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
;
25 use Wikimedia\Rdbms\Database
;
26 use Wikimedia\Rdbms\IDatabase
;
27 use Wikimedia\Rdbms\DBError
;
28 use Wikimedia\Rdbms\DBQueryError
;
29 use Wikimedia\Rdbms\DBConnectionError
;
30 use Wikimedia\Rdbms\LoadBalancer
;
31 use Wikimedia\Rdbms\TransactionProfiler
;
32 use Wikimedia\WaitConditionLoop
;
35 * Class to store objects in the database
39 class SqlBagOStuff
extends BagOStuff
{
40 /** @var array[] (server index => server config) */
41 protected $serverInfos;
42 /** @var string[] (server index => tag/host name) */
43 protected $serverTags;
45 protected $numServers;
47 protected $lastExpireAll = 0;
49 protected $purgePeriod = 100;
51 protected $shards = 1;
53 protected $tableName = 'objectcache';
55 protected $replicaOnly = false;
57 protected $syncTimeout = 3;
59 /** @var LoadBalancer|null */
60 protected $separateMainLB;
63 /** @var array UNIX timestamps */
64 protected $connFailureTimes = [];
65 /** @var array Exceptions */
66 protected $connFailureErrors = [];
69 * Constructor. Parameters are:
70 * - server: A server info structure in the format required by each
71 * element in $wgDBServers.
73 * - servers: An array of server info structures describing a set of database servers
74 * to distribute keys to. If this is specified, the "server" option will be
75 * ignored. If string keys are used, then they will be used for consistent
76 * hashing *instead* of the host name (from the server config). This is useful
77 * when a cluster is replicated to another site (with different host names)
78 * but each server has a corresponding replica in the other cluster.
80 * - purgePeriod: The average number of object cache requests in between
81 * garbage collection operations, where expired entries
82 * are removed from the database. Or in other words, the
83 * reciprocal of the probability of purging on any given
84 * request. If this is set to zero, purging will never be
87 * - tableName: The table name to use, default is "objectcache".
89 * - shards: The number of tables to use for data storage on each server.
90 * If this is more than 1, table names will be formed in the style
91 * objectcacheNNN where NNN is the shard index, between 0 and
92 * shards-1. The number of digits will be the minimum number
93 * required to hold the largest shard index. Data will be
94 * distributed across all tables by key hash. This is for
95 * MySQL bugs 61735 and 61736.
96 * - slaveOnly: Whether to only use replica DBs and avoid triggering
97 * garbage collection logic of expired items. This only
98 * makes sense if the primary DB is used and only if get()
99 * calls will be used. This is used by ReplicatedBagOStuff.
100 * - syncTimeout: Max seconds to wait for replica DBs to catch up for WRITE_SYNC.
102 * @param array $params
104 public function __construct( $params ) {
105 parent
::__construct( $params );
107 $this->attrMap
[self
::ATTR_EMULATION
] = self
::QOS_EMULATION_SQL
;
108 $this->attrMap
[self
::ATTR_SYNCWRITES
] = self
::QOS_SYNCWRITES_NONE
;
110 if ( isset( $params['servers'] ) ) {
111 $this->serverInfos
= [];
112 $this->serverTags
= [];
113 $this->numServers
= count( $params['servers'] );
115 foreach ( $params['servers'] as $tag => $info ) {
116 $this->serverInfos
[$index] = $info;
117 if ( is_string( $tag ) ) {
118 $this->serverTags
[$index] = $tag;
120 $this->serverTags
[$index] = isset( $info['host'] ) ?
