3 * Object caching using a SQL database.
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
24 use \MediaWiki\MediaWikiServices
;
27 * Class to store objects in the database
31 class SqlBagOStuff
extends BagOStuff
{
32 /** @var array[] (server index => server config) */
33 protected $serverInfos;
34 /** @var string[] (server index => tag/host name) */
35 protected $serverTags;
37 protected $numServers;
39 protected $lastExpireAll = 0;
41 protected $purgePeriod = 100;
43 protected $shards = 1;
45 protected $tableName = 'objectcache';
47 protected $replicaOnly = false;
49 protected $syncTimeout = 3;
51 /** @var LoadBalancer|null */
52 protected $separateMainLB;
55 /** @var array UNIX timestamps */
56 protected $connFailureTimes = [];
57 /** @var array Exceptions */
58 protected $connFailureErrors = [];
61 * Constructor. Parameters are:
62 * - server: A server info structure in the format required by each
63 * element in $wgDBServers.
65 * - servers: An array of server info structures describing a set of database servers
66 * to distribute keys to. If this is specified, the "server" option will be
67 * ignored. If string keys are used, then they will be used for consistent
68 * hashing *instead* of the host name (from the server config). This is useful
69 * when a cluster is replicated to another site (with different host names)
70 * but each server has a corresponding replica in the other cluster.
72 * - purgePeriod: The average number of object cache requests in between
73 * garbage collection operations, where expired entries
74 * are removed from the database. Or in other words, the
75 * reciprocal of the probability of purging on any given
76 * request. If this is set to zero, purging will never be
79 * - tableName: The table name to use, default is "objectcache".
81 * - shards: The number of tables to use for data storage on each server.
82 * If this is more than 1, table names will be formed in the style
83 * objectcacheNNN where NNN is the shard index, between 0 and
84 * shards-1. The number of digits will be the minimum number
85 * required to hold the largest shard index. Data will be
86 * distributed across all tables by key hash. This is for
87 * MySQL bugs 61735 and 61736.
88 * - slaveOnly: Whether to only use replica DBs and avoid triggering
89 * garbage collection logic of expired items. This only
90 * makes sense if the primary DB is used and only if get()
91 * calls will be used. This is used by ReplicatedBagOStuff.
92 * - syncTimeout: Max seconds to wait for replica DBs to catch up for WRITE_SYNC.
94 * @param array $params
96 public function __construct( $params ) {
97 parent
::__construct( $params );
99 $this->attrMap
[self
::ATTR_EMULATION
] = self
::QOS_EMULATION_SQL
;
100 $this->attrMap
[self
::ATTR_SYNCWRITES
] = self
::QOS_SYNCWRITES_NONE
;
102 if ( isset( $params['servers'] ) ) {
103 $this->serverInfos
= [];
104 $this->serverTags
= [];
105 $this->numServers
= count( $params['servers'] );
107 foreach ( $params['servers'] as $tag => $info ) {
108 $this->serverInfos
[$index] = $info;
109 if ( is_string( $tag ) ) {
110 $this->serverTags
[$index] = $tag;
112 $this->serverTags
[$index] = isset( $info['host'] ) ?
$info['host'] : "#$index";
116 } elseif ( isset( $params['server'] ) ) {
117 $this->serverInfos
= [ $params['server'] ];
118 $this->numServers
= count( $this->serverInfos
);
120 // Default to using the main wiki's database servers
121 $this->serverInfos
= false;
122 $this->numServers
= 1;
123 $this->attrMap
[self
::ATTR_SYNCWRITES
] = self
::QOS_SYNCWRITES_BE
;
125 if ( isset( $params['purgePeriod'] ) ) {
126 $this->purgePeriod
= intval( $params['purgePeriod'] );
128 if ( isset( $params['tableName'] ) ) {
129 $this->tableName
= $params['tableName'];
131 if ( isset( $params['shards'] ) ) {
132 $this->shards
= intval( $params['shards'] );
134 if ( isset( $params['syncTimeout'] ) ) {
135 $this->syncTimeout
= $params['syncTimeout'];
137 $this->replicaOnly
= !empty( $params['slaveOnly'] );
140 protected function getSeparateMainLB() {
143 if ( $wgDBtype === 'mysql' && $this->usesMainDB() ) {
144 if ( !$this->separateMainLB
) {
145 // We must keep a separate connection to MySQL in order to avoid deadlocks
146 $lbFactory = MediaWikiServices
::getInstance()->getDBLoadBalancerFactory();
147 $this->separateMainLB
= $lbFactory->newMainLB();
149 return $this->separateMainLB
;
151 // However, SQLite has an opposite behavior. And PostgreSQL needs to know
152 // if we are in transaction or not (@TODO: find some PostgreSQL work-around).
