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;
52 /** @var JobQueue[] (partition name => JobQueue) reverse sorted by weight */
53 protected $partitionQueues = [];
55 /** @var int Maximum number of partitions to try */
56 protected $maxPartitionsTry;
59 * @param array $params Possible keys:
60 * - sectionsByWiki : A map of wiki IDs to section names.
61 * Wikis will default to using the section "default".
62 * - partitionsBySection : Map of section names to maps of (partition name => weight).
63 * A section called 'default' must be defined if not all wikis
64 * have explicitly defined sections.
65 * - configByPartition : Map of queue partition names to configuration arrays.
66 * These configuration arrays are passed to JobQueue::factory().
67 * The options set here are overridden by those passed to this
68 * the federated queue itself (e.g. 'order' and 'claimTTL').
69 * - maxPartitionsTry : Maximum number of times to attempt job insertion using
70 * different partition queues. This improves availability
71 * during failure, at the cost of added latency and somewhat
72 * less reliable job de-duplication mechanisms.
75 protected function __construct( array $params ) {
76 parent
::__construct( $params );
77 $section = isset( $params['sectionsByWiki'][$this->wiki
] )
78 ?
$params['sectionsByWiki'][$this->wiki
]
80 if ( !isset( $params['partitionsBySection'][$section] ) ) {
81 throw new MWException( "No configuration for section '$section'." );
83 $this->maxPartitionsTry
= isset( $params['maxPartitionsTry'] )
84 ?
$params['maxPartitionsTry']
86 // Get the full partition map
87 $partitionMap = $params['partitionsBySection'][$section];
88 arsort( $partitionMap, SORT_NUMERIC
);
89 // Get the config to pass to merge into each partition queue config
90 $baseConfig = $params;
91 foreach ( [ 'class', 'sectionsByWiki', 'maxPartitionsTry',
92 'partitionsBySection', 'configByPartition', ] as $o
94 unset( $baseConfig[$o] ); // partition queue doesn't care about this
96 // The class handles all aggregator calls already
97 unset( $baseConfig['aggregator'] );
98 // Get the partition queue objects
99 foreach ( $partitionMap as $partition => $w ) {
100 if ( !isset( $params['configByPartition'][$partition] ) ) {
101 throw new MWException( "No configuration for partition '$partition'." );
103 $this->partitionQueues
[$partition] = JobQueue
::factory(
104 $baseConfig +
$params['configByPartition'][$partition] );
106 // Ring of all partitions
107 $this->partitionRing
= new HashRing( $partitionMap );
110 protected function supportedOrders() {
111 // No FIFO due to partitioning, though "rough timestamp order" is supported
112 return [ 'undefined', 'random', 'timestamp' ];
115 protected function optimalOrder() {
116 return 'undefined'; // defer to the partitions
119 protected function supportsDelayedJobs() {
120 foreach ( $this->partitionQueues
as $queue ) {
121 if ( !$queue->supportsDelayedJobs() ) {
129 protected function doIsEmpty() {
132 foreach ( $this->partitionQueues
as $queue ) {
134 $empty = $empty && $queue->doIsEmpty();
135 } catch ( JobQueueError
$e ) {
137 $this->logException( $e );
140 $this->throwErrorIfAllPartitionsDown( $failed );
145 protected function doGetSize() {
146 return $this->getCrossPartitionSum( 'size', 'doGetSize' );
149 protected function doGetAcquiredCount() {
150 return $this->getCrossPartitionSum( 'acquiredcount', 'doGetAcquiredCount' );
153 protected function doGetDelayedCount() {
154 return $this->getCrossPartitionSum( 'delayedcount', 'doGetDelayedCount' );
157 protected function doGetAbandonedCount() {
158 return $this->getCrossPartitionSum( 'abandonedcount', 'doGetAbandonedCount' );
162 * @param string $type
163 * @param string $method
166 protected function getCrossPartitionSum( $type, $method ) {
169 foreach ( $this->partitionQueues
as $queue ) {
171 $count +
= $queue->$method();
172 } catch ( JobQueueError
$e ) {
174 $this->logException( $e );
177 $this->throwErrorIfAllPartitionsDown( $failed );
182 protected function doBatchPush( array $jobs, $flags ) {
183 // Local ring variable that may be changed to point to a new ring on failure
184 $partitionRing = $this->partitionRing
;
185 // Try to insert the jobs and update $partitionsTry on any failures.
186 // Retry to insert any remaning jobs again, ignoring the bad partitions.