$info['host'] : "#$index";
124 } elseif ( isset( $params['server'] ) ) {
125 $this->serverInfos
= [ $params['server'] ];
126 $this->numServers
= count( $this->serverInfos
);
128 // Default to using the main wiki's database servers
129 $this->serverInfos
= false;
130 $this->numServers
= 1;
131 $this->attrMap
[self
::ATTR_SYNCWRITES
] = self
::QOS_SYNCWRITES_BE
;
133 if ( isset( $params['purgePeriod'] ) ) {
134 $this->purgePeriod
= intval( $params['purgePeriod'] );
136 if ( isset( $params['tableName'] ) ) {
137 $this->tableName
= $params['tableName'];
139 if ( isset( $params['shards'] ) ) {
140 $this->shards
= intval( $params['shards'] );
142 if ( isset( $params['syncTimeout'] ) ) {
143 $this->syncTimeout
= $params['syncTimeout'];
145 $this->replicaOnly
= !empty( $params['slaveOnly'] );
149 * Get a connection to the specified database
151 * @param int $serverIndex
153 * @throws MWException
155 protected function getDB( $serverIndex ) {
156 if ( !isset( $this->conns
[$serverIndex] ) ) {
157 if ( $serverIndex >= $this->numServers
) {
158 throw new MWException( __METHOD__
. ": Invalid server index \"$serverIndex\"" );
161 # Don't keep timing out trying to connect for each call if the DB is down
162 if ( isset( $this->connFailureErrors
[$serverIndex] )
163 && ( time() - $this->connFailureTimes
[$serverIndex] ) < 60
165 throw $this->connFailureErrors
[$serverIndex];
168 if ( $this->serverInfos
) {
169 // Use custom database defined by server connection info
170 $info = $this->serverInfos
[$serverIndex];
171 $type = isset( $info['type'] ) ?
$info['type'] : 'mysql';
172 $host = isset( $info['host'] ) ?
$info['host'] : '[unknown]';
173 $this->logger
->debug( __CLASS__
. ": connecting to $host" );
174 // Use a blank trx profiler to ignore expections as this is a cache
175 $info['trxProfiler'] = new TransactionProfiler();
176 $db = Database
::factory( $type, $info );
177 $db->clearFlag( DBO_TRX
); // auto-commit mode
179 // Use the main LB database
180 $lb = MediaWikiServices
::getInstance()->getDBLoadBalancer();
181 $index = $this->replicaOnly ? DB_REPLICA
: DB_MASTER
;
182 if ( $lb->getServerType( $lb->getWriterIndex() ) !== 'sqlite' ) {
183 // Keep a separate connection to avoid contention and deadlocks
184 $db = $lb->getConnection( $index, [], false, $lb::CONN_TRX_AUTOCOMMIT
);
185 // @TODO: Use a blank trx profiler to ignore expections as this is a cache
187 // However, SQLite has the opposite behavior due to DB-level locking.
188 // Stock sqlite MediaWiki installs use a separate sqlite cache DB instead.
189 $db = $lb->getConnection( $index );
193 $this->logger
->debug( sprintf( "Connection %s will be used for SqlBagOStuff", $db ) );
194 $this->conns
[$serverIndex] = $db;
197 return $this->conns
[$serverIndex];
201 * Get the server index and table name for a given key
203 * @return array Server index and table name
205 protected function getTableByKey( $key ) {
206 if ( $this->shards
> 1 ) {
207 $hash = hexdec( substr( md5( $key ), 0, 8 ) ) & 0x7fffffff;
208 $tableIndex = $hash %
$this->shards
;
212 if ( $this->numServers
> 1 ) {
213 $sortedServers = $this->serverTags
;
214 ArrayUtils
::consistentHashSort( $sortedServers, $key );
215 reset( $sortedServers );
216 $serverIndex = key( $sortedServers );
220 return [ $serverIndex, $this->getTableNameByShard( $tableIndex ) ];
224 * Get the table name for a given shard index
228 protected function getTableNameByShard( $index ) {
229 if ( $this->shards
> 1 ) {
230 $decimals = strlen( $this->shards
- 1 );
231 return $this->tableName
.