158 * Get a connection to the specified database
160 * @param int $serverIndex
162 * @throws MWException
164 protected function getDB( $serverIndex ) {
165 if ( !isset( $this->conns
[$serverIndex] ) ) {
166 if ( $serverIndex >= $this->numServers
) {
167 throw new MWException( __METHOD__
. ": Invalid server index \"$serverIndex\"" );
170 # Don't keep timing out trying to connect for each call if the DB is down
171 if ( isset( $this->connFailureErrors
[$serverIndex] )
172 && ( time() - $this->connFailureTimes
[$serverIndex] ) < 60
174 throw $this->connFailureErrors
[$serverIndex];
177 # If server connection info was given, use that
178 if ( $this->serverInfos
) {
179 $info = $this->serverInfos
[$serverIndex];
180 $type = isset( $info['type'] ) ?
$info['type'] : 'mysql';
181 $host = isset( $info['host'] ) ?
$info['host'] : '[unknown]';
182 $this->logger
->debug( __CLASS__
. ": connecting to $host" );
183 // Use a blank trx profiler to ignore expections as this is a cache
184 $info['trxProfiler'] = new TransactionProfiler();
185 $db = Database
::factory( $type, $info );
186 $db->clearFlag( DBO_TRX
);
188 $index = $this->replicaOnly ? DB_REPLICA
: DB_MASTER
;
189 if ( $this->getSeparateMainLB() ) {
190 $db = $this->getSeparateMainLB()->getConnection( $index );
191 $db->clearFlag( DBO_TRX
); // auto-commit mode
193 $db = wfGetDB( $index );
194 // Can't mess with transaction rounds (DBO_TRX) :(
197 $this->logger
->debug( sprintf( "Connection %s will be used for SqlBagOStuff", $db ) );
198 $this->conns
[$serverIndex] = $db;
201 return $this->conns
[$serverIndex];
205 * Get the server index and table name for a given key
207 * @return array Server index and table name
209 protected function getTableByKey( $key ) {
210 if ( $this->shards
> 1 ) {
211 $hash = hexdec( substr( md5( $key ), 0, 8 ) ) & 0x7fffffff;
212 $tableIndex = $hash %
$this->shards
;
216 if ( $this->numServers
> 1 ) {
217 $sortedServers = $this->serverTags
;
218 ArrayUtils
::consistentHashSort( $sortedServers, $key );
219 reset( $sortedServers );
220 $serverIndex = key( $sortedServers );
224 return [ $serverIndex, $this->getTableNameByShard( $tableIndex ) ];
228 * Get the table name for a given shard index
232 protected function getTableNameByShard( $index ) {
233 if ( $this->shards
> 1 ) {
234 $decimals = strlen( $this->shards
- 1 );
235 return $this->tableName
.
