X-Git-Url: https://git.heureux-cyclage.org/?a=blobdiff_plain;f=includes%2Ffilerepo%2Fbackend%2FFileBackendStore.php;h=30a64e298ae2e9e31518e263776e1748433501b3;hb=ed6b0a2a205afaca3f1117577c86167ce1871a48;hp=8c849cad02780311fa0759d70948f514346416c4;hpb=458f162b63d3c44bd158393d04f11d67c63b6dd4;p=lhc%2Fweb%2Fwiklou.git diff --git a/includes/filerepo/backend/FileBackendStore.php b/includes/filerepo/backend/FileBackendStore.php index 8c849cad02..30a64e298a 100644 --- a/includes/filerepo/backend/FileBackendStore.php +++ b/includes/filerepo/backend/FileBackendStore.php @@ -1,5 +1,22 @@ key => value) - protected $maxCacheSize = 100; // integer; max paths with entries + protected $maxCacheSize = 300; // integer; max paths with entries /** @var Array Map of paths to large (RAM/disk) cache items */ protected $expensiveCache = array(); // (storage path => key => value) - protected $maxExpensiveCacheSize = 10; // integer; max paths with entries + protected $maxExpensiveCacheSize = 5; // integer; max paths with entries /** @var Array Map of container names to sharding settings */ protected $shardViaHashLevels = array(); // (container name => config array) protected $maxFileSize = 4294967296; // integer bytes (4GiB) + /** + * @see FileBackend::__construct() + * + * @param $config Array + */ + public function __construct( array $config ) { + parent::__construct( $config ); + $this->memCache = new EmptyBagOStuff(); // disabled by default + } + /** * Get the maximum allowable file size given backend * medium restrictions and basic performance constraints. @@ -73,6 +103,7 @@ abstract class FileBackendStore extends FileBackend { } else { $status = $this->doCreateInternal( $params ); $this->clearCache( array( $params['dst'] ) ); + $this->deleteFileCache( $params['dst'] ); // persistent cache } wfProfileOut( __METHOD__ . '-' . $this->name ); wfProfileOut( __METHOD__ ); @@ -104,6 +135,7 @@ abstract class FileBackendStore extends FileBackend { } else { $status = $this->doStoreInternal( $params ); $this->clearCache( array( $params['dst'] ) ); + $this->deleteFileCache( $params['dst'] ); // persistent cache } wfProfileOut( __METHOD__ . '-' . $this->name ); wfProfileOut( __METHOD__ ); @@ -132,6 +164,7 @@ abstract class FileBackendStore extends FileBackend { wfProfileIn( __METHOD__ . '-' . $this->name ); $status = $this->doCopyInternal( $params ); $this->clearCache( array( $params['dst'] ) ); + $this->deleteFileCache( $params['dst'] ); // persistent cache wfProfileOut( __METHOD__ . '-' . $this->name ); wfProfileOut( __METHOD__ ); return $status; @@ -158,6 +191,7 @@ abstract class FileBackendStore extends FileBackend { wfProfileIn( __METHOD__ . '-' . $this->name ); $status = $this->doDeleteInternal( $params ); $this->clearCache( array( $params['src'] ) ); + $this->deleteFileCache( $params['src'] ); // persistent cache wfProfileOut( __METHOD__ . '-' . $this->name ); wfProfileOut( __METHOD__ ); return $status; @@ -185,6 +219,8 @@ abstract class FileBackendStore extends FileBackend { wfProfileIn( __METHOD__ . '-' . $this->name ); $status = $this->doMoveInternal( $params ); $this->clearCache( array( $params['src'], $params['dst'] ) ); + $this->deleteFileCache( $params['src'] ); // persistent cache + $this->deleteFileCache( $params['dst'] ); // persistent cache wfProfileOut( __METHOD__ . '-' . $this->name ); wfProfileOut( __METHOD__ ); return $status; @@ -371,6 +407,17 @@ abstract class FileBackendStore extends FileBackend { wfProfileIn( __METHOD__ . '-' . $this->name ); $status = Status::newGood(); + // Recursive: first delete all empty subdirs recursively + if ( !empty( $params['recursive'] ) && !$this->directoriesAreVirtual() ) { + $subDirsRel = $this->getTopDirectoryList( array( 'dir' => $params['dir'] ) ); + if ( $subDirsRel !== null ) { // no errors + foreach ( $subDirsRel as $subDirRel ) { + $subDir = $params['dir'] . "/{$subDirRel}"; // full path + $status->merge( $this->doClean( array( 'dir' => $subDir ) + $params ) ); + } + } + } + list( $fullCont, $dir, $shard ) = $this->resolveStoragePath( $params['dir'] ); if ( $dir === null ) { $status->fatal( 'backend-fail-invalidpath', $params['dir'] ); @@ -390,11 +437,13 @@ abstract class FileBackendStore extends FileBackend { if ( $shard !