188 // @codingStandardsIgnoreStart Generic.CodeAnalysis.ForLoopWithTestFunctionCall.NotAllowed
189 for ( $i = $this->maxPartitionsTry
; $i > 0 && count( $jobsLeft ); --$i ) {
190 // @codingStandardsIgnoreEnd
192 $partitionRing->getLiveRing();
193 } catch ( UnexpectedValueException
$e ) {
194 break; // all servers down; nothing to insert to
196 $jobsLeft = $this->tryJobInsertions( $jobsLeft, $partitionRing, $flags );
198 if ( count( $jobsLeft ) ) {
199 throw new JobQueueError(
200 "Could not insert job(s), {$this->maxPartitionsTry} partitions tried." );
206 * @param HashRing $partitionRing
208 * @throws JobQueueError
209 * @return array List of Job object that could not be inserted
211 protected function tryJobInsertions( array $jobs, HashRing
&$partitionRing, $flags ) {
214 // Because jobs are spread across partitions, per-job de-duplication needs
215 // to use a consistent hash to avoid allowing duplicate jobs per partition.
216 // When inserting a batch of de-duplicated jobs, QOS_ATOMIC is disregarded.
217 $uJobsByPartition = []; // (partition name => job list)
219 foreach ( $jobs as $key => $job ) {
220 if ( $job->ignoreDuplicates() ) {
221 $sha1 = sha1( serialize( $job->getDeduplicationInfo() ) );
222 $uJobsByPartition[$partitionRing->getLiveLocation( $sha1 )][] = $job;
223 unset( $jobs[$key] );
226 // Get the batches of jobs that are not de-duplicated
227 if ( $flags & self
::QOS_ATOMIC
) {
228 $nuJobBatches = [ $jobs ]; // all or nothing
230 // Split the jobs into batches and spread them out over servers if there
231 // are many jobs. This helps keep the partitions even. Otherwise, send all
232 // the jobs to a single partition queue to avoids the extra connections.
233 $nuJobBatches = array_chunk( $jobs, 300 );
236 // Insert the de-duplicated jobs into the queues...
237 foreach ( $uJobsByPartition as $partition => $jobBatch ) {
238 /** @var JobQueue $queue */
239 $queue = $this->partitionQueues
[$partition];
242 $queue->doBatchPush( $jobBatch, $flags | self
::QOS_ATOMIC
);
243 } catch ( JobQueueError
$e ) {
245 $this->logException( $e );
248 if ( !$partitionRing->ejectFromLiveRing( $partition, 5 ) ) { // blacklist
249 throw new JobQueueError( "Could not insert job(s), no partitions available." );
251 $jobsLeft = array_merge( $jobsLeft, $jobBatch ); // not inserted
255 // Insert the jobs that are not de-duplicated into the queues...
256 foreach ( $nuJobBatches as $jobBatch ) {
257 $partition = ArrayUtils
::pickRandom( $partitionRing->getLiveLocationWeights() );
258 $queue = $this->partitionQueues
[$partition];
261 $queue->doBatchPush( $jobBatch, $flags | self
::QOS_ATOMIC
);
262 } catch ( JobQueueError
$e ) {
264 $this->logException( $e );
267 if ( !$partitionRing->ejectFromLiveRing( $partition, 5 ) ) { // blacklist
268 throw new JobQueueError( "Could not insert job(s), no partitions available." );
270 $jobsLeft = array_merge( $jobsLeft, $jobBatch ); // not inserted
277 protected function doPop() {
278 $partitionsTry = $this->partitionRing
->getLiveLocationWeights(); // (partition => weight)
281 while ( count( $partitionsTry ) ) {
282 $partition = ArrayUtils
::pickRandom( $partitionsTry );
283 if ( $partition === false ) {
284 break; // all partitions at 0 weight
287 /** @var JobQueue $queue */
288 $queue = $this->partitionQueues
[$partition];
290 $job = $queue->pop();
291 } catch ( JobQueueError
$e ) {
293 $this->logException( $e );
297 $job->metadata
['QueuePartition'] = $partition;
301 unset( $partitionsTry[$partition] ); // blacklist partition
304 $this->throwErrorIfAllPartitionsDown( $failed );
309 protected function doAck( Job
$job ) {
310 if ( !isset( $job->metadata
['QueuePartition'] ) ) {
311 throw new MWException( "The given job has no defined partition name." );
314 $this->partitionQueues
[$job->metadata
['QueuePartition']]->ack( $job );
317 protected function doIsRootJobOldDuplicate( Job
$job ) {
318 $signature = $job->getRootJobParams()['rootJobSignature'];
319 $partition = $this->partitionRing
->getLiveLocation( $signature );
321 return $this->partitionQueues
[$partition]->doIsRootJobOldDuplicate( $job );