236 sprintf( "%0{$decimals}d", $index );
238 return $this->tableName
;
242 protected function doGet( $key, $flags = 0 ) {
245 return $this->getWithToken( $key, $casToken, $flags );
248 protected function getWithToken( $key, &$casToken, $flags = 0 ) {
249 $values = $this->getMulti( [ $key ] );
250 if ( array_key_exists( $key, $values ) ) {
251 $casToken = $values[$key];
252 return $values[$key];
257 public function getMulti( array $keys, $flags = 0 ) {
258 $values = []; // array of (key => value)
261 foreach ( $keys as $key ) {
262 list( $serverIndex, $tableName ) = $this->getTableByKey( $key );
263 $keysByTable[$serverIndex][$tableName][] = $key;
266 $this->garbageCollect(); // expire old entries if any
269 foreach ( $keysByTable as $serverIndex => $serverKeys ) {
271 $db = $this->getDB( $serverIndex );
272 foreach ( $serverKeys as $tableName => $tableKeys ) {
273 $res = $db->select( $tableName,
274 [ 'keyname', 'value', 'exptime' ],
275 [ 'keyname' => $tableKeys ],
277 // Approximate write-on-the-fly BagOStuff API via blocking.
278 // This approximation fails if a ROLLBACK happens (which is rare).
279 // We do not want to flush the TRX as that can break callers.
280 $db->trxLevel() ?
[ 'LOCK IN SHARE MODE' ] : []
282 if ( $res === false ) {
285 foreach ( $res as $row ) {
286 $row->serverIndex
= $serverIndex;
287 $row->tableName
= $tableName;
288 $dataRows[$row->keyname
] = $row;
291 } catch ( DBError
$e ) {
292 $this->handleReadError( $e, $serverIndex );
296 foreach ( $keys as $key ) {
297 if ( isset( $dataRows[$key] ) ) { // HIT?
298 $row = $dataRows[$key];
299 $this->debug( "get: retrieved data; expiry time is " . $row->exptime
);
302 $db = $this->getDB( $row->serverIndex
);
303 if ( $this->isExpired( $db, $row->exptime
) ) { // MISS
304 $this->debug( "get: key has expired" );
306 $values[$key] = $this->unserialize( $db->decodeBlob( $row->value
) );
308 } catch ( DBQueryError
$e ) {
309 $this->handleWriteError( $e, $db, $row->serverIndex
);
312 $this->debug( 'get: no matching rows' );
319 public function setMulti( array $data, $expiry = 0 ) {
321 foreach ( $data as $key => $value ) {
322 list( $serverIndex, $tableName ) = $this->getTableByKey( $key );
323 $keysByTable[$serverIndex][$tableName][] = $key;
326 $this->garbageCollect(); // expire old entries if any
329 $exptime = (int)$expiry;
330 foreach ( $keysByTable as $serverIndex => $serverKeys ) {
333 $db = $this->getDB( $serverIndex );
334 } catch ( DBError
$e ) {
335 $this->handleWriteError( $e, $db, $serverIndex );
340 if ( $exptime < 0 ) {
344 if ( $exptime == 0 ) {
345 $encExpiry = $this->getMaxDateTime( $db );
347 $exptime = $this->convertExpiry( $exptime );
348 $encExpiry = $db->timestamp( $exptime );
350 foreach ( $serverKeys as $tableName => $tableKeys ) {
352 foreach ( $tableKeys as $key ) {
355 'value' => $db->encodeBlob( $this->serialize( $data[$key] ) ),
356 'exptime' => $encExpiry,
367 } catch ( DBError
$e ) {
368 $this->handleWriteError( $e, $db, $serverIndex );
379 public function set( $key, $value, $exptime = 0, $flags = 0 ) {
380 $ok = $this->setMulti( [ $key => $value ], $exptime );
381 if ( ( $flags & self
::WRITE_SYNC
) == self
::WRITE_SYNC
) {
382 $ok = $this->waitForReplication() && $ok;
388 protected function cas( $casToken, $key, $value, $exptime = 0 ) {
389 list( $serverIndex, $tableName ) = $this->getTableByKey( $key );
392 $db = $this->getDB( $serverIndex );