== null ) { // confined to a single container/shard $status->merge( $this->doCleanInternal( $fullCont, $dir, $params ) ); + $this->deleteContainerCache( $fullCont ); // purge cache } else { // directory is on several shards wfDebug( __METHOD__ . ": iterating over all container shards.\n" ); list( $b, $shortCont, $r ) = self::splitStoragePath( $params['dir'] ); foreach ( $this->getContainerSuffixes( $shortCont ) as $suffix ) { $status->merge( $this->doCleanInternal( "{$fullCont}{$suffix}", $dir, $params ) ); + $this->deleteContainerCache( "{$fullCont}{$suffix}" ); // purge cache } } @@ -463,7 +512,10 @@ abstract class FileBackendStore extends FileBackend { wfProfileOut( __METHOD__ ); return false; // invalid storage path } - $latest = !empty( $params['latest'] ); + $latest = !empty( $params['latest'] ); // use latest data? + if ( !isset( $this->cache[$path]['stat'] ) ) { + $this->primeFileCache( array( $path ) ); // check persistent cache + } if ( isset( $this->cache[$path]['stat'] ) ) { // If we want the latest data, check that this cached // value was in fact fetched with the latest available data. @@ -480,9 +532,10 @@ abstract class FileBackendStore extends FileBackend { wfProfileOut( __METHOD__ . '-miss-' . $this->name ); wfProfileOut( __METHOD__ . '-miss' ); if ( is_array( $stat ) ) { // don't cache negatives + $stat['latest'] = $latest; $this->trimCache(); // limit memory $this->cache[$path]['stat'] = $stat; - $this->cache[$path]['stat']['latest'] = $latest; + $this->setFileCache( $path, $stat ); // update persistent cache } wfProfileOut( __METHOD__ . '-' . $this->name ); wfProfileOut( __METHOD__ ); @@ -579,7 +632,7 @@ abstract class FileBackendStore extends FileBackend { wfProfileIn( __METHOD__ . '-' . $this->name ); $path = $params['src']; if ( isset( $this->expensiveCache[$path]['localRef'] ) ) { - $this->pingExpensiveCache(); + $this->pingExpensiveCache( $path ); wfProfileOut( __METHOD__ . '-' . $this->name ); wfProfileOut( __METHOD__ ); return $this->expensiveCache[$path]['localRef']; @@ -646,8 +699,79 @@ abstract class FileBackendStore extends FileBackend { } /** - * @copydoc FileBackend::getFileList() - * @return Array|null|Traversable + * @see FileBackend::directoryExists() + * @return bool|null + */ + final public function directoryExists( array $params ) { + list( $fullCont, $dir, $shard ) = $this->resolveStoragePath( $params['dir'] ); + if ( $dir === null ) { + return false; // invalid storage path + } + if ( $shard !== null ) { // confined to a single container/shard + return $this->doDirectoryExists( $fullCont, $dir, $params ); + } else { // directory is on several shards + wfDebug( __METHOD__ . ": iterating over all container shards.\n" ); + list( $b, $shortCont, $r ) = self::splitStoragePath( $params['dir'] ); + $res = false; // response + foreach ( $this->getContainerSuffixes( $shortCont ) as $suffix ) { + $exists = $this->doDirectoryExists( "{$fullCont}{$suffix}", $dir, $params ); + if ( $exists ) { + $res = true; + break; // found one! + } elseif ( $exists === null ) { // error? + $res = null; // if we don't find anything, it is indeterminate + } + } + return $res; + } + } + + /** + * @see FileBackendStore::directoryExists() + * + * @param $container string Resolved container name + * @param $dir string Resolved path relative to container + * @param $params Array + * @return bool|null + */ + abstract protected function doDirectoryExists( $container, $dir, array $params ); + + /** + * @see FileBackend::getDirectoryList() + * @return Traversable|Array|null Returns null on failure + */ + final public function getDirectoryList( array $params ) { + list( $fullCont, $dir, $shard ) = $this->resolveStoragePath( $params['dir'] ); + if ( $dir === null ) { // invalid storage path + return null; + } + if ( $shard !== null ) { + // File listing is confined to a single container/shard + return $this->getDirectoryListInternal( $fullCont, $dir, $params ); + } else { + wfDebug( __METHOD__ . ": iterating over all container shards.