*
* @file
* @author Tim Starling
+ * @author Aaron Schulz
* @copyright © 2009, Tim Starling, Domas Mituzas
* @copyright © 2010, Max Sem
* @copyright © 2011, Antoine Musso
/**
* Multi dimensions array representing batches. Keys are:
* > (string) links table name
+ * > (int) batch size
* > 'numRows' : Number of rows for this link table
* > 'batches' : array( $start, $end )
*
self::$cache = new ProcessCacheLRU( 1 );
}
$dbKey = $title->getPrefixedDBkey();
- if ( !self::$cache->has( $dbKey, 'obj' ) ) {
+ if ( !self::$cache->has( $dbKey, 'obj', 3600 ) ) {
self::$cache->set( $dbKey, 'obj', new self( $title ) );
}
+
return self::$cache->get( $dbKey, 'obj' );
}
/**
* Get the backlinks for a given table. Cached in process memory only.
* @param $table String
- * @param $startId Integer or false
- * @param $endId Integer or false
+ * @param $startId Integer|false
+ * @param $endId Integer|false
+ * @param $max Integer|INF
* @return TitleArrayFromResult
*/
- public function getLinks( $table, $startId = false, $endId = false ) {
+ public function getLinks( $table, $startId = false, $endId = false, $max = INF ) {
+ return TitleArray::newFromResult( $this->queryLinks( $table, $startId, $endId, $max ) );
+ }
+
+ /**
+ * Get the backlinks for a given table. Cached in process memory only.
+ * @param $table String
+ * @param $startId Integer|false
+ * @param $endId Integer|false
+ * @param $max Integer|INF
+ * @param $select string 'all' or 'ids'
+ * @return ResultWrapper
+ */
+ protected function queryLinks( $table, $startId, $endId, $max, $select = 'all' ) {
wfProfileIn( __METHOD__ );
$fromField = $this->getPrefix( $table ) . '_from';
- if ( $startId || $endId ) {
- // Partial range, not cached
- wfDebug( __METHOD__ . ": from DB (uncacheable range)\n" );
+ if ( !$startId && !$endId && is_infinite( $max )
+ && isset( $this->fullResultCache[$table] )
+ ) {
+ wfDebug( __METHOD__ . ": got results from cache\n" );
+ $res = $this->fullResultCache[$table];
+ } else {
+ wfDebug( __METHOD__ . ": got results from DB\n" );
$conds = $this->getConditions( $table );
-
// Use the from field in the condition rather than the joined page_id,
// because databases are stupid and don't necessarily propagate indexes.
if ( $startId ) {
$conds[] = "$fromField >= " . intval( $startId );
}
-
if ( $endId ) {
$conds[] = "$fromField <= " . intval( $endId );
}
+ $options = array( 'ORDER BY' => $fromField );
+ if ( is_finite( $max ) && $max > 0 ) {
+ $options['LIMIT'] = $max;
+ }
- $res = $this->getDB()->select(
- array( $table, 'page' ),
- array( 'page_namespace', 'page_title', 'page_id' ),
- $conds,
- __METHOD__,
- array(
- 'STRAIGHT_JOIN',
- 'ORDER BY' => $fromField
- ) );
- $ta = TitleArray::newFromResult( $res );
-
- wfProfileOut( __METHOD__ );
- return $ta;
- }
+ if ( $select === 'ids' ) {
+ // Just select from the backlink table and ignore the page JOIN
+ $res = $this->getDB()->select(
+ $table,
+ array( $this->getPrefix( $table ) . '_from AS page_id' ),
+ array_filter( $conds, function ( $clause ) { // kind of janky
+ return !preg_match( '/(\b|=)page_id(\b|=)/', $clause );
+ } ),
+ __METHOD__,
+ $options
+ );
+ } else {
+ // Select from the backlink table and JOIN with page title information
+ $res = $this->getDB()->select(
+ array( $table, 'page' ),
+ array( 'page_namespace', 'page_title', 'page_id' ),
+ $conds,
+ __METHOD__,
+ array_merge( array( 'STRAIGHT_JOIN' ), $options )
+ );
+ }
- // @todo FIXME: Make this a function?
