NameTableStore.php

Go to the documentation of this file.

<?php
namespace MediaWiki\Storage;
 
use Psr\Log\LoggerInterface;
use Wikimedia\Assert\Assert;
use Wikimedia\ObjectCache\BagOStuff;
use Wikimedia\ObjectCache\WANObjectCache;
use Wikimedia\Rdbms\Database;
use Wikimedia\Rdbms\IDatabase;
use Wikimedia\Rdbms\ILoadBalancer;
use Wikimedia\Rdbms\IReadableDatabase;
 
class NameTableStore {
 
    private $tableCache = null;
 
    private readonly int $cacheTTL;
 
    private $normalizationCallback;
    private $insertCallback;
 
    public function __construct(
        private readonly ILoadBalancer $loadBalancer,
        private readonly WANObjectCache $cache,
        private readonly LoggerInterface $logger,
        private readonly string $table,
        private readonly string $idField,
        private readonly string $nameField,
        ?callable $normalizationCallback = null,
        private readonly bool|string $domain = false,
        ?callable $insertCallback = null,
    ) {
        $this->normalizationCallback = $normalizationCallback;
        $this->cacheTTL = BagOStuff::TTL_MONTH;
        $this->insertCallback = $insertCallback;
    }
 
    private function getDBConnection( $index, $flags = 0 ) {
        return $this->loadBalancer->getConnection( $index, [], $this->domain, $flags );
    }
 
    private function getCacheKey() {
        return $this->cache->makeGlobalKey(
            'NameTableSqlStore',
            $this->table,
            $this->loadBalancer->resolveDomainID( $this->domain )
        );
    }
 
    private function normalizeName( $name ) {
        if ( $this->normalizationCallback === null ) {
            return $name;
        }
        return ( $this->normalizationCallback )( $name );
    }
 
    public function acquireId( string $name ) {
        $name = $this->normalizeName( $name );
 
        $table = $this->getTableFromCachesOrReplica();
        $searchResult = array_search( $name, $table, true );
        if ( $searchResult === false ) {
            $id = $this->store( $name );
 
            if ( isset( $table[$id] ) ) {
                // This can happen when a name is assigned an ID within a transaction due to
                // CONN_TRX_AUTOCOMMIT being unable to use a separate connection (e.g. SQLite).
                // The right thing to do in this case is to discard the old value. According to
                // the contract of acquireId, the caller should not have used it outside the
                // transaction, so it should not be persisted anywhere after the rollback.
                $m = "Got ID $id for '$name' from insert"
                    . " into '{$this->table}', but ID $id was previously associated with"
                    . " the name '{$table[$id]}'. Overriding the old value, which presumably"
                    . " has been removed from the database due to a transaction rollback.";
                $this->logger->warning( $m );
            }
 
            $table[$id] = $name;
            $searchResult = $id;
 
            $this->tableCache = $table;
        }
 
        return $searchResult;
    }
 
    public function reloadMap( $connFlags = 0 ) {
        $dbw = $this->getDBConnection( DB_PRIMARY, $connFlags );
        $this->tableCache = $this->loadTable( $dbw );
        $dbw->onTransactionPreCommitOrIdle( function () {
            $this->cache->delete( $this->getCacheKey() );
        }, __METHOD__ );
 
        return $this->tableCache;
    }
 
    public function getId( string $name ) {
        $name = $this->normalizeName( $name );
 
        $table = $this->getTableFromCachesOrReplica();
        $searchResult = array_search( $name, $table, true );
 
        if ( $searchResult !== false ) {
            return $searchResult;
        }
 
        throw NameTableAccessException::newFromDetails( $this->table, 'name', $name );
    }
 
    public function getName( int $id ) {
        $table = $this->getTableFromCachesOrReplica();
        if ( array_key_exists( $id, $table ) ) {
            return $table[$id];
        }
        $fname = __METHOD__;
 