322 } catch ( JobQueueError
$e ) {
323 if ( $this->partitionRing
->ejectFromLiveRing( $partition, 5 ) ) {
324 $partition = $this->partitionRing
->getLiveLocation( $signature );
325 return $this->partitionQueues
[$partition]->doIsRootJobOldDuplicate( $job );
332 protected function doDeduplicateRootJob( IJobSpecification
$job ) {
333 $signature = $job->getRootJobParams()['rootJobSignature'];
334 $partition = $this->partitionRing
->getLiveLocation( $signature );
336 return $this->partitionQueues
[$partition]->doDeduplicateRootJob( $job );
337 } catch ( JobQueueError
$e ) {
338 if ( $this->partitionRing
->ejectFromLiveRing( $partition, 5 ) ) {
339 $partition = $this->partitionRing
->getLiveLocation( $signature );
340 return $this->partitionQueues
[$partition]->doDeduplicateRootJob( $job );
347 protected function doDelete() {
349 /** @var JobQueue $queue */
350 foreach ( $this->partitionQueues
as $queue ) {
353 } catch ( JobQueueError
$e ) {
355 $this->logException( $e );
358 $this->throwErrorIfAllPartitionsDown( $failed );
362 protected function doWaitForBackups() {
364 /** @var JobQueue $queue */
365 foreach ( $this->partitionQueues
as $queue ) {
367 $queue->waitForBackups();
368 } catch ( JobQueueError
$e ) {
370 $this->logException( $e );
373 $this->throwErrorIfAllPartitionsDown( $failed );
376 protected function doFlushCaches() {
377 /** @var JobQueue $queue */
378 foreach ( $this->partitionQueues
as $queue ) {
379 $queue->doFlushCaches();
383 public function getAllQueuedJobs() {
384 $iterator = new AppendIterator();
386 /** @var JobQueue $queue */
387 foreach ( $this->partitionQueues
as $queue ) {
388 $iterator->append( $queue->getAllQueuedJobs() );
394 public function getAllDelayedJobs() {
395 $iterator = new AppendIterator();
397 /** @var JobQueue $queue */
398 foreach ( $this->partitionQueues
as $queue ) {
399 $iterator->append( $queue->getAllDelayedJobs() );
405 public function getAllAcquiredJobs() {
406 $iterator = new AppendIterator();
408 /** @var JobQueue $queue */
409 foreach ( $this->partitionQueues
as $queue ) {
410 $iterator->append( $queue->getAllAcquiredJobs() );
416 public function getAllAbandonedJobs() {
417 $iterator = new AppendIterator();
419 /** @var JobQueue $queue */
420 foreach ( $this->partitionQueues
as $queue ) {
421 $iterator->append( $queue->getAllAbandonedJobs() );
427 public function getCoalesceLocationInternal() {
428 return "JobQueueFederated:wiki:{$this->wiki}" .
429 sha1( serialize( array_keys( $this->partitionQueues
) ) );
432 protected function doGetSiblingQueuesWithJobs( array $types ) {
436 /** @var JobQueue $queue */
437 foreach ( $this->partitionQueues
as $queue ) {
439 $nonEmpty = $queue->doGetSiblingQueuesWithJobs( $types );
440 if ( is_array( $nonEmpty ) ) {
441 $result = array_unique( array_merge( $result, $nonEmpty ) );
443 return null; // not supported on all partitions; bail
445 if ( count( $result ) == count( $types ) ) {
446 break; // short-circuit
448 } catch ( JobQueueError
$e ) {
450 $this->logException( $e );
453 $this->throwErrorIfAllPartitionsDown( $failed );
455 return array_values( $result );
458 protected function doGetSiblingQueueSizes( array $types ) {
461 /** @var JobQueue $queue */
462 foreach ( $this->partitionQueues
as $queue ) {
464 $sizes = $queue->doGetSiblingQueueSizes( $types );
465 if ( is_array( $sizes ) ) {
466 foreach ( $sizes as $type => $size ) {
467 $result[$type] = isset( $result[$type] ) ?
$result[$type] +
$size : $size;
470 return null; // not supported on all partitions; bail
472 } catch ( JobQueueError
$e ) {
474 $this->logException( $e );
477 $this->throwErrorIfAllPartitionsDown( $failed );
482 protected function logException( Exception
$e ) {
483 wfDebugLog( 'JobQueueFederated', $e->getMessage() . "\n" . $e->getTraceAsString() );
487 * Throw an error if no partitions available
489 * @param int $down The number of up partitions down
491 * @throws JobQueueError
493 protected function throwErrorIfAllPartitionsDown( $down ) {
494 if ( $down >= count( $this->partitionQueues
) ) {
495 throw new JobQueueError( 'No queue partitions available.' );