232 sprintf( "%0{$decimals}d", $index );
234 return $this->tableName
;
238 protected function doGet( $key, $flags = 0 ) {
241 return $this->getWithToken( $key, $casToken, $flags );
244 protected function getWithToken( $key, &$casToken, $flags = 0 ) {
245 $values = $this->getMulti( [ $key ] );
246 if ( array_key_exists( $key, $values ) ) {
247 $casToken = $values[$key];
248 return $values[$key];
253 public function getMulti( array $keys, $flags = 0 ) {
254 $values = []; // array of (key => value)
257 foreach ( $keys as $key ) {
258 list( $serverIndex, $tableName ) = $this->getTableByKey( $key );
259 $keysByTable[$serverIndex][$tableName][] = $key;
262 $this->garbageCollect(); // expire old entries if any
265 foreach ( $keysByTable as $serverIndex => $serverKeys ) {
267 $db = $this->getDB( $serverIndex );
268 foreach ( $serverKeys as $tableName => $tableKeys ) {
269 $res = $db->select( $tableName,
270 [ 'keyname', 'value', 'exptime' ],
271 [ 'keyname' => $tableKeys ],
273 // Approximate write-on-the-fly BagOStuff API via blocking.
274 // This approximation fails if a ROLLBACK happens (which is rare).
275 // We do not want to flush the TRX as that can break callers.
276 $db->trxLevel() ?
[ 'LOCK IN SHARE MODE' ] : []
278 if ( $res === false ) {
281 foreach ( $res as $row ) {
282 $row->serverIndex
= $serverIndex;
283 $row->tableName
= $tableName;
284 $dataRows[$row->keyname
] = $row;
287 } catch ( DBError
$e ) {
288 $this->handleReadError( $e, $serverIndex );
292 foreach ( $keys as $key ) {
293 if ( isset( $dataRows[$key] ) ) { // HIT?
294 $row = $dataRows[$key];
295 $this->debug( "get: retrieved data; expiry time is " . $row->exptime
);
298 $db = $this->getDB( $row->serverIndex
);
299 if ( $this->isExpired( $db, $row->exptime
) ) { // MISS
300 $this->debug( "get: key has expired" );
302 $values[$key] = $this->unserialize( $db->decodeBlob( $row->value
) );
304 } catch ( DBQueryError
$e ) {
305 $this->handleWriteError( $e, $db, $row->serverIndex
);
308 $this->debug( 'get: no matching rows' );
315 public function setMulti( array $data, $expiry = 0 ) {
317 foreach ( $data as $key => $value ) {
318 list( $serverIndex, $tableName ) = $this->getTableByKey( $key );
319 $keysByTable[$serverIndex][$tableName][] = $key;
322 $this->garbageCollect(); // expire old entries if any
325 $exptime = (int)$expiry;
326 foreach ( $keysByTable as $serverIndex => $serverKeys ) {
329 $db = $this->getDB( $serverIndex );
330 } catch ( DBError
$e ) {
331 $this->handleWriteError( $e, $db, $serverIndex );
336 if ( $exptime < 0 ) {
340 if ( $exptime == 0 ) {
341 $encExpiry = $this->getMaxDateTime( $db );
343 $exptime = $this->convertExpiry( $exptime );
344 $encExpiry = $db->timestamp( $exptime );
346 foreach ( $serverKeys as $tableName => $tableKeys ) {
348 foreach ( $tableKeys as $key ) {
351 'value' => $db->encodeBlob( $this->serialize( $data[$key] ) ),
352 'exptime' => $encExpiry,
363 } catch ( DBError
$e ) {
364 $this->handleWriteError( $e, $db, $serverIndex );
375 public function set( $key, $value, $exptime = 0, $flags = 0 ) {
376 $ok = $this->setMulti( [ $key => $value ], $exptime );
377 if ( ( $flags & self
::WRITE_SYNC
) == self
::WRITE_SYNC
) {
378 $ok = $this->waitForReplication() && $ok;
384 protected function cas( $casToken, $key, $value, $exptime = 0 ) {
385 list( $serverIndex, $tableName ) = $this->getTableByKey( $key );
388 $db = $this->getDB( $serverIndex );