- if ( !isset( $this->fullResultCache[$table] ) ) {
- wfDebug( __METHOD__ . ": from DB\n" );
- $res = $this->getDB()->select(
- array( $table, 'page' ),
- array( 'page_namespace', 'page_title', 'page_id' ),
- $this->getConditions( $table ),
- __METHOD__,
- array(
- 'STRAIGHT_JOIN',
- 'ORDER BY' => $fromField,
- ) );
- $this->fullResultCache[$table] = $res;
+ if ( $select === 'all' && !$startId && !$endId && $res->numRows() < $max ) {
+ // The full results fit within the limit, so cache them
+ $this->fullResultCache[$table] = $res;
+ } else {
+ wfDebug( __METHOD__ . ": results from DB were uncacheable\n" );
+ }
}
- $ta = TitleArray::newFromResult( $this->fullResultCache[$table] );
-
wfProfileOut( __METHOD__ );
- return $ta;
+
+ return $res;
}
/**
} else {
$prefix = null;
wfRunHooks( 'BacklinkCacheGetPrefix', array( $table, &$prefix ) );
- if( $prefix ) {
+ if ( $prefix ) {
return $prefix;
} else {
throw new MWException( "Invalid table \"$table\" in " . __CLASS__ );
case 'imagelinks':
$conds = array(
'il_to' => $this->title->getDBkey(),
- 'page_id=il_from'
+ "page_id={$prefix}_from"
);
break;
case 'categorylinks':
$conds = array(
'cl_to' => $this->title->getDBkey(),
- 'page_id=cl_from',
+ "page_id={$prefix}_from"
);
break;
default:
$conds = null;
wfRunHooks( 'BacklinkCacheGetConditions', array( $table, $this->title, &$conds ) );
- if( !$conds ) {
+ if ( !$conds ) {
throw new MWException( "Invalid table \"$table\" in " . __CLASS__ );
}
}
/**
* Get the approximate number of backlinks
* @param $table String
- * @param $max integer Only count up to this many backlinks
+ * @param $max integer|INF Only count up to this many backlinks
* @return integer
*/
public function getNumLinks( $table, $max = INF ) {
- global $wgMemc;
+ global $wgMemc, $wgUpdateRowsPerJob;
// 1) try partition cache ...
if ( isset( $this->partitionCache[$table] ) ) {
$entry = reset( $this->partitionCache[$table] );
+
return min( $max, $entry['numRows'] );
}
}
// 4) fetch from the database ...
- if ( is_infinite( $max ) ) { // full count
- $count = $this->getLinks( $table )->count();
- $wgMemc->set( $memcKey, $count, self::CACHE_EXPIRY );
- } else { // with limit
- $count = $this->getDB()->select(
- array( $table, 'page' ),
- '1',
- $this->getConditions( $table ),
- __METHOD__,
- array( 'LIMIT' => $max )
- )->numRows();
+ if ( is_infinite( $max ) ) { // no limit at all
+ // Use partition() since it will batch the query and skip the JOIN.
+ // Use $wgUpdateRowsPerJob just to encourage cache reuse for jobs.
+ $this->partition( $table, $wgUpdateRowsPerJob ); // updates $this->partitionCache
+ return $this->partitionCache[$table][$wgUpdateRowsPerJob]['numRows'];
+ } else { // probably some sane limit
+ // Fetch the full title info, since the caller will likely need it next
+ $count = $this->getLinks( $table, false, false, $max )->count();
+ if ( $count < $max ) { // full count
+ $wgMemc->set( $memcKey, $count, self::CACHE_EXPIRY );
+ }
}
- return $count;
+ return min( $max, $count );
}
/**
// 1) try partition cache ...