\n" ); + // File listing spans multiple containers/shards + list( $b, $shortCont, $r ) = self::splitStoragePath( $params['dir'] ); + return new FileBackendStoreShardDirIterator( $this, + $fullCont, $dir, $this->getContainerSuffixes( $shortCont ), $params ); + } + } + + /** + * Do not call this function from places outside FileBackend + * + * @see FileBackendStore::getDirectoryList() + * + * @param $container string Resolved container name + * @param $dir string Resolved path relative to container + * @param $params Array + * @return Traversable|Array|null Returns null on failure + */ + abstract public function getDirectoryListInternal( $container, $dir, array $params ); + + /** + * @see FileBackend::getFileList() + * @return Traversable|Array|null Returns null on failure */ final public function getFileList( array $params ) { list( $fullCont, $dir, $shard ) = $this->resolveStoragePath( $params['dir'] ); @@ -661,7 +785,7 @@ abstract class FileBackendStore extends FileBackend { wfDebug( __METHOD__ . ": iterating over all container shards.\n" ); // File listing spans multiple containers/shards list( $b, $shortCont, $r ) = self::splitStoragePath( $params['dir'] ); - return new FileBackendStoreShardListIterator( $this, + return new FileBackendStoreShardFileIterator( $this, $fullCont, $dir, $this->getContainerSuffixes( $shortCont ), $params ); } } @@ -674,7 +798,7 @@ abstract class FileBackendStore extends FileBackend { * @param $container string Resolved container name * @param $dir string Resolved path relative to container * @param $params Array - * @return Traversable|Array|null + * @return Traversable|Array|null Returns null on failure */ abstract public function getFileListInternal( $container, $dir, array $params ); @@ -705,7 +829,7 @@ abstract class FileBackendStore extends FileBackend { * @return Array List of FileOp objects * @throws MWException */ - final public function getOperations( array $ops ) { + final public function getOperationsInternal( array $ops ) { $supportedOps = $this->supportedOperations(); $performOps = array(); // array of FileOp objects @@ -719,13 +843,36 @@ abstract class FileBackendStore extends FileBackend { // Append the FileOp class $performOps[] = new $class( $this, $params ); } else { - throw new MWException( "Operation `$opName` is not supported." ); + throw new MWException( "Operation '$opName' is not supported." ); } } return $performOps; } + /** + * Get a list of storage paths to lock for a list of operations + * Returns an array with 'sh' (shared) and 'ex' (exclusive) keys, + * each corresponding to a list of storage paths to be locked. + * + * @param $performOps Array List of FileOp objects + * @return Array ('sh' => list of paths, 'ex' => list of paths) + */ + final public function getPathsToLockForOpsInternal( array $performOps ) { + // Build up a list of files to lock... + $paths = array( 'sh' => array(), 'ex' => array() ); + foreach ( $performOps as $fileOp ) { + $paths['sh'] = array_merge( $paths['sh'], $fileOp->storagePathsRead() ); + $paths['ex'] = array_merge( $paths['ex'], $fileOp->storagePathsChanged() ); + } + // Optimization: if doing an EX lock anyway, don't also set an SH one + $paths['sh'] = array_diff( $paths['sh'], $paths['ex'] ); + // Get a shared lock on the parent directory of each path changed + $paths['sh'] = array_merge( $paths['sh'], array_map( 'dirname', $paths['ex'] ) ); + + return $paths; + } + /** * @see FileBackend::doOperationsInternal() * @return Status @@ -736,23 +883,15 @@ abstract class FileBackendStore extends FileBackend { $status = Status::newGood(); // Build up a list of FileOps... - $performOps = $this->getOperations( $ops ); + $performOps = $this->getOperationsInternal( $ops ); // Acquire any locks as needed... if ( empty( $opts['nonLocking'] ) ) { // Build up a list of files to lock... - $filesLockEx = $filesLockSh = array(); - foreach ( $performOps as $fileOp ) { - $filesLockSh = array_merge( $filesLockSh, $fileOp->storagePathsRead() ); - $filesLockEx = array_merge( $filesLockEx, $fileOp->storagePathsChanged() ); - } - // Optimization: if doing an EX lock anyway, don't also set an SH one - $filesLockSh = array_diff( $filesLockSh, $filesLockEx ); - // Get a shared lock on the parent directory of each path changed - $filesLockSh = array_merge( $filesLockSh, array_map( 'dirname', $filesLockEx ) ); + $paths = $this->getPathsToLockForOpsInternal( $performOps ); // Try to lock those files for the scope of this function... - $scopeLockS = $this->getScopedFileLocks( $filesLockSh, LockManager::LOCK_UW, $status ); - $scopeLockE = $this->getScopedFileLocks( $filesLockEx, LockManager::LOCK_EX, $status ); + $scopeLockS = $this->getScopedFileLocks( $paths['sh'], LockManager::LOCK_UW, $status ); + $scopeLockE = $this->getScopedFileLocks( $paths['ex'], LockManager::LOCK_EX, $status ); if ( !$status->isOK() ) { wfProfileOut( __METHOD__ . '-' . $this->name ); wfProfileOut( __METHOD__ ); @@ -760,9 +899,13 @@ abstract class FileBackendStore extends FileBackend { } } - // Clear any cache entries (after locks acquired) + // Clear any file cache entries (after locks acquired) $this->clearCache(); + // Load from the persistent file and container caches + $this->primeFileCache( $performOps ); + $this->primeContainerCache( $performOps ); + // Actually attempt the operation batch... $subStatus = FileOp::attemptBatch( $performOps, $opts, $this->fileJournal ); @@ -805,10 +948,20 @@ abstract class FileBackendStore extends FileBackend { */ protected function doClearCache( array $paths = null ) {} + /** + * Is this a key/value store where directories are just virtual? + * Virtual directories exists in so much as files exists that are + * prefixed with the directory path followed by a forward slash. + * + * @return bool + */ + abstract protected function directoriesAreVirtual(); + /** * Move a cache entry to the top (such as when accessed) * * @param $path string Storage path + * @return void */ protected function pingCache( $path ) { if ( isset( $this->cache[$path] ) ) { @@ -834,6 +987,7 @@ abstract class FileBackendStore extends FileBackend { * Move a cache entry to the top (such as when accessed) * * @param $path string Storage path + * @return void */ protected function pingExpensiveCache( $path ) { if ( isset( $this->expensiveCache[$path] ) ) { @@ -962,6 +1116,19 @@ abstract class FileBackendStore extends FileBackend { return ''; // no sharding } + /** + * Check if a storage path maps to a single shard. + * Container dirs like "a", where the container shards on "x/xy", + * can reside on several shards. Such paths are tricky to handle. + * + * @param $storagePath string Storage path + * @return bool + */ + final public function isSingleShardPathInternal( $storagePath ) { + list( $c, $r, $shard ) = $this->resolveStoragePath( $storagePath ); + return ( $shard !== null ); + } + /** * Get the sharding config for a container. * If greater than 0, then all file storage paths within @@ -1041,29 +1208,198 @@ abstract class FileBackendStore extends FileBackend { protected function resolveContainerPath( $container, $relStoragePath ) { return $relStoragePath; } + + /** + * Get the cache key for a container + * + * @param $container Resolved container name + * @return string + */ + private function containerCacheKey( $container ) { + return wfMemcKey( 'backend', $this->getName(), 'container', $container ); + } + + /** + * Set the cached info for a container + * + * @param $container Resolved container name + * @param $val mixed Information to cache + * @return void + */ + final protected function setContainerCache( $container, $val ) { + $this->memCache->set( $this->containerCacheKey( $container ), $val, 14*86400 ); + } + + /** + * Delete the cached info for a container + * + * @param $container Resolved container name + * @return void + */ + final protected function deleteContainerCache( $container ) { + for ( $attempts=1; $attempts <= 3; $attempts++ ) { + if ( $this->memCache->delete( $this->containerCacheKey( $container ) ) ) { + return; // done! + } + } + trigger_error( "Unable to delete stat cache for container $container." ); + } + + /** + * Do a batch lookup from cache for container stats for all containers + * used in a list of container names, storage paths, or FileOp objects. + * + * @param $items Array + * @return void + */ + final protected function primeContainerCache( array $items ) { + wfProfileIn( __METHOD__ ); + wfProfileIn( __METHOD__ . '-' . $this->name ); + $paths = array(); // list of storage paths + $contNames = array(); // (cache key => resolved container name) + // Get all the paths/containers from the items... + foreach ( $items as $item ) { + if ( $item instanceof FileOp ) { + $paths = array_merge( $paths, $item->storagePathsRead() ); + $paths = array_merge( $paths, $item->storagePathsChanged() ); + } elseif ( self::isStoragePath( $item ) ) { + $paths[] = $item; + } elseif ( is_string( $item ) ) { // full container name + $contNames[$this->containerCacheKey( $item )] = $item; + } + } + // Get all the corresponding cache keys for paths... + foreach ( $paths as $path ) { + list( $fullCont, $r, $s ) = $this->resolveStoragePath( $path ); + if ( $fullCont !== null ) { // valid path for this backend + $contNames[$this->containerCacheKey( $fullCont )] = $fullCont; + } + } + + $contInfo = array(); // (resolved container name => cache value) + // Get all cache entries for these container cache keys... + $values = $this->memCache->getBatch( array_keys( $contNames ) ); + foreach ( $values as $cacheKey => $val ) { + $contInfo[$contNames[$cacheKey]] = $val; + } + + // Populate the container process cache for the backend... + $this->doPrimeContainerCache( array_filter( $contInfo, 'is_array' ) ); + wfProfileOut( __METHOD__ . '-' . $this->name ); + wfProfileOut( __METHOD__ ); + } + + /** + * Fill the backend-specific process cache given an array of + * resolved container names and their corresponding cached info. + * Only containers that actually exist should appear in the map. + * + * @param $containerInfo Array Map of resolved container names to cached info + * @return void + */ + protected function doPrimeContainerCache( array $containerInfo ) {} + + /** + * Get the cache key for a file path + * + * @param $path Storage path + * @return string + */ + private function fileCacheKey( $path ) { + return wfMemcKey( 'backend', $this->getName(), 'file', sha1( $path ) ); + } + + /** + * Set the cached stat info for a file path + * + * @param $path Storage path + * @param $val mixed Information to cache + * @return void + */ + final protected function setFileCache( $path, $val ) { + $this->memCache->set( $this->fileCacheKey( $path ), $val, 7*86400 ); + } + + /** + * Delete the cached stat info for a file path + * + * @param $path Storage path + * @return void + */ + final protected function deleteFileCache( $path ) { + for ( $attempts=1; $attempts <= 3; $attempts++ ) { + if ( $this->memCache->delete( $this->fileCacheKey( $path ) ) ) { + return; // done! + } + } + trigger_error( "Unable to delete stat cache for file $path." ); + } + + /** + * Do a batch lookup from cache for file stats for all paths + * used in a list of storage paths or FileOp objects. + * + * @param $items Array List of storage paths or FileOps + * @return void + */ + final protected function primeFileCache( array $items ) { + wfProfileIn( __METHOD__ ); + wfProfileIn( __METHOD__ . '-' . $this->name ); + $paths = array(); // list of storage paths + $pathNames = array(); // (cache key => storage path) + // Get all the paths/containers from the items... + foreach ( $items as $item ) { + if ( $item instanceof FileOp ) { + $paths = array_merge( $paths, $item->storagePathsRead() ); + $paths = array_merge( $paths, $item->storagePathsChanged() ); + } elseif ( self::isStoragePath( $item ) ) { + $paths[] = $item; + } + } + // Get all the corresponding cache keys for paths... + foreach ( $paths as $path ) { + list( $cont, $rel, $s ) = $this->resolveStoragePath( $path ); + if ( $rel !