389 $exptime = intval( $exptime );
391 if ( $exptime < 0 ) {
395 if ( $exptime == 0 ) {
396 $encExpiry = $this->getMaxDateTime( $db );
398 $exptime = $this->convertExpiry( $exptime );
399 $encExpiry = $db->timestamp( $exptime );
401 // (T26425) use a replace if the db supports it instead of
402 // delete/insert to avoid clashes with conflicting keynames
407 'value' => $db->encodeBlob( $this->serialize( $value ) ),
408 'exptime' => $encExpiry
412 'value' => $db->encodeBlob( $this->serialize( $casToken ) )
416 } catch ( DBQueryError
$e ) {
417 $this->handleWriteError( $e, $db, $serverIndex );
422 return (bool)$db->affectedRows();
425 public function delete( $key ) {
426 list( $serverIndex, $tableName ) = $this->getTableByKey( $key );
429 $db = $this->getDB( $serverIndex );
432 [ 'keyname' => $key ],
434 } catch ( DBError
$e ) {
435 $this->handleWriteError( $e, $db, $serverIndex );
442 public function incr( $key, $step = 1 ) {
443 list( $serverIndex, $tableName ) = $this->getTableByKey( $key );
446 $db = $this->getDB( $serverIndex );
447 $step = intval( $step );
448 $row = $db->selectRow(
450 [ 'value', 'exptime' ],
451 [ 'keyname' => $key ],
454 if ( $row === false ) {
459 $db->delete( $tableName, [ 'keyname' => $key ], __METHOD__
);
460 if ( $this->isExpired( $db, $row->exptime
) ) {
461 // Expired, do not reinsert
466 $oldValue = intval( $this->unserialize( $db->decodeBlob( $row->value
) ) );
467 $newValue = $oldValue +
$step;
468 $db->insert( $tableName,
471 'value' => $db->encodeBlob( $this->serialize( $newValue ) ),
472 'exptime' => $row->exptime
473 ], __METHOD__
, 'IGNORE' );
475 if ( $db->affectedRows() == 0 ) {
476 // Race condition. See T30611
479 } catch ( DBError
$e ) {
480 $this->handleWriteError( $e, $db, $serverIndex );
487 public function merge( $key, callable
$callback, $exptime = 0, $attempts = 10, $flags = 0 ) {
488 $ok = $this->mergeViaCas( $key, $callback, $exptime, $attempts );
489 if ( ( $flags & self
::WRITE_SYNC
) == self
::WRITE_SYNC
) {
490 $ok = $this->waitForReplication() && $ok;
496 public function changeTTL( $key, $expiry = 0 ) {
497 list( $serverIndex, $tableName ) = $this->getTableByKey( $key );
500 $db = $this->getDB( $serverIndex );
503 [ 'exptime' => $db->timestamp( $this->convertExpiry( $expiry ) ) ],
504 [ 'keyname' => $key, 'exptime > ' . $db->addQuotes( $db->timestamp( time() ) ) ],
507 if ( $db->affectedRows() == 0 ) {
510 } catch ( DBError
$e ) {
511 $this->handleWriteError( $e, $db, $serverIndex );
519 * @param IDatabase $db
520 * @param string $exptime
523 protected function isExpired( $db, $exptime ) {
524 return $exptime != $this->getMaxDateTime( $db ) && wfTimestamp( TS_UNIX
, $exptime ) < time();
528 * @param IDatabase $db
531 protected function getMaxDateTime( $db ) {
532 if ( time() > 0x7fffffff ) {
533 return $db->timestamp( 1 << 62 );
535 return $db->timestamp( 0x7fffffff );
539 protected function garbageCollect() {
540 if ( !$this->purgePeriod ||
$this->replicaOnly
) {
544 // Only purge on one in every $this->purgePeriod requests.
545 if ( $this->purgePeriod
!== 1 && mt_rand( 0, $this->purgePeriod
- 1 ) ) {
549 // Avoid repeating the delete within a few seconds
550 if ( $now > ( $this->lastExpireAll +
1 ) ) {
551 $this->lastExpireAll
= $now;
556 public function expireAll() {
557 $this->deleteObjectsExpiringBefore( wfTimestampNow() );
561 * Delete objects from the database which expire before a certain date.