        $table = $this->cache->getWithSetCallback(
            $this->getCacheKey(),
            $this->cacheTTL,
            function ( $oldValue, &$ttl, &$setOpts ) use ( $id, $fname ) {
                // Check if cached value is up-to-date enough to have $id
                if ( is_array( $oldValue ) && array_key_exists( $id, $oldValue ) ) {
                    // Completely leave the cache key alone
                    $ttl = WANObjectCache::TTL_UNCACHEABLE;
                    // Use the old value
                    return $oldValue;
                }
                // Regenerate from replica DB, and primary DB if needed
                foreach ( [ DB_REPLICA, DB_PRIMARY ] as $source ) {
                    // Log a fallback to primary
                    if ( $source === DB_PRIMARY ) {
                        $this->logger->info(
                            $fname . ' falling back to primary select from ' .
                            $this->table . ' with id ' . $id
                        );
                    }
                    $db = $this->getDBConnection( $source );
                    $cacheSetOpts = Database::getCacheSetOptions( $db );
                    $table = $this->loadTable( $db );
                    if ( array_key_exists( $id, $table ) ) {
                        break; // found it
                    }
                }
                // Use the value from last source checked
                $setOpts += $cacheSetOpts;
 
                return $table;
            },
            [ 'minAsOf' => INF ] // force callback run
        );
 
        $this->tableCache = $table;
 
        if ( array_key_exists( $id, $table ) ) {
            return $table[$id];
        }
 
        throw NameTableAccessException::newFromDetails( $this->table, 'id', $id );
    }
 
    public function getMap() {
        return $this->getTableFromCachesOrReplica();
    }
 
    private function getTableFromCachesOrReplica() {
        if ( $this->tableCache !== null ) {
            return $this->tableCache;
        }
 
        $table = $this->cache->getWithSetCallback(
            $this->getCacheKey(),
            $this->cacheTTL,
            function ( $oldValue, &$ttl, &$setOpts ) {
                $dbr = $this->getDBConnection( DB_REPLICA );
                $setOpts += Database::getCacheSetOptions( $dbr );
                return $this->loadTable( $dbr );
            }
        );
 
        $this->tableCache = $table;
 
        return $table;
    }
 
    private function loadTable( IReadableDatabase $db ) {
        $result = $db->newSelectQueryBuilder()
            ->select( [
                'id' => $this->idField,
                'name' => $this->nameField
            ] )
            ->from( $this->table )
            ->orderBy( 'id' )
            ->caller( __METHOD__ )->fetchResultSet();
 
        $assocArray = [];
        foreach ( $result as $row ) {
            $assocArray[(int)$row->id] = $row->name;
        }
 
        return $assocArray;
    }
 
    private function store( string $name ) {
        Assert::parameter( $name !== '', '$name', 'should not be an empty string' );
        // Note: this is only called internally so normalization of $name has already occurred.
 
        $dbw = $this->getDBConnection( DB_PRIMARY, ILoadBalancer::CONN_TRX_AUTOCOMMIT );
 
        $dbw->newInsertQueryBuilder()
            ->insertInto( $this->table )
            ->ignore()
            ->row( $this->getFieldsToStore( $name ) )
            ->caller( __METHOD__ )->execute();
 
        if ( $dbw->affectedRows() > 0 ) {
            $id = $dbw->insertId();
            // As store returned an ID we know we inserted so delete from WAN cache
            $dbw->onTransactionPreCommitOrIdle(
                function () {
                    $this->cache->delete( $this->getCacheKey() );
                },
                __METHOD__
            );
 
            return $id;
        }
 
        $this->logger->info(
            'Tried to insert name into table ' . $this->table . ', but value already existed.'
        );
 
        // Note that in MySQL, even if this method somehow runs in a transaction, a plain
        // (non-locking) SELECT will see the new row created by the other transaction, even
        // with REPEATABLE-READ. This is due to how "consistent reads" works: the latest
        // version of rows become visible to the snapshot after the transaction sees those
        // rows as either matching an update query or conflicting with an insert query.
        $id = $dbw->newSelectQueryBuilder()
            ->select( [ 'id' => $this->idField ] )
            ->from( $this->table )
            ->where( [ $this->nameField => $name ] )
            ->caller( __METHOD__ )->fetchField();
 
        if ( $id === false ) {
            // Insert failed due to IGNORE flag, but DB_PRIMARY didn't give us the data
            $m = "No insert possible but primary DB didn't give us a record for " .
                "'{$name}' in '{$this->table}'";
            $this->logger->error( $m );
            throw new NameTableAccessException( $m );
        }
 
        return (int)$id;
    }
 
    private function getFieldsToStore( $name, $id = null ) {
        $fields = [];
 
        $fields[$this->nameField] = $name;
 
        if ( $id !== null ) {
            $fields[$this->idField] = $id;
        }
 
        if ( $this->insertCallback !== null ) {
            $fields = ( $this->insertCallback )( $fields );
        }
        return $fields;
    }
 
}