3 * Job queue code for federated queues.
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.
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.
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
21 * @author Aaron Schulz
25 * Class to handle enqueueing and running of background jobs for federated queues
27 * This class allows for queues to be partitioned into smaller queues.
28 * A partition is defined by the configuration for a JobQueue instance.
29 * For example, one can set $wgJobTypeConf['refreshLinks'] to point to a
30 * JobQueueFederated instance, which itself would consist of three JobQueueRedis
31 * instances, each using their own redis server. This would allow for the jobs
32 * to be split (evenly or based on weights) accross multiple servers if a single
33 * server becomes impractical or expensive. Different JobQueue classes can be mixed.
35 * The basic queue configuration (e.g. "order", "claimTTL") of a federated queue
36 * is inherited by the partition queues. Additional configuration defines what
37 * section each wiki is in, what partition queues each section uses (and their weight),
38 * and the JobQueue configuration for each partition. Some sections might only need a
39 * single queue partition, like the sections for groups of small wikis.
41 * If used for performance, then $wgMainCacheType should be set to memcached/redis.
42 * Note that "fifo" cannot be used for the ordering, since the data is distributed.
43 * One can still use "timestamp" instead, as in "roughly timestamp ordered". Also,
44 * queue classes used by this should ignore down servers (with TTL) to avoid slowness.
49 class JobQueueFederated
extends JobQueue
{
51 protected $partitionRing;
53 protected $partitionPushRing;
54 /** @var array (partition name => JobQueue) reverse sorted by weight */
55 protected $partitionQueues = array();
60 /** @var int Maximum number of partitions to try */
61 protected $maxPartitionsTry;
63 const CACHE_TTL_SHORT
= 30; // integer; seconds to cache info without re-validating
64 const CACHE_TTL_LONG
= 300; // integer; seconds to cache info that is kept up to date
68 * - sectionsByWiki : A map of wiki IDs to section names.
69 * Wikis will default to using the section "default".
70 * - partitionsBySection : Map of section names to maps of (partition name => weight).
71 * A section called 'default' must be defined if not all wikis
72 * have explicitly defined sections.
73 * - configByPartition : Map of queue partition names to configuration arrays.
74 * These configuration arrays are passed to JobQueue::factory().
75 * The options set here are overriden by those passed to this
76 * the federated queue itself (e.g. 'order' and 'claimTTL').
77 * - partitionsNoPush : List of partition names that can handle pop() but not push().
78 * This can be used to migrate away from a certain partition.
79 * - maxPartitionsTry : Maximum number of times to attempt job insertion using
80 * different partition queues. This improves availability
81 * during failure, at the cost of added latency and somewhat
82 * less reliable job de-duplication mechanisms.
83 * @param array $params
86 protected function __construct( array $params ) {
87 parent
::__construct( $params );
88 $section = isset( $params['sectionsByWiki'][$this->wiki
] )
89 ?
$params['sectionsByWiki'][$this->wiki
]
91 if ( !isset( $params['partitionsBySection'][$section] ) ) {
92 throw new MWException( "No configuration for section '$section'." );
94 $this->maxPartitionsTry
= isset( $params['maxPartitionsTry'] )
95 ?
$params['maxPartitionsTry']
97 // Get the full partition map
98 $partitionMap = $params['partitionsBySection'][$section];
99 arsort( $partitionMap, SORT_NUMERIC
);
100 // Get the partitions jobs can actually be pushed to
101 $partitionPushMap = $partitionMap;
102 if ( isset( $params['partitionsNoPush'] ) ) {
103 foreach ( $params['partitionsNoPush'] as $partition ) {
104 unset( $partitionPushMap[$partition] );
107 // Get the config to pass to merge into each partition queue config
108 $baseConfig = $params;
109 foreach ( array( 'class', 'sectionsByWiki', 'maxPartitionsTry',
110 'partitionsBySection', 'configByPartition', 'partitionsNoPush' ) as $o
112 unset( $baseConfig[$o] ); // partition queue doesn't care about this
114 // Get the partition queue objects
115 foreach ( $partitionMap as $partition => $w ) {
116 if ( !isset( $params['configByPartition'][$partition] ) ) {
117 throw new MWException( "No configuration for partition '$partition'." );
119 $this->partitionQueues
[$partition] = JobQueue
::factory(
120 $baseConfig +
$params['configByPartition'][$partition] );
122 // Ring of all partitions
123 $this->partitionRing
= new HashRing( $partitionMap );
124 // Get the ring of partitions to push jobs into
125 if ( count( $partitionPushMap ) === count( $partitionMap ) ) {
126 $this->partitionPushRing
= clone $this->partitionRing
; // faster
128 $this->partitionPushRing
= new HashRing( $partitionPushMap );
130 // Aggregate cache some per-queue values if there are multiple partition queues
131 $this->cache
= count( $partitionMap ) > 1 ?