562 * @param string $timestamp
563 * @param bool|callable $progressCallback
566 public function deleteObjectsExpiringBefore( $timestamp, $progressCallback = false ) {
567 for ( $serverIndex = 0; $serverIndex < $this->numServers
; $serverIndex++
) {
570 $db = $this->getDB( $serverIndex );
571 $dbTimestamp = $db->timestamp( $timestamp );
572 $totalSeconds = false;
573 $baseConds = [ 'exptime < ' . $db->addQuotes( $dbTimestamp ) ];
574 for ( $i = 0; $i < $this->shards
; $i++
) {
578 if ( $maxExpTime !== false ) {
579 $conds[] = 'exptime >= ' . $db->addQuotes( $maxExpTime );
582 $this->getTableNameByShard( $i ),
583 [ 'keyname', 'exptime' ],
586 [ 'LIMIT' => 100, 'ORDER BY' => 'exptime' ] );
587 if ( $rows === false ||
!$rows->numRows() ) {
591 $row = $rows->current();
592 $minExpTime = $row->exptime
;
593 if ( $totalSeconds === false ) {
594 $totalSeconds = wfTimestamp( TS_UNIX
, $timestamp )
595 - wfTimestamp( TS_UNIX
, $minExpTime );
597 foreach ( $rows as $row ) {
598 $keys[] = $row->keyname
;
599 $maxExpTime = $row->exptime
;
603 $this->getTableNameByShard( $i ),
605 'exptime >= ' . $db->addQuotes( $minExpTime ),
606 'exptime < ' . $db->addQuotes( $dbTimestamp ),
611 if ( $progressCallback ) {
612 if ( intval( $totalSeconds ) === 0 ) {
615 $remainingSeconds = wfTimestamp( TS_UNIX
, $timestamp )
616 - wfTimestamp( TS_UNIX
, $maxExpTime );
617 if ( $remainingSeconds > $totalSeconds ) {
618 $totalSeconds = $remainingSeconds;
620 $processedSeconds = $totalSeconds - $remainingSeconds;
621 $percent = ( $i +
$processedSeconds / $totalSeconds )
622 / $this->shards
* 100;
624 $percent = ( $percent / $this->numServers
)
625 +
( $serverIndex / $this->numServers
* 100 );
626 call_user_func( $progressCallback, $percent );
630 } catch ( DBError
$e ) {
631 $this->handleWriteError( $e, $db, $serverIndex );
639 * Delete content of shard tables in every server.
640 * Return true if the operation is successful, false otherwise.
643 public function deleteAll() {
644 for ( $serverIndex = 0; $serverIndex < $this->numServers
; $serverIndex++
) {
647 $db = $this->getDB( $serverIndex );
648 for ( $i = 0; $i < $this->shards
; $i++
) {
649 $db->delete( $this->getTableNameByShard( $i ), '*', __METHOD__
);
651 } catch ( DBError
$e ) {
652 $this->handleWriteError( $e, $db, $serverIndex );
660 * Serialize an object and, if possible, compress the representation.
661 * On typical message and page data, this can provide a 3X decrease
662 * in storage requirements.
664 * @param mixed &$data
667 protected function serialize( &$data ) {
668 $serial = serialize( $data );
670 if ( function_exists( 'gzdeflate' ) ) {
671 return gzdeflate( $serial );
678 * Unserialize and, if necessary, decompress an object.
679 * @param string $serial
682 protected function unserialize( $serial ) {
683 if ( function_exists( 'gzinflate' ) ) {
684 Wikimedia\
suppressWarnings();
685 $decomp = gzinflate( $serial );
686 Wikimedia\restoreWarnings
();
688 if ( false !== $decomp ) {
693 $ret = unserialize( $serial );
699 * Handle a DBError which occurred during a read operation.