== null ) { // valid path for this backend + $pathNames[$this->fileCacheKey( $path )] = $path; + } + } + // Get all cache entries for these container cache keys... + $values = $this->memCache->getBatch( array_keys( $pathNames ) ); + foreach ( $values as $cacheKey => $val ) { + if ( is_array( $val ) ) { + $this->trimCache(); // limit memory + $this->cache[$pathNames[$cacheKey]]['stat'] = $val; + } + } + wfProfileOut( __METHOD__ . '-' . $this->name ); + wfProfileOut( __METHOD__ ); + } } /** - * FileBackendStore helper function to handle file listings that span container shards. + * FileBackendStore helper function to handle listings that span container shards. * Do not use this class from places outside of FileBackendStore. * * @ingroup FileBackend */ -class FileBackendStoreShardListIterator implements Iterator { - /* @var FileBackendStore */ +abstract class FileBackendStoreShardListIterator implements Iterator { + /** @var FileBackendStore */ protected $backend; - /* @var Array */ + /** @var Array */ protected $params; - /* @var Array */ + /** @var Array */ protected $shardSuffixes; - protected $container; // string - protected $directory; // string + protected $container; // string; full container name + protected $directory; // string; resolved relative path - /* @var Traversable */ + /** @var Traversable */ protected $iter; protected $curShard = 0; // integer protected $pos = 0; // integer + /** @var Array */ + protected $multiShardPaths = array(); // (rel path => 1) + /** * @param $backend FileBackendStore * @param $container string Full storage container name @@ -1112,6 +1448,8 @@ class FileBackendStoreShardListIterator implements Iterator { } else { $this->iter->next(); } + // Filter out items that we already listed + $this->filterViaNext(); // Find the next non-empty shard if no elements are left $this->nextShardIteratorIfNotValid(); } @@ -1124,6 +1462,8 @@ class FileBackendStoreShardListIterator implements Iterator { $this->pos = 0; $this->curShard = 0; $this->setIteratorFromCurrentShard(); + // Filter out items that we already listed + $this->filterViaNext(); // Find the next non-empty shard if this one has no elements $this->nextShardIteratorIfNotValid(); } @@ -1133,7 +1473,7 @@ class FileBackendStoreShardListIterator implements Iterator { * @return bool */ public function valid() { - if ( $this->iter == null ) { + if ( $this->iter === null ) { return false; // some failure? } elseif ( is_array( $this->iter ) ) { return ( current( $this->iter ) !== false ); // no paths can have this value @@ -1142,6 +1482,25 @@ class FileBackendStoreShardListIterator implements Iterator { } } + /** + * Filter out duplicate items by advancing to the next ones + */ + protected function filterViaNext() { + while ( $this->iter->valid() ) { + $rel = $this->iter->current(); // path relative to given directory + $path = $this->params['dir'] . "/{$rel}"; // full storage path + if ( !$this->backend->isSingleShardPathInternal( $path ) ) { + // Don't keep listing paths that are on multiple shards + if ( isset( $this->multiShardPaths[$rel] ) ) { + $this->iter->next(); // we already listed this path + } else { + $this->multiShardPaths[$rel] = 1; + break; + } + } + } + } + /** * If the list iterator for this container shard is out of items, * then move on to the next container that has items. @@ -1161,7 +1520,35 @@ class FileBackendStoreShardListIterator implements Iterator { */ protected function setIteratorFromCurrentShard() { $suffix = $this->shardSuffixes[$this->curShard]; - $this->iter = $this->backend->getFileListInternal( + $this->iter = $this->listFromShard( "{$this->container}{$suffix}", $this->directory, $this->params ); } + + /** + * Get the list for a given container shard + * + * @param $container string Resolved container name + * @param $dir string Resolved path relative to container + * @param $params Array + * @return Traversable|Array|null + */ + abstract protected function listFromShard( $container, $dir, array $params ); +} + +/** + * Iterator for listing directories + */ +class FileBackendStoreShardDirIterator extends FileBackendStoreShardListIterator { + protected function listFromShard( $container, $dir, array $params ) { + return $this->backend->getDirectoryListInternal( $container, $dir, $params ); + } +} + +/** + * Iterator for listing regular files + */ +class FileBackendStoreShardFileIterator extends FileBackendStoreShardListIterator { + protected function listFromShard( $container, $dir, array $params ) { + return $this->backend->getFileListInternal( $container, $dir, $params ); + } }