 // in MySQL if there are many rows for some reason. This uses a small OFFSET
300 // instead of job_random for reducing excess claim retries.
301 $row = $dbw->selectRow( 'job', '*', // find a random job
302 array(
303 'job_cmd' => $this->type,
304 'job_token' => '', // unclaimed
305 ),
306 __METHOD__,
307 array( 'OFFSET' => mt_rand( 0, self::MAX_OFFSET ) )
308 );
309 if ( !$row ) {
310 $tinyQueue = true; // we know the queue must have <= MAX_OFFSET rows
311 $wgMemc->set( $this->getCacheKey( 'small' ), 1, 30 );
312 continue; // use job_random
313 }
314 }
315 if ( $row ) { // claim the job
316 $dbw->update( 'job', // update by PK
317 array(
318 'job_token' => $uuid,
319 'job_token_timestamp' => $dbw->timestamp(),
320 'job_attempts = job_attempts+1' ),
321 array( 'job_cmd' => $this->type, 'job_id' => $row->job_id, 'job_token' => '' ),
322 __METHOD__
323 );
324 // This might get raced out by another runner when claiming the previously
325 // selected row. The use of job_random should minimize this problem, however.
326 if ( !$dbw->affectedRows() ) {
327 $row = false; // raced out
328 }
329 } else {
330 break; // nothing to do
331 }
332 } while ( !$row );
333
334 return $row;
335 }
336
337 /**
338 * Reserve a row with a single UPDATE without holding row locks over RTTs...
339 *
340 * @param $uuid string 32 char hex string
341 * @return Row|false
342 */
343 protected function claimOldest( $uuid ) {
344 list( $dbw, $scope ) = $this->getMasterDB();
345
346 $row = false; // the row acquired
347 do {
348 if ( $dbw->getType() === 'mysql' ) {
349 // Per http://bugs.mysql.com/bug.php?id=6980, we can't use subqueries on the
350 // same table being changed in an UPDATE query in MySQL (gives Error: 1093).
351 // Oracle and Postgre have no such limitation. However, MySQL offers an
352 // alternative here by supporting ORDER BY + LIMIT for UPDATE queries.
353 $dbw->query( "UPDATE {$dbw->tableName( 'job' )} " .
354 "SET " .
355 "job_token = {$dbw->addQuotes( $uuid ) }, " .
356 "job_token_timestamp = {$dbw->addQuotes( $dbw->timestamp() )}, " .
357 "job_attempts = job_attempts+1 " .
358 "WHERE ( " .
359 "job_cmd = {$dbw->addQuotes( $this->type )} " .
360 "AND job_token = {$dbw->addQuotes( '' )} " .
361 ") ORDER BY job_id ASC LIMIT 1",
362 __METHOD__
363 );
364 } else {
365 // Use a subquery to find the job, within an UPDATE to claim it.
366 // This uses as much of the DB wrapper functions as possible.
367 $dbw->update( 'job',
368 array(
369 'job_token' => $uuid,
370 'job_token_timestamp' => $dbw->timestamp(),
371 'job_attempts = job_attempts+1' ),
372 array( 'job_id = (' .
373 $dbw->selectSQLText( 'job', 'job_id',
374 array( 'job_cmd' => $this->type, 'job_token' => '' ),
375 __METHOD__,
376 array( 'ORDER BY' => 'job_id ASC', 'LIMIT' => 1 ) ) .
377 ')'
378 ),
379 __METHOD__
380 );
381 }
382 // Fetch any row that we just reserved...
383 if ( $dbw->affectedRows() ) {
384 $row = $dbw->selectRow( 'job', '*',
385 array( 'job_cmd' => $this->type, 'job_token' => $uuid ), __METHOD__
386 );
387 if ( !$row ) { // raced out by duplicate job removal
388 wfDebugLog( 'JobQueueDB', "Row deleted as duplicate by another process." );
389 }
390 } else {
391 break; // nothing to do
392 }
393 } while ( !$row );
394
395 return $row;
396 }
397
398 /**
399 * Recycle or destroy any jobs that have been claimed for too long
400 *
401 * @return integer Number of jobs recycled/deleted
402 */
403 protected function recycleStaleJobs() {
404 $now = time();
405 list( $dbw, $scope ) = $this->getMasterDB();
406 $count = 0; // affected rows
407
408 if ( !$dbw->lock( "jobqueue-recycle-{$this->type}", __METHOD__, 1 ) ) {
409 return $count; // already in progress
410 }
411
412 // Remove claims on jobs acquired for too long if enabled...