701 * @param DBError $exception
702 * @param int $serverIndex
704 protected function handleReadError( DBError
$exception, $serverIndex ) {
705 if ( $exception instanceof DBConnectionError
) {
706 $this->markServerDown( $exception, $serverIndex );
708 $this->logger
->error( "DBError: {$exception->getMessage()}" );
709 if ( $exception instanceof DBConnectionError
) {
710 $this->setLastError( BagOStuff
::ERR_UNREACHABLE
);
711 $this->logger
->debug( __METHOD__
. ": ignoring connection error" );
713 $this->setLastError( BagOStuff
::ERR_UNEXPECTED
);
714 $this->logger
->debug( __METHOD__
. ": ignoring query error" );
719 * Handle a DBQueryError which occurred during a write operation.
721 * @param DBError $exception
722 * @param IDatabase|null $db DB handle or null if connection failed
723 * @param int $serverIndex
726 protected function handleWriteError( DBError
$exception, IDatabase
$db = null, $serverIndex ) {
728 $this->markServerDown( $exception, $serverIndex );
729 } elseif ( $db->wasReadOnlyError() ) {
730 if ( $db->trxLevel() && $this->usesMainDB() ) {
731 // Errors like deadlocks and connection drops already cause rollback.
732 // For consistency, we have no choice but to throw an error and trigger
733 // complete rollback if the main DB is also being used as the cache DB.
738 $this->logger
->error( "DBError: {$exception->getMessage()}" );
739 if ( $exception instanceof DBConnectionError
) {
740 $this->setLastError( BagOStuff
::ERR_UNREACHABLE
);
741 $this->logger
->debug( __METHOD__
. ": ignoring connection error" );
743 $this->setLastError( BagOStuff
::ERR_UNEXPECTED
);
744 $this->logger
->debug( __METHOD__
. ": ignoring query error" );
749 * Mark a server down due to a DBConnectionError exception
751 * @param DBError $exception
752 * @param int $serverIndex
754 protected function markServerDown( DBError
$exception, $serverIndex ) {
755 unset( $this->conns
[$serverIndex] ); // bug T103435
757 if ( isset( $this->connFailureTimes
[$serverIndex] ) ) {
758 if ( time() - $this->connFailureTimes
[$serverIndex] >= 60 ) {
759 unset( $this->connFailureTimes
[$serverIndex] );
760 unset( $this->connFailureErrors
[$serverIndex] );
762 $this->logger
->debug( __METHOD__
. ": Server #$serverIndex already down" );
767 $this->logger
->info( __METHOD__
. ": Server #$serverIndex down until " . ( $now +
60 ) );
768 $this->connFailureTimes
[$serverIndex] = $now;
769 $this->connFailureErrors
[$serverIndex] = $exception;
773 * Create shard tables. For use from eval.php.
775 public function createTables() {
776 for ( $serverIndex = 0; $serverIndex < $this->numServers
; $serverIndex++
) {
777 $db = $this->getDB( $serverIndex );
778 if ( $db->getType() !== 'mysql' ) {
779 throw new MWException( __METHOD__
. ' is not supported on this DB server' );
782 for ( $i = 0; $i < $this->shards
; $i++
) {
784 'CREATE TABLE ' . $db->tableName( $this->getTableNameByShard( $i ) ) .
785 ' LIKE ' . $db->tableName( 'objectcache' ),
792 * @return bool Whether the main DB is used, e.g. wfGetDB( DB_MASTER )
794 protected function usesMainDB() {
795 return !$this->serverInfos
;
798 protected function waitForReplication() {
799 if ( !$this->usesMainDB() ) {
800 // Custom DB server list; probably doesn't use replication
804 $lb = MediaWikiServices
::getInstance()->getDBLoadBalancer();
805 if ( $lb->getServerCount() <= 1 ) {
806 return true; // no replica DBs
809 // Main LB is used; wait for any replica DBs to catch up
810 $masterPos = $lb->getMasterPos();
812 return true; // not applicable
815 $loop = new WaitConditionLoop(
816 function () use ( $lb, $masterPos ) {
817 return $lb->waitForAll( $masterPos, 1 );
823 return ( $loop->invoke() === $loop::CONDITION_REACHED
);