wfGetMainCache() : new EmptyBagOStuff();
134 protected function supportedOrders() {
135 // No FIFO due to partitioning, though "rough timestamp order" is supported
136 return array( 'undefined', 'random', 'timestamp' );
139 protected function optimalOrder() {
140 return 'undefined'; // defer to the partitions
143 protected function supportsDelayedJobs() {
144 return true; // defer checks to the partitions
147 protected function doIsEmpty() {
148 $key = $this->getCacheKey( 'empty' );
150 $isEmpty = $this->cache
->get( $key );
151 if ( $isEmpty === 'true' ) {
153 } elseif ( $isEmpty === 'false' ) {
159 foreach ( $this->partitionQueues
as $queue ) {
161 $empty = $empty && $queue->doIsEmpty();
162 } catch ( JobQueueError
$e ) {
164 MWExceptionHandler
::logException( $e );
167 $this->throwErrorIfAllPartitionsDown( $failed );
169 $this->cache
->add( $key, $empty ?
'true' : 'false', self
::CACHE_TTL_LONG
);
173 protected function doGetSize() {
174 return $this->getCrossPartitionSum( 'size', 'doGetSize' );
177 protected function doGetAcquiredCount() {
178 return $this->getCrossPartitionSum( 'acquiredcount', 'doGetAcquiredCount' );
181 protected function doGetDelayedCount() {
182 return $this->getCrossPartitionSum( 'delayedcount', 'doGetDelayedCount' );
185 protected function doGetAbandonedCount() {
186 return $this->getCrossPartitionSum( 'abandonedcount', 'doGetAbandonedCount' );
190 * @param string $type
191 * @param string $method
194 protected function getCrossPartitionSum( $type, $method ) {
195 $key = $this->getCacheKey( $type );
197 $count = $this->cache
->get( $key );
198 if ( is_int( $count ) ) {
203 foreach ( $this->partitionQueues
as $queue ) {
205 $count +
= $queue->$method();
206 } catch ( JobQueueError
$e ) {
208 MWExceptionHandler
::logException( $e );
211 $this->throwErrorIfAllPartitionsDown( $failed );
213 $this->cache
->set( $key, $count, self
::CACHE_TTL_SHORT
);
218 protected function doBatchPush( array $jobs, $flags ) {
219 // Local ring variable that may be changed to point to a new ring on failure
220 $partitionRing = $this->partitionPushRing
;
221 // Try to insert the jobs and update $partitionsTry on any failures.
222 // Retry to insert any remaning jobs again, ignoring the bad partitions.
224 // @codingStandardsIgnoreStart Generic.CodeAnalysis.ForLoopWithTestFunctionCall.NotAllowed
225 for ( $i = $this->maxPartitionsTry
; $i > 0 && count( $jobsLeft ); --$i ) {
226 // @codingStandardsIgnoreEnd
228 $partitionRing->getLiveRing();
229 } catch ( UnexpectedValueException
$e ) {
230 break; // all servers down; nothing to insert to
232 $jobsLeft = $this->tryJobInsertions( $jobsLeft, $partitionRing, $flags );
234 if ( count( $jobsLeft ) ) {
235 throw new JobQueueError(
236 "Could not insert job(s), {$this->maxPartitionsTry} partitions tried." );
244 * @param HashRing $partitionRing
246 * @throws JobQueueError
247 * @return array List of Job object that could not be inserted
249 protected function tryJobInsertions( array $jobs, HashRing
&$partitionRing, $flags ) {
252 // Because jobs are spread across partitions, per-job de-duplication needs
253 // to use a consistent hash to avoid allowing duplicate jobs per partition.
254 // When inserting a batch of de-duplicated jobs, QOS_ATOMIC is disregarded.