if ( isset( $this->partitionCache[$table][$batchSize] ) ) {
wfDebug( __METHOD__ . ": got from partition cache\n" );
+
return $this->partitionCache[$table][$batchSize]['batches'];
}
if ( isset( $this->fullResultCache[$table] ) ) {
$cacheEntry = $this->partitionResult( $this->fullResultCache[$table], $batchSize );
wfDebug( __METHOD__ . ": got from full result cache\n" );
+
return $cacheEntry['batches'];
}
if ( is_array( $memcValue ) ) {
$cacheEntry = $memcValue;
wfDebug( __METHOD__ . ": got from memcached $memcKey\n" );
+
return $cacheEntry['batches'];
}
// 4) ... finally fetch from the slow database :(
- $this->getLinks( $table );
- $cacheEntry = $this->partitionResult( $this->fullResultCache[$table], $batchSize );
+ $cacheEntry = array( 'numRows' => 0, 'batches' => array() ); // final result
+ // Do the selects in batches to avoid client-side OOMs (bug 43452).
+ // Use a LIMIT that plays well with $batchSize to keep equal sized partitions.
+ $selectSize = max( $batchSize, 200000 - ( 200000 % $batchSize ) );
+ $start = false;
+ do {
+ $res = $this->queryLinks( $table, $start, false, $selectSize, 'ids' );
+ $partitions = $this->partitionResult( $res, $batchSize, false );
+ // Merge the link count and range partitions for this chunk
+ $cacheEntry['numRows'] += $partitions['numRows'];
+ $cacheEntry['batches'] = array_merge( $cacheEntry['batches'], $partitions['batches'] );
+ if ( count( $partitions['batches'] ) ) {
+ list( $lStart, $lEnd ) = end( $partitions['batches'] );
+ $start = $lEnd + 1; // pick up after this inclusive range
+ }
+ } while ( $partitions['numRows'] >= $selectSize );
+ // Make sure the first range has start=false and the last one has end=false
+ if ( count( $cacheEntry['batches'] ) ) {
+ $cacheEntry['batches'][0][0] = false;
+ $cacheEntry['batches'][count( $cacheEntry['batches'] ) - 1][1] = false;
+ }
+
// Save partitions to memcached
$wgMemc->set( $memcKey, $cacheEntry, self::CACHE_EXPIRY );
$wgMemc->set( $memcKey, $cacheEntry['numRows'], self::CACHE_EXPIRY );
wfDebug( __METHOD__ . ": got from database\n" );
+
return $cacheEntry['batches'];
}
* Partition a DB result with backlinks in it into batches
* @param $res ResultWrapper database result
* @param $batchSize integer
+ * @param $isComplete bool Whether $res includes all the backlinks
* @throws MWException
- * @return array @see
+ * @return array
*/
- protected function partitionResult( $res, $batchSize ) {
+ protected function partitionResult( $res, $batchSize, $isComplete = true ) {
$batches = array();
$numRows = $res->numRows();
$numBatches = ceil( $numRows / $batchSize );
for ( $i = 0; $i < $numBatches; $i++ ) {
- if ( $i == 0 ) {
+ if ( $i == 0 && $isComplete ) {
$start = false;
} else {
- $rowNum = intval( $numRows * $i / $numBatches );
+ $rowNum = $i * $batchSize;
$res->seek( $rowNum );
$row = $res->fetchObject();
- $start = $row->page_id;
+ $start = (int)$row->page_id;
}
- if ( $i == $numBatches - 1 ) {
+ if ( $i == ( $numBatches - 1 ) && $isComplete ) {
$end = false;
} else {
- $rowNum = intval( $numRows * ( $i + 1 ) / $numBatches );
+ $rowNum = min( $numRows - 1, ( $i + 1 ) * $batchSize - 1 );
$res->seek( $rowNum );
$row = $res->fetchObject();
- $end = $row->page_id - 1;
+ $end = (int)$row->page_id;
}
# Sanity check order