413 if ( $this->claimTTL > 0 ) {
414 $claimCutoff = $dbw->timestamp( $now - $this->claimTTL );
415 // Get the IDs of jobs that have be claimed but not finished after too long.
416 // These jobs can be recycled into the queue by expiring the claim. Selecting
417 // the IDs first means that the UPDATE can be done by primary key (less deadlocks).
418 $res = $dbw->select( 'job', 'job_id',
419 array(
420 'job_cmd' => $this->type,
421 "job_token != {$dbw->addQuotes( '' )}", // was acquired
422 "job_token_timestamp < {$dbw->addQuotes( $claimCutoff )}", // stale
423 "job_attempts < {$dbw->addQuotes( self::MAX_ATTEMPTS )}" ), // retries left
424 __METHOD__
425 );
426 $ids = array_map( function( $o ) { return $o->job_id; }, iterator_to_array( $res ) );
427 if ( count( $ids ) ) {
428 // Reset job_token for these jobs so that other runners will pick them up.
429 // Set the timestamp to the current time, as it is useful to now that the job
430 // was already tried before (the timestamp becomes the "released" time).
431 $dbw->update( 'job',
432 array(
433 'job_token' => '',
434 'job_token_timestamp' => $dbw->timestamp( $now ) ), // time of release
435 array(
436 'job_id' => $ids ),
437 __METHOD__
438 );
439 $count += $dbw->affectedRows();
440 }
441 }
442
443 // Just destroy any stale jobs...
444 $pruneCutoff = $dbw->timestamp( $now - self::MAX_AGE_PRUNE );
445 $conds = array(
446 'job_cmd' => $this->type,
447 "job_token != {$dbw->addQuotes( '' )}", // was acquired
448 "job_token_timestamp < {$dbw->addQuotes( $pruneCutoff )}" // stale
449 );
450 if ( $this->claimTTL > 0 ) { // only prune jobs attempted too many times...
451 $conds[] = "job_attempts >= {$dbw->addQuotes( self::MAX_ATTEMPTS )}";
452 }
453 // Get the IDs of jobs that are considered stale and should be removed. Selecting
454 // the IDs first means that the UPDATE can be done by primary key (less deadlocks).
455 $res = $dbw->select( 'job', 'job_id', $conds, __METHOD__ );
456 $ids = array_map( function( $o ) { return $o->job_id; }, iterator_to_array( $res ) );
457 if ( count( $ids ) ) {
458 $dbw->delete( 'job', array( 'job_id' => $ids ), __METHOD__ );
459 $count += $dbw->affectedRows();
460 }
461
462 $dbw->unlock( "jobqueue-recycle-{$this->type}", __METHOD__ );
463
464 return $count;
465 }
466
467 /**
468 * @see JobQueue::doAck()
469 * @param Job $job
470 * @throws MWException
471 * @return Job|bool
472 */
473 protected function doAck( Job $job ) {
474 if ( !$job->getId() ) {
475 throw new MWException( "Job of type '{$job->getType()}' has no ID." );
476 }
477
478 list( $dbw, $scope ) = $this->getMasterDB();
479 $dbw->commit( __METHOD__, 'flush' ); // flush existing transaction
480
481 // Delete a row with a single DELETE without holding row locks over RTTs...
482 $dbw->delete( 'job',
483 array( 'job_cmd' => $this->type, 'job_id' => $job->getId() ), __METHOD__ );
484
485 return true;
486 }
487
488 /**
489 * @see JobQueue::doDeduplicateRootJob()
490 * @param Job $job
491 * @throws MWException
492 * @return bool
493 */
494 protected function doDeduplicateRootJob( Job $job ) {
495 $params = $job->getParams();
496 if ( !isset( $params['rootJobSignature'] ) ) {
497 throw new MWException( "Cannot register root job; missing 'rootJobSignature'." );
498 } elseif ( !isset( $params['rootJobTimestamp'] ) ) {
499 throw new MWException( "Cannot register root job; missing 'rootJobTimestamp'." );
500 }
501 $key = $this->getRootJobCacheKey( $params['rootJobSignature'] );
502 // Callers should call batchInsert() and then this function so that if the insert
503 // fails, the de-duplication registration will be aborted. Since the insert is
504 // deferred till "transaction idle", do the same here, so that the ordering is
505 // maintained. Having only the de-duplication registration succeed would cause
506 // jobs to become no-ops without any actual jobs that made them redundant.