255 $uJobsByPartition = array(); // (partition name => job list)
257 foreach ( $jobs as $key => $job ) {
258 if ( $job->ignoreDuplicates() ) {
259 $sha1 = sha1( serialize( $job->getDeduplicationInfo() ) );
260 $uJobsByPartition[$partitionRing->getLiveLocation( $sha1 )][] = $job;
261 unset( $jobs[$key] );
264 // Get the batches of jobs that are not de-duplicated
265 if ( $flags & self
::QOS_ATOMIC
) {
266 $nuJobBatches = array( $jobs ); // all or nothing
268 // Split the jobs into batches and spread them out over servers if there
269 // are many jobs. This helps keep the partitions even. Otherwise, send all
270 // the jobs to a single partition queue to avoids the extra connections.
271 $nuJobBatches = array_chunk( $jobs, 300 );
274 // Insert the de-duplicated jobs into the queues...
275 foreach ( $uJobsByPartition as $partition => $jobBatch ) {
276 /** @var JobQueue $queue */
277 $queue = $this->partitionQueues
[$partition];
280 $queue->doBatchPush( $jobBatch, $flags | self
::QOS_ATOMIC
);
281 } catch ( JobQueueError
$e ) {
283 MWExceptionHandler
::logException( $e );
286 $key = $this->getCacheKey( 'empty' );
287 $this->cache
->set( $key, 'false', JobQueueDB
::CACHE_TTL_LONG
);
289 if ( !$partitionRing->ejectFromLiveRing( $partition, 5 ) ) { // blacklist
290 throw new JobQueueError( "Could not insert job(s), no partitions available." );
292 $jobsLeft = array_merge( $jobsLeft, $jobBatch ); // not inserted
296 // Insert the jobs that are not de-duplicated into the queues...
297 foreach ( $nuJobBatches as $jobBatch ) {
298 $partition = ArrayUtils
::pickRandom( $partitionRing->getLiveLocationWeights() );
299 $queue = $this->partitionQueues
[$partition];
302 $queue->doBatchPush( $jobBatch, $flags | self
::QOS_ATOMIC
);
303 } catch ( JobQueueError
$e ) {
305 MWExceptionHandler
::logException( $e );
308 $key = $this->getCacheKey( 'empty' );
309 $this->cache
->set( $key, 'false', JobQueueDB
::CACHE_TTL_LONG
);
311 if ( !$partitionRing->ejectFromLiveRing( $partition, 5 ) ) { // blacklist
312 throw new JobQueueError( "Could not insert job(s), no partitions available." );
314 $jobsLeft = array_merge( $jobsLeft, $jobBatch ); // not inserted
321 protected function doPop() {
322 $key = $this->getCacheKey( 'empty' );
324 $isEmpty = $this->cache
->get( $key );
325 if ( $isEmpty === 'true' ) {
329 $partitionsTry = $this->partitionRing
->getLiveLocationWeights(); // (partition => weight)
332 while ( count( $partitionsTry ) ) {
333 $partition = ArrayUtils
::pickRandom( $partitionsTry );
334 if ( $partition === false ) {
335 break; // all partitions at 0 weight
338 /** @var JobQueue $queue */
339 $queue = $this->partitionQueues
[$partition];
341 $job = $queue->pop();
342 } catch ( JobQueueError
$e ) {
344 MWExceptionHandler
::logException( $e );
348 $job->metadata
['QueuePartition'] = $partition;
352 unset( $partitionsTry[$partition] ); // blacklist partition
355 $this->throwErrorIfAllPartitionsDown( $failed );
357 $this->cache
->set( $key, 'true', JobQueueDB
::CACHE_TTL_LONG
);
362 protected function doAck( Job
$job ) {
363 if ( !isset( $job->metadata
['QueuePartition'] ) ) {
364 throw new MWException( "The given job has no defined partition name." );
367 return $this->partitionQueues
[$job->metadata
['QueuePartition']]->ack( $job );
370 protected function doIsRootJobOldDuplicate( Job
$job ) {
371 $params = $job->getRootJobParams();
372 $sigature = $params['rootJobSignature'];
373 $partition = $this->partitionPushRing
->getLiveLocation( $sigature );
375 return $this->partitionQueues
[$partition]->doIsRootJobOldDuplicate( $job );
376 } catch ( JobQueueError
$e ) {
377 if ( $this->partitionPushRing
->ejectFromLiveRing( $partition, 5 ) ) {
378 $partition = $this->partitionPushRing
->getLiveLocation( $sigature );
379 return $this->partitionQueues
[$partition]->doIsRootJobOldDuplicate( $job );
386 protected function doDeduplicateRootJob( Job
$job ) {
387 $params = $job->getRootJobParams();