393 $exptime = intval( $exptime );
395 if ( $exptime < 0 ) {
399 if ( $exptime == 0 ) {
400 $encExpiry = $this->getMaxDateTime( $db );
402 $exptime = $this->convertExpiry( $exptime );
403 $encExpiry = $db->timestamp( $exptime );
405 // (bug 24425) use a replace if the db supports it instead of
406 // delete/insert to avoid clashes with conflicting keynames
411 'value' => $db->encodeBlob( $this->serialize( $value ) ),
412 'exptime' => $encExpiry
416 'value' => $db->encodeBlob( $this->serialize( $casToken ) )
420 } catch ( DBQueryError
$e ) {
421 $this->handleWriteError( $e, $db, $serverIndex );
426 return (bool)$db->affectedRows();
429 public function delete( $key ) {
430 list( $serverIndex, $tableName ) = $this->getTableByKey( $key );
433 $db = $this->getDB( $serverIndex );
436 [ 'keyname' => $key ],
438 } catch ( DBError
$e ) {
439 $this->handleWriteError( $e, $db, $serverIndex );
446 public function incr( $key, $step = 1 ) {
447 list( $serverIndex, $tableName ) = $this->getTableByKey( $key );
450 $db = $this->getDB( $serverIndex );
451 $step = intval( $step );
452 $row = $db->selectRow(
454 [ 'value', 'exptime' ],
455 [ 'keyname' => $key ],
458 if ( $row === false ) {
463 $db->delete( $tableName, [ 'keyname' => $key ], __METHOD__
);
464 if ( $this->isExpired( $db, $row->exptime
) ) {
465 // Expired, do not reinsert
470 $oldValue = intval( $this->unserialize( $db->decodeBlob( $row->value
) ) );
471 $newValue = $oldValue +
$step;
472 $db->insert( $tableName,
475 'value' => $db->encodeBlob( $this->serialize( $newValue ) ),
476 'exptime' => $row->exptime
477 ], __METHOD__
, 'IGNORE' );
479 if ( $db->affectedRows() == 0 ) {
480 // Race condition. See bug 28611
483 } catch ( DBError
$e ) {
484 $this->handleWriteError( $e, $db, $serverIndex );
491 public function merge( $key, callable
$callback, $exptime = 0, $attempts = 10, $flags = 0 ) {
492 $ok = $this->mergeViaCas( $key, $callback, $exptime, $attempts );
493 if ( ( $flags & self
::WRITE_SYNC
) == self
::WRITE_SYNC
) {
494 $ok = $this->waitForReplication() && $ok;
500 public function changeTTL( $key, $expiry = 0 ) {
501 list( $serverIndex, $tableName ) = $this->getTableByKey( $key );
504 $db = $this->getDB( $serverIndex );
507 [ 'exptime' => $db->timestamp( $this->convertExpiry( $expiry ) ) ],
508 [ 'keyname' => $key, 'exptime > ' . $db->addQuotes( $db->timestamp( time() ) ) ],
511 if ( $db->affectedRows() == 0 ) {
514 } catch ( DBError
$e ) {
515 $this->handleWriteError( $e, $db, $serverIndex );
523 * @param IDatabase $db
524 * @param string $exptime
527 protected function isExpired( $db, $exptime ) {
528 return $exptime != $this->getMaxDateTime( $db ) && wfTimestamp( TS_UNIX
, $exptime ) < time();
532 * @param IDatabase $db
535 protected function getMaxDateTime( $db ) {
536 if ( time() > 0x7fffffff ) {
537 return $db->timestamp( 1 << 62 );
539 return $db->timestamp( 0x7fffffff );
543 protected function garbageCollect() {
544 if ( !$this->purgePeriod ||
$this->replicaOnly
) {
548 // Only purge on one in every $this->purgePeriod requests.
549 if ( $this->purgePeriod
!== 1 && mt_rand( 0, $this->purgePeriod
- 1 ) ) {
553 // Avoid repeating the delete within a few seconds
554 if ( $now > ( $this->lastExpireAll +
1 ) ) {
555 $this->lastExpireAll
= $now;
560 public function expireAll() {
561 $this->deleteObjectsExpiringBefore( wfTimestampNow() );
565 * Delete objects from the database which expire before a certain date.