507 list( $dbw, $scope ) = $this->getMasterDB();
508 $dbw->onTransactionIdle( function() use ( $params, $key, $scope ) {
509 global $wgMemc;
510
511 $timestamp = $wgMemc->get( $key ); // current last timestamp of this job
512 if ( $timestamp && $timestamp >= $params['rootJobTimestamp'] ) {
513 return true; // a newer version of this root job was enqueued
514 }
515
516 // Update the timestamp of the last root job started at the location...
517 return $wgMemc->set( $key, $params['rootJobTimestamp'], 14*86400 ); // 2 weeks
518 } );
519
520 return true;
521 }
522
523 /**
524 * Check if the "root" job of a given job has been superseded by a newer one
525 *
526 * @param $job Job
527 * @return bool
528 */
529 protected function isRootJobOldDuplicate( Job $job ) {
530 global $wgMemc;
531
532 $params = $job->getParams();
533 if ( !isset( $params['rootJobSignature'] ) ) {
534 return false; // job has no de-deplication info
535 } elseif ( !isset( $params['rootJobTimestamp'] ) ) {
536 trigger_error( "Cannot check root job; missing 'rootJobTimestamp'." );
537 return false;
538 }
539
540 // Get the last time this root job was enqueued
541 $timestamp = $wgMemc->get( $this->getRootJobCacheKey( $params['rootJobSignature'] ) );
542
543 // Check if a new root job was started at the location after this one's...
544 return ( $timestamp && $timestamp > $params['rootJobTimestamp'] );
545 }
546
547 /**
548 * @see JobQueue::doWaitForBackups()
549 * @return void
550 */
551 protected function doWaitForBackups() {
552 wfWaitForSlaves();
553 }
554
555 /**
556 * @return Array (DatabaseBase, ScopedCallback)
557 */
558 protected function getSlaveDB() {
559 return $this->getDB( DB_SLAVE );
560 }
561
562 /**
563 * @return Array (DatabaseBase, ScopedCallback)
564 */
565 protected function getMasterDB() {
566 return $this->getDB( DB_MASTER );
567 }
568
569 /**
570 * @param $index integer (DB_SLAVE/DB_MASTER)
571 * @return Array (DatabaseBase, ScopedCallback)
572 */
573 protected function getDB( $index ) {
574 $lb = ( $this->cluster !== false )
575 ? wfGetLBFactory()->getExternalLB( $this->cluster, $this->wiki )
576 : wfGetLB( $this->wiki );
577 $conn = $lb->getConnection( $index );
578 return array(
579 $conn,
580 new ScopedCallback( function() use ( $lb, $conn ) {
581 $lb->reuseConnection( $conn );
582 } )
583 );
584 }
585
586 /**
587 * @param $job Job
588 * @return array
589 */
590 protected function insertFields( Job $job ) {
591 list( $dbw, $scope ) = $this->getMasterDB();
592 return array(
593 // Fields that describe the nature of the job
594 'job_cmd' => $job->getType(),
595 'job_namespace' => $job->getTitle()->getNamespace(),
596 'job_title' => $job->getTitle()->getDBkey(),
597 'job_params' => self::makeBlob( $job->getParams() ),
598 // Additional job metadata
599 'job_id' => $dbw->nextSequenceValue( 'job_job_id_seq' ),
600 'job_timestamp' => $dbw->timestamp(),
601 'job_sha1' => wfBaseConvert(
602 sha1( serialize( $job->getDeduplicationInfo() ) ),
603 16, 36, 31
604 ),
605 'job_random' => mt_rand( 0, self::MAX_JOB_RANDOM )
606 );
607 }
608
609 /**
610 * @return string
611 */
612 private function getCacheKey( $property ) {
613 list( $db, $prefix ) = wfSplitWikiID( $this->wiki );
614 return wfForeignMemcKey( $db, $prefix, 'jobqueue', $this->type, $property );
615 }
616
617 /**
618 * @param string $signature Hash identifier of the root job
619 * @return string
620 */
621 private function getRootJobCacheKey( $signature ) {
622 list( $db, $prefix ) = wfSplitWikiID( $this->wiki );
623 return wfForeignMemcKey( $db, $prefix, 'jobqueue', $this->type, 'rootjob', $signature );
624 }
625
626 /**
627 * @param $params
628 * @return string
629 */
630 protected static function makeBlob( $params ) {
631 if ( $params !== false ) {
632 return serialize( $params );
633 } else {
634 return '';
635 }
636 }
637
638 /**
639 * @param $blob
640 * @return bool|mixed
641 */
642 protected static function extractBlob( $blob ) {
643 if ( (string)$blob !== '' ) {
644 return unserialize( $blob );
645 } else {
646 return false;
647 }
648 }
649 }