388 $sigature = $params['rootJobSignature'];
389 $partition = $this->partitionPushRing
->getLiveLocation( $sigature );
391 return $this->partitionQueues
[$partition]->doDeduplicateRootJob( $job );
392 } catch ( JobQueueError
$e ) {
393 if ( $this->partitionPushRing
->ejectFromLiveRing( $partition, 5 ) ) {
394 $partition = $this->partitionPushRing
->getLiveLocation( $sigature );
395 return $this->partitionQueues
[$partition]->doDeduplicateRootJob( $job );
402 protected function doDelete() {
404 /** @var JobQueue $queue */
405 foreach ( $this->partitionQueues
as $queue ) {
408 } catch ( JobQueueError
$e ) {
410 MWExceptionHandler
::logException( $e );
413 $this->throwErrorIfAllPartitionsDown( $failed );
417 protected function doWaitForBackups() {
419 /** @var JobQueue $queue */
420 foreach ( $this->partitionQueues
as $queue ) {
422 $queue->waitForBackups();
423 } catch ( JobQueueError
$e ) {
425 MWExceptionHandler
::logException( $e );
428 $this->throwErrorIfAllPartitionsDown( $failed );
431 protected function doGetPeriodicTasks() {
433 /** @var JobQueue $queue */
434 foreach ( $this->partitionQueues
as $partition => $queue ) {
435 foreach ( $queue->getPeriodicTasks() as $task => $def ) {
436 $tasks["{$partition}:{$task}"] = $def;
443 protected function doFlushCaches() {
444 static $types = array(
452 foreach ( $types as $type ) {
453 $this->cache
->delete( $this->getCacheKey( $type ) );
456 /** @var JobQueue $queue */
457 foreach ( $this->partitionQueues
as $queue ) {
458 $queue->doFlushCaches();
462 public function getAllQueuedJobs() {
463 $iterator = new AppendIterator();
465 /** @var JobQueue $queue */
466 foreach ( $this->partitionQueues
as $queue ) {
467 $iterator->append( $queue->getAllQueuedJobs() );
473 public function getAllDelayedJobs() {
474 $iterator = new AppendIterator();
476 /** @var JobQueue $queue */
477 foreach ( $this->partitionQueues
as $queue ) {
478 $iterator->append( $queue->getAllDelayedJobs() );
484 public function getCoalesceLocationInternal() {
485 return "JobQueueFederated:wiki:{$this->wiki}" .
486 sha1( serialize( array_keys( $this->partitionQueues
) ) );
489 protected function doGetSiblingQueuesWithJobs( array $types ) {
493 /** @var JobQueue $queue */
494 foreach ( $this->partitionQueues
as $queue ) {
496 $nonEmpty = $queue->doGetSiblingQueuesWithJobs( $types );
497 if ( is_array( $nonEmpty ) ) {
498 $result = array_unique( array_merge( $result, $nonEmpty ) );
500 return null; // not supported on all partitions; bail
502 if ( count( $result ) == count( $types ) ) {
503 break; // short-circuit
505 } catch ( JobQueueError
$e ) {
507 MWExceptionHandler
::logException( $e );
510 $this->throwErrorIfAllPartitionsDown( $failed );
512 return array_values( $result );
515 protected function doGetSiblingQueueSizes( array $types ) {
518 /** @var JobQueue $queue */
519 foreach ( $this->partitionQueues
as $queue ) {
521 $sizes = $queue->doGetSiblingQueueSizes( $types );
522 if ( is_array( $sizes ) ) {
523 foreach ( $sizes as $type => $size ) {
524 $result[$type] = isset( $result[$type] ) ?
$result[$type] +
$size : $size;
527 return null; // not supported on all partitions; bail
529 } catch ( JobQueueError
$e ) {
531 MWExceptionHandler
::logException( $e );
534 $this->throwErrorIfAllPartitionsDown( $failed );
540 * Throw an error if no partitions available
542 * @param int $down The number of up partitions down
544 * @throws JobQueueError
546 protected function throwErrorIfAllPartitionsDown( $down ) {
547 if ( $down >= count( $this->partitionQueues
) ) {
548 throw new JobQueueError( 'No queue partitions available.' );
552 public function setTestingPrefix( $key ) {
553 /** @var JobQueue $queue */
554 foreach ( $this->partitionQueues
as $queue ) {
555 $queue->setTestingPrefix( $key );
560 * @param string $property
563 private function getCacheKey( $property ) {
564 list( $db, $prefix ) = wfSplitWikiID( $this->wiki
);
566 return wfForeignMemcKey( $db, $prefix, 'jobqueue', $this->type
, $property );