566 * @param string $timestamp
567 * @param bool|callable $progressCallback
570 public function deleteObjectsExpiringBefore( $timestamp, $progressCallback = false ) {
571 for ( $serverIndex = 0; $serverIndex < $this->numServers
; $serverIndex++
) {
574 $db = $this->getDB( $serverIndex );
575 $dbTimestamp = $db->timestamp( $timestamp );
576 $totalSeconds = false;
577 $baseConds = [ 'exptime < ' . $db->addQuotes( $dbTimestamp ) ];
578 for ( $i = 0; $i < $this->shards
; $i++
) {
582 if ( $maxExpTime !== false ) {
583 $conds[] = 'exptime > ' . $db->addQuotes( $maxExpTime );
586 $this->getTableNameByShard( $i ),
587 [ 'keyname', 'exptime' ],
590 [ 'LIMIT' => 100, 'ORDER BY' => 'exptime' ] );
591 if ( $rows === false ||
!$rows->numRows() ) {
595 $row = $rows->current();
596 $minExpTime = $row->exptime
;
597 if ( $totalSeconds === false ) {
598 $totalSeconds = wfTimestamp( TS_UNIX
, $timestamp )
599 - wfTimestamp( TS_UNIX
, $minExpTime );
601 foreach ( $rows as $row ) {
602 $keys[] = $row->keyname
;
603 $maxExpTime = $row->exptime
;
607 $this->getTableNameByShard( $i ),
609 'exptime >= ' . $db->addQuotes( $minExpTime ),
610 'exptime < ' . $db->addQuotes( $dbTimestamp ),
615 if ( $progressCallback ) {
616 if ( intval( $totalSeconds ) === 0 ) {
619 $remainingSeconds = wfTimestamp( TS_UNIX
, $timestamp )
620 - wfTimestamp( TS_UNIX
, $maxExpTime );
621 if ( $remainingSeconds > $totalSeconds ) {
622 $totalSeconds = $remainingSeconds;
624 $processedSeconds = $totalSeconds - $remainingSeconds;
625 $percent = ( $i +
$processedSeconds / $totalSeconds )
626 / $this->shards
* 100;
628 $percent = ( $percent / $this->numServers
)
629 +
( $serverIndex / $this->numServers
* 100 );
630 call_user_func( $progressCallback, $percent );
634 } catch ( DBError
$e ) {
635 $this->handleWriteError( $e, $db, $serverIndex );
643 * Delete content of shard tables in every server.
644 * Return true if the operation is successful, false otherwise.
647 public function deleteAll() {
648 for ( $serverIndex = 0; $serverIndex < $this->numServers
; $serverIndex++
) {
651 $db = $this->getDB( $serverIndex );
652 for ( $i = 0; $i < $this->shards
; $i++
) {
653 $db->delete( $this->getTableNameByShard( $i ), '*', __METHOD__
);
655 } catch ( DBError
$e ) {
656 $this->handleWriteError( $e, $db, $serverIndex );
664 * Serialize an object and, if possible, compress the representation.
665 * On typical message and page data, this can provide a 3X decrease
666 * in storage requirements.
671 protected function serialize( &$data ) {
672 $serial = serialize( $data );
674 if ( function_exists( 'gzdeflate' ) ) {
675 return gzdeflate( $serial );
682 * Unserialize and, if necessary, decompress an object.
683 * @param string $serial
686 protected function unserialize( $serial ) {
687 if ( function_exists( 'gzinflate' ) ) {
688 MediaWiki\
suppressWarnings();
689 $decomp = gzinflate( $serial );
690 MediaWiki\restoreWarnings
();
692 if ( false !== $decomp ) {
697 $ret = unserialize( $serial );
703 * Handle a DBError which occurred during a read operation.
705 * @param DBError $exception
706 * @param int $serverIndex
708 protected function handleReadError( DBError
$exception, $serverIndex ) {
709 if ( $exception instanceof DBConnectionError
) {
710 $this->markServerDown( $exception, $serverIndex );
712 $this->logger
->error( "DBError: {$exception->getMessage()}" );
713 if ( $exception instanceof DBConnectionError
) {
714 $this->setLastError( BagOStuff
::ERR_UNREACHABLE
);
715 $this->logger
->debug( __METHOD__
. ": ignoring connection error" );
717 $this->setLastError( BagOStuff
::ERR_UNEXPECTED
);
718 $this->logger
->debug( __METHOD__
. ": ignoring query error" );
723 * Handle a DBQueryError which occurred during a write operation.
725 * @param DBError $exception
726 * @param IDatabase|null $db DB handle or null if connection failed
727 * @param int $serverIndex
730 protected function handleWriteError( DBError
$exception, IDatabase
$db = null, $serverIndex ) {
732 $this->markServerDown( $exception, $serverIndex );
733 } elseif ( $db->wasReadOnlyError() ) {
734 if ( $db->trxLevel() && $this->usesMainDB() ) {
735 // Errors like deadlocks and connection drops already cause rollback.
736 // For consistency, we have no choice but to throw an error and trigger
737 // complete rollback if the main DB is also being used as the cache DB.
742 $this->logger
->error( "DBError: {$exception->getMessage()}" );
743 if ( $exception instanceof DBConnectionError
) {
744 $this->setLastError( BagOStuff
::ERR_UNREACHABLE
);
745 $this->logger
->debug( __METHOD__
. ": ignoring connection error" );
747 $this->setLastError( BagOStuff
::ERR_UNEXPECTED
);
748 $this->logger
->debug( __METHOD__
. ": ignoring query error" );
753 * Mark a server down due to a DBConnectionError exception
755 * @param DBError $exception
756 * @param int $serverIndex
758 protected function markServerDown( DBError
$exception, $serverIndex ) {
759 unset( $this->conns
[$serverIndex] ); // bug T103435
761 if ( isset( $this->connFailureTimes
[$serverIndex] ) ) {
762 if ( time() - $this->connFailureTimes
[$serverIndex] >= 60 ) {
763 unset( $this->connFailureTimes
[$serverIndex] );
764 unset( $this->connFailureErrors
[$serverIndex] );
766 $this->logger
->debug( __METHOD__
. ": Server #$serverIndex already down" );
771 $this->logger
->info( __METHOD__
. ": Server #$serverIndex down until " . ( $now +
60 ) );
772 $this->connFailureTimes
[$serverIndex] = $now;
773 $this->connFailureErrors
[$serverIndex] = $exception;
777 * Create shard tables. For use from eval.php.
779 public function createTables() {
780 for ( $serverIndex = 0; $serverIndex < $this->numServers
; $serverIndex++
) {
781 $db = $this->getDB( $serverIndex );
782 if ( $db->getType() !== 'mysql' ) {
783 throw new MWException( __METHOD__
. ' is not supported on this DB server' );
786 for ( $i = 0; $i < $this->shards
; $i++
) {
788 'CREATE TABLE ' . $db->tableName( $this->getTableNameByShard( $i ) ) .
789 ' LIKE ' . $db->tableName( 'objectcache' ),
796 * @return bool Whether the main DB is used, e.g. wfGetDB( DB_MASTER )
798 protected function usesMainDB() {
799 return !$this->serverInfos
;
802 protected function waitForReplication() {
803 if ( !$this->usesMainDB() ) {
804 // Custom DB server list; probably doesn't use replication
808 $lb = $this->getSeparateMainLB()
809 ?
: MediaWikiServices
::getInstance()->getDBLoadBalancer();
811 if ( $lb->getServerCount() <= 1 ) {
812 return true; // no replica DBs
815 // Main LB is used; wait for any replica DBs to catch up
816 $masterPos = $lb->getMasterPos();
818 $loop = new WaitConditionLoop(
819 function () use ( $lb, $masterPos ) {
820 return $lb->waitForAll( $masterPos, 1 );
826 return ( $loop->invoke() === $loop::CONDITION_REACHED
);